(PDF) An Enquiry into the Causes and Nature of the Transmission Mechanisms between Labour-Based Rural Roads, Sustainable Growth, and Agricultural Trade in Zambia‟s Eastern Province

Poverty is largely found in the rural areas and rural employment is still mainly found in the agricultural sector. These facts should lend credence to the perception of rural transport infrastructure (RTI) as an important policy instrument for rural development. Thus, the objective of the thesis is to empirically establish the structural relationship between RTI development, pro-poor rural growth and agricultural trade. Conceptually the thesis extracts insights gained from growth theories and the traditional cross-country literature on the relationships between infrastructure provision and growth, while recognizing that this macro approach is inadequate for evaluating the effects at the sub-national level. Instead by relying on a micro-level approach our key findings are the following: At the district level improved accessibility led to changes in land allocation and in yields to Eastern Province of Zambia‘s most important cash crop – cotton. Although, the mean cotton sales share of household income more than doubled, the estimation results only show small gains to mean consumption. Through a qualitative analysis we find that improved access to external markets is a critical determinant of the households‘ ability to increase their income and break out of the poverty trap. Moreover, we find that it is primarily small private companies that are more likely to have relocated as a consequence of the Eastern Province Feeder Road Project (EPFRP). The entry by the large South African Clark Cotton Company in 1995 was, however, linked to Zambia‘s and South Africa‘s SADC membership. The thesis contributes to the methodological framework used in rural projects evaluation by proposing that household and firm survey analysis has the potential of providing insights that go beyond what is revealed by aggregate cross-country regressions and traditional cost-benefit analysis. However, our impact evaluation approach is a-theoretical and reduced form and it should therefore be considered to be complementary to the more detailed structural approach used by other authors referred in the literature review. Finally, the thesis recommends that future research should be devoted to further advancing appraisal techniques for rural road projects and to discuss how quantitative impact evaluations of rural development policy instruments such as feeder roads can best influence policy in order to help make progress towards the Millennium Development Goals.

An Enquiry into the Causes and Nature of the Transmission Mechanisms between Labour-Based Rural Roads, Sustainable Growth, and Agricultural Trade in Zambia‟s Eastern Province Christian Kitenge Moembo Kingombe A Thesis Submitted for the Degree of Doctor of Philosophy University of London, Imperial College Supervisors: Colin Thirtle, Salvatore di Falco and Jonathan Kydd Date: 26th of March 2011 i Declaration I herewith certify that all material in this dissertation which is not my own work has been properly acknowledged. Name: Christian Kingombe ii Abstract An Enquiry into the Causes and Nature of the Transmission Mechanisms between Labour-Based Rural Roads, Sustainable Growth, and Agricultural Trade in Zambia‟s Eastern Province Christian Kitenge Moembo Kingombe Thesis Submitted for the Degree of Doctor of Philosophy in Economics Imperial College, University of London and Centre for Environmental Policy Trinity Term 2010 Poverty is largely found in the rural areas and rural employment is still mainly found in the agricultural sector. These facts should lend credence to the perception of rural transport infrastructure (RTI) as an important policy instrument for rural development. Thus, the objective of the thesis is to empirically establish the structural relationship between RTI development, pro-poor rural growth and agricultural trade. Conceptually the thesis extracts insights gained from growth theories and the traditional cross-country literature on the relationships between infrastructure provision and growth, while recognizing that this macro approach is inadequate for evaluating the effects at the sub-national level. Instead by relying on a micro-level approach our key findings are the following: At the district level improved accessibility led to changes in land allocation and in yields to Eastern Province of Zambia‘s most important cash crop – cotton. Although, the mean cotton sales share of household income more than doubled, the estimation results only show small gains to mean consumption. Through a qualitative analysis we find that improved access to external markets is a critical determinant of the households‘ ability to increase their income and break out of the poverty trap. Moreover, we find that it is primarily small private companies that are more likely to have relocated as a consequence of the Eastern Province Feeder Road Project (EPFRP). The entry by the large South African Clark Cotton Company in 1995 was, however, linked to Zambia‘s and South Africa‘s SADC membership. The thesis contributes to the methodological framework used in rural projects evaluation by proposing that household and firm survey analysis has the potential of providing insights that go beyond what is revealed by aggregate cross-country regressions and traditional cost-benefit analysis. However, our impact evaluation approach is a-theoretical and reduced form and it should therefore be considered to be complementary to the more detailed structural approach used by other authors referred in the literature review. Finally, the thesis recommends that future research should be devoted to further advancing appraisal techniques for rural road projects and to discuss how quantitative impact evaluations of rural development policy instruments such as feeder roads can best influence policy in order to help make progress towards the Millennium Development Goals. iii Dedication This thesis is above all dedicated to my beloved partner Ariane Laloe for her support, patience, and never fading love during this long and painful process. I would also like to dedicate this dissertation to my family: To my mother Elisabeth Siggaard Kingombe to whom I owe everything and to my younger but inspirational brother Ndjadi Kingombe who has shown through his own career how much you can achieve through hard work and sacrifice when you wholeheartedly believe in an innovative idea. iv Acknowledgements I am grateful to many people and colleagues for their advice, encouragement and contribution to the immensely difficult task of completing this dissertation part-time over the past six years. It all started in 2003 when I was stationed in the ILO Area Office in Lusaka, Zambia. During the pre-departure briefing that I received in the ILO Headquarters in Geneva Terje Tessem (ILO/EIIP) introduced me to the notion of Labour Based Technology (LBT) Infrastructure Construction Methods. Subsequently I carried out a number of field visits in rural Zambia together with my ILO colleague Carl-Erik Hedstrom (Road Training School) and Simon Tembo (EPFRP national engineer), which converted me into believing in this idea of maximizing the economic benefits of LBT infrastructure development projects. A special thanks also goes to Frans Blokhuis (EPFRP international engineer) for providing me with all the necessary district road maps and project data and for telling me the whole story about the implementation of the EPFRP. I would like to thank my first supervisor Professor Jonathan Kydd, who despite my choice of not using a NIE approach against his advice, still took an interest in my work and selflessly gave a massive amount of his time during the 15 months from the MPhil registration to the PhD Transfer Report and oral examination through weekly sessions in which he shared his enormous knowledge about rural development in Zambia and neighbouring Malawi. I should also thank my Progress Review Panel, especially Professor Andrew Dorward for his comments and suggestions on the Transfer report and chapter 7. Notwithstanding the excellent econometric training I have received from during the years by: Erling B. Andersen; Ellen Andersen; Marcia Schafgans; Hidehiko Ichimura; Monica Costa Dias; Andrew Chesher; Sokbae (Simon) Lee; and Richard Blundell, I‘m very grateful for the indispensable econometric advice that I received from my methodological advisor Dr.Salvatore di Falco. vi I‘m grateful to Professor Colin Thirtle, who took over as principal supervisor from Professor Jonathan Kydd, for his support in the task of completing this dissertation and for his willingness to provide comments and suggestions for the entire thesis. Many others have contributed to improving sections of this thesis, most notably Professor Michael Grimm, who commented on chapter 4 presented at the ISS ωonference ‗Development Dialogue‘ and Professor Marcel Fafchamps who commented on chapter 6 and chapter 3 both presented at the CSAE Seminar series in Oxford. The study would not have been possible without the support of a number of people in Zambia. I would like to thank the staff at both the LCMS-unit and the Agricultural & Environment Division both at Zambia‘s ωentral Statistical ηffice for providing me with access to both the LCMS and PHS datasets; my four Eastern Province field assistants: Prince Mwenge; Lenai Msundwe (R.I.P.); Mwima Mfiwe as well as the knowledgeable Amos Phiri; and last but not least Patrick Banda for setting up all the interviews with the key informants in Lusaka as well as providing additional research assistance during the first field visit to Chipata, Eastern Province, in August 2004. The Orwin Building at Imperial College Wye Campus was my institutional home during the first 18 months of my thesis and the CEP PhD common room 311 at the Mechanical Engineering Building at Imperial College South Kensington Campus, during the last year of my doctoral studies. I would like to thank staff and fellow researchers at both campuses for ensuring a supportive and friendly working environment. In between beginning and the end I had the great privilege of gaining PhD relevant professional experience in the areas of: Rural development statistics at UNECE Statistical Division (e.g. Jan Karlsson & Prof. Berkeley Hill); Transport Investment Appraisal at COWI A/S & Institute for Transport Studies - University of Leeds (e.g. Prof. Peter Mackie); Economic Growth in Africa at OECD Development Centre (e.g. Dr.Kenneth Ruffing) & CREI Univ. Pompeu Fabra (Prof. Sala-i-Martin); Regional Integration at UNCTAD‘s Africa Division (Prof. Charles Gore). Finally, I would like to thank both Prof. Pierre Biscourp and especially Prof. Paul Collier for kindly inviting me to visit respectively ENSAE-CREST ParisTech and CSAE Oxford University during the write-up. vii Despite living away from my partner Ariane Laloe in Geneva; my mother Elisabeth Siggaard Kingombe in Copenhagen/Florence and brother Ndjadi Kingombe in Copenhagen/Malmö throughout the entire process, I would never have made it without their love, encouragement, and financial support. Christian Kingombe Oxford, June 2010. I would like to thank my VIVA examiners Professor Peter Hazell and Professor Bhavani Shankar for their helpful comments and suggestions in the their separate preliminary-Viva reports, during the VIVA held on the 7th of January 2011 and subsequently in their joint VIVA report. Furthermore, I would like to thank my current colleagues at the Overseas Development Institute (ODI) who attended an internal ODI seminar in particular Dr.Anna McCord; Dr.Sobona Mtisi; Claire Melamed and Dr.Jessica Hagen-Zanker for their comments. Christian Kingombe Geneva, March 2011 viii Table of Contents Declaration......................................................................................................................... ii Abstract ............................................................................................................................. iii Dedication ......................................................................................................................... iv Acknowledgements .......................................................................................................... vi Figures ............................................................................................................................. xiii Tables .............................................................................................................................. xiv Maps ............................................................................................................................... xvii Annexes ......................................................................................................................... xviii Abbreviations ................................................................................................................. xix Overview ........................................................................................................................ xxii Part One ............................................................................................................................. 0 Introduction and Approaches to Assess the Effect of Public Capital Infrastructure on Economic Growth ........................................................................................................ 0 Chapter 1: Policy Evaluation of the Investment-Trade Nexus ..................................... 1 1.1. Introduction ............................................................................................................ 2 1.2. Overview of the Major Issues ............................................................................... 3 1.2.1. Long-Run Growth .......................................................................................... 4 1.2.2. Rural Development through Public Works Programmes ........................... 5 1.2.3. The Role of Agriculture in Development and Structural Transformation 7 1.3. Analytical Research Approach and Scope of Research.................................... 11 1.4. A Brief Summary of the Key Findings ............................................................... 17 1.5. An Overview of the Structure of the Thesis ...................................................... 18 Chapter 2: Theories of Economic Growth and Empirics of Growth: An Exposition and Assessment of the Macro-level Theoretical and Empirical foundations of the Dynamic Models of Production and Consumption Growth at the Micro-level ........ 20 2.1. Introduction .............................................................................................................. 21 2.2. Theoretical Foundations: Mainstream Theories of Economic Growth .......... 24 2.3. The Empirical Challenge: Cross-Country Patterns of Economic Growth ..... 29 2.3.1. Identified Regressors by the Empirical Growth Literature ..................... 29 2.3.2. Other Issues in the Study of Growth .......................................................... 33 β.4. Supply Side: Productive Government Expenditures‟ Impact on Economic Growth ............................................................................................................................. 36 2.4.1. Theoretical Approach to Rurality ............................................................... 36 2.4.2. Productive Government Expenditures and Rural Development ............. 37 viii 2.4.2.1. Poverty Reducing Employment Intensive Investment Programmes ........... 38 2.4.2.2. Structural Models of Road Impacts ............................................................. 42 2.4.2.3. Impact Evaluation of Rural Road Infrastructure .......................................... 42 2.5. Demand Side: Rural Growth Through Agricultural Trade ............................ 47 2.6. Conclusions and Policy Implications .................................................................. 51 Chapter γ: Zambia‟s Eastern Province and Eastern Province Feeder Road Project ........................................................................................................................................... 55 3.1. Key contextual information about Zambia‟s Eastern Province .......................... 56 3.2. The Eastern Province Feeder Road Project .......................................................... 58 3.2.1. The Implementation of the Feeder Road Programme................................... 63 3.3. Justification of Analytical Approach ..................................................................... 71 Part Two .......................................................................................................................... 79 Empirical Investigations of the Investment-led Rural Development Approach through Rural Transport Infrastructure Development in Zambia‟s Eastern Province ........................................................................................................................... 79 Chapter 4: The Impact of Rural Roads on Cash Crop Production in Zambia‟s Eastern Province between 1997 and 2002 ..................................................................... 80 4.1. Introduction .............................................................................................................. 81 4.2. Background and Setting ...................................................................................... 84 4.3. Framework ........................................................................................................... 89 4.3.1. Agricultural Productivity: Estimation Strategy ............................................ 90 4.3.1.1. A Counterfactual Setting.............................................................................. 90 4.3.1.2. Estimating Model of Cotton Productivity .................................................... 94 4.3.2. Agricultural Productivity: Identification Strategy ........................................ 96 4.4. Data ..................................................................................................................... 103 4.4.1. The 1990 & 2000 Censuses of Population, Housing and Agriculture ........ 103 4.4.2. Review of Sample Design for 1996/1997 - 1999/2000 Post-Harvest Survey104 4.5. Estimation Results and Discussion ....................................................................... 109 4.5.1. Descriptive Statistics ....................................................................................... 109 4.5.β. Evaluation of the EPFRP‟s impact on Cotton Productivity ....................... 110 4.5.2.1. Matching and Propensity Score Estimators ............................................... 111 4.5.2.2. Differences-in-Differences Estimators ...................................................... 117 4.5.2.3. Further Specification .................................................................................. 124 4.5.3. Robustness Checks .......................................................................................... 130 4.5.3.1. Tests of the Matching Assumption and Sensitivity of Estimates .............. 130 4.5.3.2. Specification tests and model diagnostic ................................................... 134 4.6. Conclusions and Policy Implications .................................................................... 136 ix Chapter 5: Regional Analysis of Eastern Province Feeder Road Project. District level estimation of the Poverty Alleviation Effects of Rural Roads Improvements in Zambia‟s Eastern Province .......................................................................................... 139 5.1. Introduction ............................................................................................................ 140 5.2. Background and setting ......................................................................................... 143 5.3. Theoretical Framework and Estimation methods .............................................. 147 5.3.1. Analytical Framework .................................................................................... 147 5.3.2. Models and Estimators ................................................................................... 148 5.3.2.1. Semiparametric models .............................................................................. 149 5.3.2.2. Partially linear models ............................................................................... 150 5.3.2.3. Model using a Times Series of Cross Sections .......................................... 152 5.4. Estimation Results ................................................................................................. 161 5.4.1. Model Diagnostics ........................................................................................... 162 5.4.2. Estimation results of Parametric and Semiparametric Models ................. 164 5.4.3. Estimation results of Panel Data from Successive Cross Sections ............. 167 5.5. Discussion of Estimation Results .......................................................................... 174 5.5.1. Specification tests of the functional form...................................................... 174 5.5.2. Instrumental-variable estimators .................................................................. 178 5.5.3. Testing Linear Panel-Data Models ................................................................ 188 5.6. Conclusions ............................................................................................................. 190 Chapter 6: Poverty over Time in Zambia‟s Eastern Province ................................. 193 6.1. Introduction ............................................................................................................ 194 6.2. Household Survey Data ......................................................................................... 196 6.2.1. Living Condition Monitoring Survey II 1998 ............................................... 196 6.2.2. Living Conditions Monitoring Survey IV 2004 ............................................ 198 6.3. Average Monthly Household Expenditure in 1998 and 2004 ............................ 199 6.4. Poverty Trends in Rural Eastern Province ......................................................... 203 6.4.1. Poverty Lines in Zambia ................................................................................ 203 6.4.2. Poverty Measures and Trends ....................................................................... 207 6.5. Conclusion .............................................................................................................. 218 Chapter 7: Rural Growth and Poverty Reduction through Feeder Road Rehabilitation in Chipata & Lundazi Districts of Zambia‟s Eastern Province, 19962005................................................................................................................................. 219 7.1. Introduction ............................................................................................................ 220 7.2. Context: Chipata and Lundazi Districts .......................................................... 222 7.2.1. Field Sites of the EPFRP ................................................................................ 223 x 7.2.2. Description of the Survey Sites and of the Data Collection Procedures .... 225 7.3. Theoretical Framework ..................................................................................... 231 7.3.1. Conceptual Framework .................................................................................. 231 7.3.2. Methodological Framework ........................................................................... 233 7.4. Qualitative Findings........................................................................................... 239 7.4.1. 1996 Chiwele‟s Baseline Survey Findings ..................................................... 239 7.4.2. 2005 Eastern Province Community Survey Findings .................................. 243 7.5. Quantitative Findings ........................................................................................ 254 7.5.1. Changes in Socio-Economic Characteristics, 1996-2005 ............................. 254 7.5.2. Linkage between EPFRP and Household Welfare ...................................... 258 7.5.2.1. Poverty Levels and Changes from 1996 to 2005 ....................................... 258 7.5.2.2. Decomposing poverty changes, 1996 and 2005 ........................................ 265 7.5.3. Regression Analysis ........................................................................................ 267 7.5.3.1. Consumption Growth Determinants .......................................................... 267 7.5.3.2. Poverty Determinants................................................................................. 271 7.6. Robustness of the Results ...................................................................................... 276 7.6.1. Robustness of Poverty Levels and Changes.................................................. 276 7.6.2. Sensitivity of Poverty Measures to Price Changes ....................................... 278 7.6.3. Robustness of Regression Analysis ................................................................ 280 Chapter 8: The Rural Transport Infrastructure and Marketing Linkage within the Context of the Sub-Regional Zambia-Malawi-Mozambique Growth Triangle ...... 290 8.1. Introduction ............................................................................................................ 291 8.2. Background: Sub-Regional Zambia-Malawi-Mozambique Growth Triangle. 294 8.3. Framework ............................................................................................................. 296 8.3.1. Theoretical Effects .......................................................................................... 296 8.3.1.1. Theoretical Effects of the network performance ....................................... 297 8.3.1.2. Location theory: Explaining the firm relocation process ........................... 297 8.3.2. The 1996 and 2005 Agribusiness Survey Methodologies............................. 299 8.3.2.1. 1996 Baseline Survey Methodology .......................................................... 300 8.3.2.2. 2005 Follow-up Survey Methodology ....................................................... 301 8.4. Identification of Geographical Movement of Firms due to the EPFRP ........... 303 8.4.1.1. Produce Traders Profile ............................................................................. 305 8.4.1.2. Millers Profile ............................................................................................ 308 8.4.1.3. NGOs Profile ............................................................................................. 309 8.5. The role of Regional Trade Agreements on the Market Participants............... 310 8.5.1. The Southern African Development Community ........................................ 310 8.5.2. The Sub-Regional Zambia-Malawi-Mozambique Growth Triangle ......... 312 xi 8.5.3. From the Growth Triangle to the Nacala Development Corridor ............. 313 8.6. Model/Empirical Strategy and Main Results .................................................. 315 8.6.1. Model/Empirical Strategy .......................................................................... 315 8.6.2. The Impact of the EPFRP on the Market Participants .......................... 318 8.6.2.1. 1996 Baseline Findings ......................................................................... 319 8.6.2.2. 2005 Follow-up Survey Findings .......................................................... 322 8.6.3. Robustness Checks by Comparing Models .............................................. 331 8.7. Conclusions and Policy Implications .................................................................... 333 Part Three ...................................................................................................................... 336 Conclusions and Policy Implications ........................................................................... 336 Chapter 9: The Modus Operandi of the Investment-Trade Nexus .......................... 337 9.1. Introduction ............................................................................................................ 338 9.2. Methods of Measuring the Impact of Rural Feeder Roads................................ 340 9.2.1. Generic issues in assessing the impacts of rural roads ................................ 340 9.2.2. Evaluation Methodology ................................................................................ 342 9.3. The Empirical Findings ......................................................................................... 346 9.4. Changes in policy on transport infrastructure investment in Zambia ............. 349 9.5. Policy Implications ................................................................................................. 353 9.6. Future Research ..................................................................................................... 356 References ...................................................................................................................... 360 Annexes .......................................................................................................................... 379 Annex: Chapter 1 .......................................................................................................... 380 Annex: Chapter 2 .......................................................................................................... 382 Annex: Chapter 3 .......................................................................................................... 387 Annex: Chapter 4 .......................................................................................................... 398 Annex: Chapter 5 .......................................................................................................... 411 Annex: Chapter 6 .......................................................................................................... 422 Annex: Chapter 7 .......................................................................................................... 442 Annex: Chapter 8 .......................................................................................................... 458 Annex: Chapter 9 .......................................................................................................... 468 Fieldwork Documentation ............................................................................................ 471 xii Figures Chapter 1 Figure 1.1: The Transmission Mechanisms between Infrastructure, Sustainable Pro-poor Economic Growth, and Agricultural Trade Figure 1.2: The Basic causality paradigm of the relationship between transport infrastructure investment and economic development Chapter 2 Figure 2.1: The Relationship between Trade, the Development of Productive Capacities, Employment and Poverty Chapter 3 Figure 3.1: Rainfall pattern in Zambia‘s Eastern Province, 1994-2005 Chapter 4 Figure 4.1: Yield of Selected Cash Crops in Zambia, 1996-2005 Fig. 4.2a: ‗Trends‘ in the Log of Cotton Yield for Treatment & Control group, 1996-2001 Fig. 4.2b: ‗Trends‘ in the log of Cotton Production for treatment & control group, 19962001 Chapter 5 Fig.5.1: Household Expenditure by Age, 1998 & 2004 Fig.5.2: Household Expenditure by cohort, 1998 & 2004 Chapter 6 Figure 6.1: Cumulative Density of pae expenditure in 1996 Figure 6.2: Cumulative Density of pae expenditure in 2005 Figure 6.3: Cumulative Density of pae expenditure in 1996 Figure 6.4: Cumulative Density of pae expenditure in 2005 Chapter 8 Figure 8.1: Conceptual model of an individual firm in a physical environment Figure 8.2: Distribution of firms in the 2005 Agribusiness Survey dataset xiii Tables Chapter 2 Table 2.1: Long-term effect of rural roads identified by literature Chapter 3 Table 3.1: Eastern Province Road Sector Network, (km) Table 3.2: Eastern Province Feeder Road Network Table 3.3: Types of endogeneity of programme placement associated with the existence of third variables (X) Chapter 4 Table 4.1: Share of Cotton Area in Total Cropped Area, 1993-1998 Table 4.2: Percentage of Farmers Growing Cotton in Eastern Province, 1997 – 2002 Table 4.3: Fraction of Land Allocated to Cotton, 1997 - 2002 Table 4.4: Yields per Hectare in Cotton (MT/HA), 1997 - 2002 Table 4.5: Percentage of Households that Grow Maize, 1997 – 2002 Table 4.6: Frequency of Holdings in Eastern Province, 1996-2002 Table 4.7: Post Harvest Survey, Sample sizes by District, 1997-2002 Table 4.8 Descriptive Statistics, 1996/1997 – 2001/2002 Table 4.10c: Matching and Propensity Score Estimators Table 4.10d: Matching estimators for Average Treatment Effects Table 4.11 Basic Log-Productivity & Log-Production Regressions Table 4.12: Tobit Model Comparisons of Log Likelihood Table 4.13 Cotton Yields: Impacts of the EPFRP Baseline Regressions Table 4.14: Comparison of Cotton Productivity Estimation Models Table 4.15: Cotton Productivity Non-Linearity of Unobserved Productivity Table 4.16: Covariate imbalance testing Table 4.17a: The Baseline ATT estimation (with no simulated confounder) Table 4.17b: ATT estimation with simulated confounder Table 4.18: Sensitivity to the Definition of EPFRP and Reassignment of Implementation Year Chapter 5 Table 5.1: Sample Allocation (Standard Enumeration Areas -Primary Sampling Units) Table 5.2: Principal Economic Activity of Household Head, Rural Areas, Numbers of Household Heads by Quintile of Consumption Table 5.3: Percentage of Households in Rural Areas Owning Particular Assets by Quintile Table 5.4: Mean distance to services and Community Assets, by Household, Rural Areas Table 5.5: Ranking of Consumption Growth and Poverty Change at the Constituency level, 1998 & 2004 Table 5.6: Number of Persons in Selected Cohorts by Survey Year, 1998 & 2004 Table 5.7: Descriptive Statistics of the Data covering Rural Eastern Province Table 5.8: Descriptive Statistics of covariates, 1998 and 2004 xiv Table 5.9: Comparison of Generalized linear models, 1998 and 2004 Table 5.10: Partial Linear regression models, 1998 and 2004 Table 5.11: Pooled OLS with and without cluster-robust standard errors Table 5.12: Between Estimator with Default Standard Errors Table 5.13: RE Estimator: Comparison of cross-sectional time-series regression models. Table 5.14a Panel Estimator, Comparison, Catchment Table 5.14b Panel Estimator Comparison, Control Table 5.15: First-Differences Estimator with Cluster-Robust Standard Errors Table 5.16: Testing for heteroskedasticity in the IV context in 1998 and 2004 Table 5.17: Weighted Least-Squares Estimator vs OLS estimator, 1998 & 2004 Table 5.18: Durbin-Wu-Hausmann Tests for Endogeneity in IV estimation Table 5.19.1 Quantile 1: 20%, 1998 Table 5.19.2 Quantile 1: 20%, 2004 Table 5.19.3 Quantile 4: 80%, 1998 Table 5.19.4 Quantile 4: 80%, 2004 Table 5.20a: Hausman Test for Fixed Effects, Catchment areas Table 5.20b: Hausman Test for Fixed Effects, Control Areas Chapter 6 Table 6.1: Relevant Zambian Household Surveys since 1991 Table 6.2: Sample Allocation (Standard Enumeration Areas - Primary Sampling Units) Table 6.3: Average Monthly Household Expenditure by Rural/Urban, Rural Stratum and District (ZMK), 1998 & 2004 Table 6.4a: Mean Per Adult Equivalent for different groups in Eastern Province Table 6.4b: Mean per adult equivalent for different groups in Eastern Province Table 6.4c: Mean per-capita incomes in real terms in Zambia‘s Eastern θrovince Table 6.4d: Mean per-capita incomes in real terms in Zambia‘s Eastern θrovince Table 6.5: Zambian Poverty Lines Table 6.6: Headcount Poverty Estimates in Rural Eastern Province, 2004 and 1998 Table 6.7: Poverty Gaps in Rural Eastern Province, 2004 and 1998 Table 6.8: Squared Poverty Gaps (Severity of Poverty) in Eastern Province, 2004 & 1998 Table 6.9: Cumulative Rural total household expenditure by decile, 2004 Table 6.10: Cumulative Rural total household expenditure by decile, 1998 Table 6.11: Cumulative Rural total household expenditure by Quintile, 2004 Table 6.12: Cumulative Rural total household expenditure by Quintile, 1998 Chapter 7 Table 7.1: Population Size and Average Growth Rates Table 7.2: Total Length of Primary Roads Rehabilitated in Chipata District Table 7.3: Total Length of Feeder Roads Rehabilitated in Lundazi District Table 7.4: Criteria for Stratification of rural Households Table 7.6: Survey Sites used by the LCMS I 1996 and EPRHS 2005 Table 7.7: Geographic & Demographic Characteristic of Communities, 2005 Table 7.8: Survey Data for Evaluation of the EPFRP, 1996 & 2005 Table 7.9: Poverty Lines and Implied Inflation Rates Table 7.10: Distances to Public Services, 2005 xv Table 7.11: Transportation in Survey Communities, 2004/2005 Table 7.12: Impact of Eastern Province Feeder Road Project, 2004/2005 Table 7.13: Labour Market Issues in Survey Communities, 2004/2005 Table 7.14: Agricultural Situation in Communities, 2004/2005 Table 7.15: Communities Perception of Determinants of Standard of Living, 2005 Table 7.16: Communities own Perception of the Determinants of Poverty, 2005 Table 7.17: Descriptive Statistics for the Treatment and Comparison households, 1996 and 2005 (%) Table 7.18: Poverty Levels, 16 PSUs (1996: n=241 & 2005: n=88) Table 7.19: Changes in Comprehensive Total Expenditure and Poverty, 1996 and 2005 (%) Table 7.20: Measures of Inequality of per adult Expenditure, 1996 and 2005 Table 7.21: Decomposing poverty changes per adult equivalent, 1996 and 2005 Table 7.22: Determinants of Total Household Expenditure in per adult equivalent, 1996 & 2005 Table 7.23: Determinants of Poverty Table 7.24: Predictions after Tobit, 1996 & 2005 Table 7.25: Food poverty levels 1996-2005; 16 PSUs (1996: n=241; 2005: n=88) Table 7.26: Total poverty levels 1996-2005; 16 PSUs (1996: n=241; 2005: n=88) Table 7.27: Upper poverty levels 1996-2005; 16 PSUs (1996: n=241; 2005: n=88) Table 7.28: Lower poverty levels 1996-2005; 16 PSUs (1996: n=241; 2005: n=88) Table 7.29: Sensitivity of Poverty Measures to Price Changes Table 7.30a: Poverty Comparisons with a common Food Poverty line for ‗real‘ comprehensive household Expenditure Table 7.30b: Poverty Comparisons with a common Total Poverty line for ‗real‘ comprehensive household Expenditure Table 7.31: Tobit on Log Poverty Gap Transformation Table 7.32: Tests of Normality and Homoskedasticity, 1996 & 2005 Table 7.33: Two-part model in logs, 1996 & 2005 Table 7.34: Heteroskedasticiy & non-normality, 1996 & 2005 Chapter 8 Table 8.1: Mobility profiles in Chipata and Lundazi Districts, 2005 Table 8.2: Overlapping Membership in Regional Integration Groups, 2009 Table 8.3: Major Regional Integration Initiatives Table 8.4: Frequency of the propensity to migrate (Migration) Table 8.5: Firm relocation by stratum, 2005 Table 8.6a: Empirical results Table 8.6b: Empirical results Table 8.7: Comparing Full with Reduced Logistic regression Models Chapter 9 Table 9.1: Summary of Empirical Findings xvi Maps Chapter 3 Map 3.1: Illustration of the Eastern Province Feeder Roads Annex Chapter 4 εap A1μ Zambia Eastern θrovince‘s District Roadmap Annex Chapter 5 Map A1: The major soil types of Eastern Zambia Map A2: Agroecological zones and agricultural districts, Eastern Province, Zambia Annex Chapter 6 Map A1: LCMS-I 1996 & EPRHS 2005 SEAs εap Aβμ ωhipata District, Zambia‘s Eastern θrovince Map A3: Lundazi District, Zambia‘s Eastern θrovince Map A4: Chipata Survey Sites Map A5: Lundazi Survey Sites xvii Annexes Annex: Chapter 1 Annex: Chapter 2 Annex: Chapter 3 Annex: Chapter 4 Annex: Chapter 5 Annex: Chapter 6 Annex: Chapter 7 Annex: Chapter 8 Annex: Chapter 9 xviii Abbreviations ADB AEC AfDB AMIS AMRF ASIP BoZ BWIs CBA CBO CEA CGIAR COMESA CPI CSO CTA DAC DBZ DC DCU DDCC DDP DISS EAGER EB ECA EGS EIP EPCU EPFRP EPZ ERP ESAF EU FAO FFW FNDP FRP FRS FTA FTC GATT GDCF GDP GoRZ HDI HDR HIPC HS IBRD IDA IFPRI IFIs ILO Asian Development Bank African Economic Community African Development Bank Agriculture and Market Information System Agricultural Marketing Revolving Fund Agricultural Sector Investment Programme Bank of Zambia Bretton Woods Institutions (World Bank & International Monetary Fund) Cost Benefit Analysis Community Based Organisation Cost Effectiveness Analysis Consultative Group on International Agricultural Research Common Market for East and Southern Africa Consumer Price Index Central Statistical Office Chief Technical Advisor Development Assistance Committee (OECD) Development Bank of Zambia District Council District Cooperative Union District Development Consultative Committee District Development Project Department of Infrastructure and Support Services Equity and Growth through Economic Research Equipment Based (construction methods) Economic Commission for Africa Employment Guarantee Scheme (India) Employment Intensive Programme Eastern Province Cooperative Union Eastern Province Feeder Roads Project Export Processing Zone Economic Recovery Programme Enhanced Structural Adjustment Facility European Union Food and Agricultural Organization Food-for-work (programmes) Fifth National Development Plan (2006-2011) Feeder Road Project (in Eastern Province – EPFRP) Feeder Roads Section, DISS/MLGH Free Trade Area First Trial Contract General Agreement on Tariffs and Trade Gross Domestic Capital Formation Gross Domestic Product Government of the Republic of Zambia Human Development Index Human Development Report (UNDP) Highly Indebted Poor Countries Harmonized System (of Tariffs) International Bank for Reconstruction and Development International Development Association International Food Policy Research Institute International Financial Institutions (World Bank and IMF) International Labour Organization xix ILO/ASIST IMF I-PRSP IRR LBT LDCs LFA LINTCO LOR MAFF (MACO) MDGs MMD MoF (MFNP) MWS MTEF NAMBoard NCC NDCC NCDP NEPAD NERP NGO NMT NORAD NPV NTE NRB NRFA ODA OECD OER PAM PHS PIP PMU PPP PRA PRGF PRGS PRSP PTA PWP RAMB RCP RCS RCT REER RMI RNFE ROADSIP RTS SADC SAM SAP SIO SIP Advisory Support Information Services and Training International Monetary Fund Interim Poverty Reduction Strategy Paper Internal Rate of Return Labour Based Technology (construction methods) Least Developed Countries Logical Framework Analysis Lint Company Line of Rail Ministry of Agriculture, Food and Fisheries (Ministry of Agriculture and Cooperatives) Millennium Development Goals Movement for Multi-Party Democracy Ministry of Finance (Ministry of Finance and National Planning) Ministry of Works and Supply Medium Term Expenditure Framework National Agricultural Marketing Board National Council for Construction National Development Coordinating Committee National Commission on Development Planning New Partnership for African Development (African Union) New Economic Recovery Programme Non-Governmental Organization Non-Motorised Transport Norwegian Development Agency Net Present Value Non-Traditional Export National Roads Board National Road Fund Agency Official Development Assistance Organisation for Economic Cooperation and Development Official Exchange Rate Programme Against Malnutrition Post-Harvey Surveys Public Investment Programmes Project Management Unit Purchasing Power Parity Participatory Rural Appraisal method Poverty Reduction and Growth Facility Poverty Reduction and Growth Strategy Poverty Reduction Strategy Paper Preferential Trade Area Public Work Programmes Rural Agricultural Marketing Board Regional Co-operation Policy Repeated Cross-Section data Randomized control group trial Real Effective Exchange Rate Road Maintenance Initiative Rural Non-Farm Economy Road Sector Investment Programme Roads Department Training School Southern Africa Development Community Social Accounting Matrix Structural Adjustment Programme Semi-Input-Output Model Sector Investment Programme xx SSA STC UN UNCDF UNCTAD UNDP UNIP VAT WFP WTO ZAMSIF ZANU ZCCM Sub-Saharan Africa Second Trial Contract United Nations United Nations Capital Development Fund United Nations Conference on Trade and Development United Nations Development Programme United National Independence Party Value Added Tax World Food Programme World Trade Organization Zambia Social Investment Fund Zambia National Farmers Union Zambia Consolidated Copper Mines Table of Units C.i.f. MTs Km HA HH Kcal US$ ZMK Cost, insurance, freight Metric Tonnes Kilometres Hectare (10,000 sq. m) Household Kilo Calories United States Dollar (USD) Zambian Kwacha (ROE: USD = ZMK4300 – March 2002) xxi Overview Abstract of Chapter 2: Ravallion and Jalan(1996) propose that theories of economic growth offer some clues as to why some areas are persistently poor over a long period, whereas others have escaped the spatial poverty trap. For this reason we start our literature review by looking at the mainstream economic growth theories, including the standard Solow-Swan ―Exogenous‖ growth model and a somewhat richer set of explanations for poor areas, which according to Ravallion and Jalan(1996, 1997, 1998) can be found in the more recent ―endogenous growth models.‖ Sala-I-Martin(1997) finds that of the 62 (59) growth determinants found in the cross-country regression empirical growth literature ββ regressors appear to be ―significant.‖ Although the influential study by Barro(1990) triggered an avalanche of macroeconometric studies of the long-run effects of public infrastructure investment on macroeconomic performance, the empirical estimates of the productivity impact are inconclusive, ranging from no effect to rates of return in excess of 100% per annum. In the end the cross-country regressions don‘t seem to be the best tools for analyzing the problem of understanding this linkage. Micro-level studies have also led to doubt about the benefits to the poor of the assets created through Public Works Programmes. This doubt could according to Van de Walle( 2002, 2009) be explained by the existing methods of project appraisal for rural roads, which don‘t properly reflect the potential benefits to the poor. The methodological framework used in public projects evaluation has been rehabilitated considerably thanks to the recent introduction of propensity score matching techniques developed by Rosenbaum and Rubin(1983), which has been incorporated into to the analysis of the social and economic impacts derived from rural roads rehabilitation and maintenance projects. Estache(2010) concludes based on the most recent survey of the impact of infrastructure that not all infrastructure interventions are suitable for impact evaluations based on experiments or quasi experiments. For those unsuitable cases, there are other ways of generating robust quantitative evaluation of the effectiveness of infrastructure interventions which can be just as effective in increasing accountability for intervention selection, implementation and sustainability. Abstract of Chapter 4: This chapter investigates the dynamic impacts of rural road improvements on farm productivity and crop choices in Zambia‘s Eastern θrovince. There are several channels through which the feeder road improvements impact on farmers. Our aim is to estimate whether the differential outcome in the five treatment districts and three control districts generated by the expansion of market agricultural activities among small to medium scale farmers could be explained by rural road improvements that took place after the new Chiluba MMD government in 1995 had completed an IMF rights accumulation programme bringing the principal marketing agent system to an end (Pletcher, 2000). Our district-level empirical analysis is an extension to the Brambilla and Porto(2005, 2007) cross-provincial level approach which proposes a dynamic approach accounting for entry and exit into the agricultural cotton sector to avoid biases in the estimates of aggregate productivity, when measuring productivity in agriculture applied to a repeated cross-sections of farm-level data from the Zambian post-harvest survey (PHS). We use PHS data covering the period from 1996/1997 to 2001/2002 when the Eastern Province Feeder Road Project (EPFRP) was being implemented. The identification strategy relies on differences-indifferences of outcomes (i.e., cotton productivity) approach across two phases (pre-treatment and posttreatment). We use maize productivity to difference out unobserved household and aggregate agricultural year effects. By implementing respectively matching estimators for average treatment effects; differences-todifferences estimators as well as censored regression models we find at the district level that the improved accessibility from the EPFRP led to changes in land allocation and in yields to the Eastern θrovince‘s most important cash crop – cotton. Key words: Pooled Repeated cross-section post-harvest surveys, Impact evaluation of cash crop choice and cotton productivity; Africa; Zambia (Eastern Province). JEL-codes: C2; C83; D2; O12; O13; Q12; R3. xxii Abstract of Chapter 5: Remarkably little is known about the long-term impacts of project aid to lagging poor areas (Chen, Mu et al. 2006, 2008). This chapter contributes to the debate about the role of rural transport infrastructure development in explaining the long-term rural development. In line with Grimm and Klasen(2008) we agree that there is value-added to consider this debate at the micro level within a country as particularly questions of parameter heterogeneity and unobserved heterogeneity are likely to be smaller than between countries. Moreover, at the micro level it is possible to identify more precise transmission mechanisms from rural transport infrastructure to socio-economic development outcomes. This is done empirically by analyzing a UζDθ&UζωDF financed rural development project in Zambia‘s Eastern θrovince running from 1997-2002. The secondary datasets consist of respectively a series of repeated cross-sectional living conditions monitoring surveys (LCMSs). The LCMSs were collected in 1998 (baseline) and 2004 (follow-up), that is both prior, during and after the project implementation. Our aim is to assess the ability of a few parametric and semi-parametric models as well as using a time- series of cross-sections to provide an adequate description of the logarithm of per adult equivalent consumption of rural household conditional on few covariates, including an infrastructure treatment dummy variable. Although, the mean cotton sales share of household income has more than doubled despite the fact that the mean distance to the input market remained unchanged from 1998 to 2004, the parametric and semi-parametric estimation results are only small and statistically insignificant in terms of gains to mean consumption emerged in the longer-term. The main results are robust to corrections for various sources of selection bias. Keywords: Parametric and Semi-parametric Regression Models; Time-series Model from Successive cross sections; Cohort Data; Poverty Measures, poor rural area development projects, feeder roads, household surveys, impact evaluation, Zambia. JEL: C14, C21, D12, I32, O1, Q1. Abstract of Chapter 6: This chapter explores the linkage between poverty and rural roads improvements through an analysis of the poverty impact of the rehabilitation of a large share of the feeder road network in Zambia‘s Eastern Province. This is done by using the Deaton(1997) Stata codes complemented by the use of the World ψank‘s Automated DEω‘s θoverty Tables (ADeθT) 4.1 θoverty εodule to estimate poverty rates, and inequality and welfare indices for Zambia‘s Eastern θrovince covering the period from 1998 to β004. We show that: (1) the pro-poor growth rate of rural Eastern province was positive. (2) The reduction of the headcount poverty rate from 1998 to 2004 was much faster in the rural areas compared to the urban areas. (3) The very unequal expenditure distribution in 1998 had not improved much in 2004. (4) The inequality was higher in the pre-treatment region in 1998 compared to the control region, but this was no longer the case in 2004 after the Eastern Province Feeder Road Project (EPFRP) treatment from 1998 to 2001. (5) Finally, the biggest decline in inequality in the rural areas between 1998 and 2004 is found in the decile dispersion ratio. Key words: Africa; Zambia (Eastern Province); Poverty; Inequality; Rural households; Comparison of two independent cross-sectional Surveys; Inter-temporal poverty and inequality analysis; Geographic distribution. JEL: C8; I3; R2. xxiii Abstract of Chapter 7: ―Assessing changes in poverty levels over time is bedevilled by problems in questionnaire design, the choice of the poverty line, the exact timing of the survey and uncertainty about the appropriate cost-ofliving deflators (Dercon & Krishnan, 1998).‖ In this chapter, we adopt a methodology of estimating the contribution of a rural feeder roads project to regional development, which requires a counterfactual situation at the sub-district level to be set up to answer questions pertaining to the nature of the spatial structure in the absence of the rural feeder roads project (EPFRP 1996-2001). We focus on testing the robustness of measured changes in poverty to these common problems, using a constructed rural household pseudo-panel dataset based upon a baseline cross-section Living Conditions Monitoring Survey (LCMS-I) dataset collected by Zambia‘s ωentral Statistical ηffice in 199θ and our own Eastern Province Rural Household Survey (EPRHS), which was carried out in 2005 in exactly the same 16 Standard Enumeration Areas as the baseline LCMS I. Through a qualitative analysis we find that between 1996 and 2005 access to external markets through the rehabilitated feeder road network, is a critical determinant of the rural households‘ ability to increase their income and improve their chances of breaking out of the poverty trap in the medium to long run. However, there are indications of considerable impact heterogeneity between the survey sites. Unfortunately, it‘s much harder to come to the same clear-cut conclusion through our quantitative microeconometric approach, partly because it is based upon a very small sample size that together with the usual measurement errors doesn‘t validate the qualitative findings in a robust manner. The difficult identification of impact is also partly explained by the number of roads (78%) that remained outside the EPFRP. Notwithstanding these possible explanations, the statistical significance of the association is not robust to different choices of price deflators and specifications tests for the Tobit Model. Therefore any policy implications from these ambiguous results should be done with great caution. Key words: Africa; Zambia (Eastern Province: Chipata & Lundazi districts); Poverty; Inequality; Rural households; Pseudo-Panel data at Standard Enumeration Area level; Comparison of two independent crosssectional household surveys; Inter-temporal poverty analysis; Rural Transport Infrastructure and Economic Development Impact; Geographic distribution. JEL: C8; D1; H4; H5; I3; O12; Q1; R2. Abstract of Chapter 8: In our overall quest to evaluate the long-term direct benefits of the Eastern Province Feeder Road Project (EPFRP) carried out in the period from 1996 to 2001, we want to investigate whether the accessibility improvements have translated into increased marketing activities. Were more traders of different sizes induced to visit the villages within the zone of influence of the feeder roads e.g. through their decision to either operate (start-ups), relocate (attraction), expand activities in or stay (retention) in ωhipata and/or δundazi districts in Zambia‘s Eastern θrovince. We consider how the reductions in the cost – broadly defined – of movement has affected the economic activities of private companies, individual traders, input suppliers, millers and NGOs in Chipata and Lundazi districts. The 1996 baseline study by Chiwele et al.,(1996) find that the nature of the infrastructure and the road network connecting the surplus remote areas and the deficit regions in 1996 was one of the key reasons why the new grain marketing system at the end of this period remained undeveloped at the end of 1996. From our own 2005 follow-up survey we find that it is primarily small private companies that are more likely to have moved into these two districts as a consequence of the EPFRP. On the other hand, the entry by the large South African Clark Cotton Company in 1995 was linked to Zambia and South Africa‘s SADC membership. In addition, it was mainly the ‗small-scale private companies‘ engaged in ‗agricultural marketing‘ which had made a positive relocation decision. The Growth Triangle doesn‘t seem to have played any significant role on the trade prospects of the business community from 1999 to 2005. This seems bound to change now that the last phase of the Nacala corridor from Mchinji to Chipata has been completed in 2010. Key words: (Discrete Choice) Logit Model; Business survey (descriptive) statistical approach; Zambia (Eastern Province: Chipata & Lundazi districts); Transport Infrastructure; Accessibility; Localization; Site Selection; Decision Making; Agricultural Marketing; Transportation; Regionalism; Firm level survey data. JEL-codes: C1; D21; F14; F15; L91; N57; N77; 055; Q13; Q17; R. xxiv Part One Introduction and Approaches to Assess the Effect of Public Capital Infrastructure on Economic Growth 0 Chapter 1: Policy Evaluation of the InvestmentTrade Nexus 1 Where a road passes, development follows right on its heels The World Bank, 2006. 1.1. Introduction The thesis aims to deconstruct the rural growth process in order to empirically establish the structural relationships between rural transport infrastructure (RTI) development, pro-poor rural growth and agricultural trade in the medium to longterm. In other words, the purpose of the thesis is to focus on the important developmental role that rural roads improvements can play in transforming a predominantly subsistence agriculture-based economy characterized by generalized poverty into a market-based rural economy where agricultural no longer is considered as a way of life, but rather as a commercial outward oriented profession. The potential linkages between rural roads improvements and transformative rural development as measured by three specific dimensions, namely, agricultural productivity; poverty alleviation; and agricultural trade (i.e. agribusiness relocation) are tested by using a range of different micro-econometric reduced-form methods. The importance of infrastructure as an instrument of economic development and, potentially, poverty reduction, is reflected in the high level of investment, which national governments and international donor agencies put into infrastructure development.1 For example, the United Nations (UN) identified 5 clusters to support and respond to the ζew θartnership for Africa‘s Development (ζEθAD) Action plan, from which Infrastructure Development, and Agriculture, Trade and Market Access constitute two of the most important clusters. Likewise the report of the Commission for Africa(2005) called for a doubling of official development assistance (ODA) to Sub-Saharan Africa (SSA),2 including an investment of $150 billion in infrastructure over a decade. Recently, a consortium of donors, which has established the African Infrastructure Country Diagnostic (AICD) similarly calls for $31 billion annually on infrastructure spending in Africa (Gollin and Rogerson, 2010; Foster and BriceñoGarmendia, 2010). 1 Investment in infrastructure was found to represent around 20 per cent of total investment in a sample of low and middle income countries, and to account for 40 - 60 per cent of public investment (UNCHS, 1996). Individuals, private firms and NGOs have also made substantial investments. 2 ωf. UζωTAD. β00θ. Economic Development in Africa. Doubling Aidμ εaking the ―ψig θush‖ work. 2 The Commission for Africa in particular argues that investing in assets such as rural roads,3 and a transport network, in addition to health and education, can lead to growth and job creation, helping Africa make progress towards the Millennium Development Goals (MDGs). However, in order to achieve the reduction of poverty by half in the year 2015 (i.e. MDG1), the Economic Commission for Africa (ECA) in 1999 estimated the growth rate of GDP per capita required for Southern African countries to be 6.2% per annum (UNECA, 2003). This according to Estache(2006) also corresponds to annual estimated new infrastructure and maintenance requirements for all of Africa of about 9% of GDP, or equivalently US $40 billion between 2005 and 2015 (Estache, 2006). Despite the fact that rural infrastructure has become a major development priority (World Bank, 1994, Commission for Africa, 2005; Foster and BriceñoGarmendia, 2010; G20, 2010), yet little is known about the size and especially the distribution of benefits from such investments in Least Developed Countries (LDCs). Roads are particularly important forms of rural infrastructure, providing cheap access to both markets for agricultural output and for modern inputs (see figure A2). Given limited policy instruments for reaching the remote rural poor, road building at first glance seems desirable on distributional grounds (Jacoby, 2002). Along this line of thought, the ILO(2004) argues that if poverty is largely rural and rural employment is mainly in agriculture, then it seems likely that the parts of the growth process that are linked to rural areas, and especially those related to agriculture, may have more immediate and direct effects on poverty reduction than would growth outside rural areas. Whether there is such a linkage between RTI, Agricultural Trade, Rural Growth and Poverty Reduction, this is the key question the thesis attempts to demonstrate empirically. 1.2. Overview of the Major Issues Improved road infrastructure can create opportunities for economic growth and poverty reduction through a range of transmission mechanisms. While these linkages ―Rural roads are defined as those where improving mobility settlements are an important feature …within any country the number of major roads will be low so that for most rural people it is access to a larger village or district town which is important (Howe & Richards, 1984μ1).‖ 3 3 are often direct, many are indirect and visible only over time, which leads us to address issues of long-term dynamics. Given Zambia‘s developmental challenges, especially the high poverty levels, there is a real need for stepping up efforts aimed at strengthening and broadening the growth process. Zambia‘s Fifth ζational Development θlan (FζDθ) covering the period 2006-2011 therefore identified two critical areas where public spending should be focused if growth is to be accelerated and broadened. These were: (a) Strengthening the relevant economic and social infrastructure; and (b) Enhancing agriculture and rural development (GoRZ, 2005), which are testimony to the policy relevance of this thesis. We consider in turn the role of economic growth, employment-intensive public works programmes (PWP) and agriculture. 1.2.1. Long-Run Growth One of the central debates in poverty reduction concerns the role of economic growth. For decades academics hotly debate the size of the relationship between level of national income and poverty, and between growth of income and reduction in poverty (chapter 2). While it is generally agreed that poverty reduction strategies cannot succeed if they are not accompanied by policies to sustain rapid growth and improve income distribution (Dagdeviren et al., 2001, Lubker, 2002), it is precisely the content and nature of those policies that are at the heart of the debate (UNCTAD, 2002a). According to Winters et al.,(2004) sustained growth requires increases in productivity, and most of the evidence suggests that trade liberalization operates through this route (chapter 4). The long-term growth rate depends on governmental actions, such as provision of infrastructure services, and regulations of international trade. The government therefore has great potential for good or ill through its influence on the long-term rate of growth (Barro and Sala-i-Martin, 1995). While acknowledging that the role of government in promoting growth and development continues to be ideologically polarized (Akyuz and Gore, 1994), we are interested in exploring the potential role of RTI in a strategy of growth based on farm and nonfarm rural sectors (Khandker et al., 2006). 4 The benefits that roads bring to rural areas are often seen as so obvious in the development literature that they are listed rather than discussed as in (Wilson, 2004).4 Nevertheless, there are three main reasons for taking the infrastructure and the construction sector as a catalyser for labour-based pro-poor growth:    The lack of up-gradation of productive, social and access infrastructure retards economic development, and generally isolates poorer remote communities. Infrastructure represents a significant proportion of GDP, public investment, and donor support in developing countries (World Bank, 1994; UNCTAD, 2008).5 The potential for labour absorption is particularly high in this sector: LabourBased Technology (LBT) methods account for 50-60% of total costs in several country case studies (Islam&Majeres, 2001:14-15 qtd. in Devereux, 2002:26). 1.2.2. Rural Development through Public Works Programmes Trade development depends on macroeconomic policies, non-trade policies as well as trade policies. Particularly important in this regard are the so-called complementary non-trade policies, which promote the development of productive capabilities through capital investment (UNCTAD, 2004). Public investment programmes (PIP) are generally considered as public policy instruments aimed at producing economic, social and transport infrastructure outputs. They present one of the few remaining government policy instruments through which productive employment opportunities can be created and more economically and socially balanced development promoted (ASIST, 2003b; Edmonds and Howe, 1980, von Braun et al., 1992)). Thus, with appropriate policies, the poor can participate in growth and contribute to it, and when they do, rapid declines in poverty are consistent with sustained growth (UNECA, 2003). Although, the PRSPs have underlined the need to explore the interconnections among growth, an explicit focus on the distributional impact or the poverty dynamics of government expenditure side of the budget is an 4 Fan&Chan-Kang(2004) notice that rural road projects do not always improve the well-being of local communities and help the poor as discussed e.g. in Riverson et al (1991). 5 Rural transport infrastructure in SSA carries about γγ percent of the region‘s GDθ, mainly agricultural goods, and 40% of the export products (Malmberg-Calvo, 1998 ref. in Banjo, 1999). 5 idea, which started being promoted within the PRSP process.6 That is to know whether investment in the rehabilitation of feeder roads will benefit the very poor (Bigsten and Shimeles, 2003, Levinsohn, 2003, UNCTAD, 2002a; NRFA, 2010b). A country‘s rural road network is normally made up of tracks, trails, footpaths and earth roads. The tracks, trails and footpaths defined as „non-motorized (rural) roads‟ are characterized by low quality standards and limited transit. A second type of road, which is studied in this thesis, is the „motorized (rural) roads‟ also known as rural feeder roads. These roads are engineered earth roads used to connect rural areas by public transport or cargo trucks to the urban districts centres via the connection to secondary district roads (Escobal and Ponce, 2003, Barwell, 1996, Howe and Richards, 1984). In fact, road investment constitutes a major portfolio of public investment in rural areas, reinforcing the notion that rural income and productivity growth depend critically on roads and other public investments (Khandker et al., 2006, Van de Walle, 2002, Howe, 2001). Currently, some kind of PWP exists in virtually every country of the world, but Keddeman(1998) emphasises that differences in purpose, scope, content and procedures may well be more significant now than are similarities between the countries and programmes. A basic distinction must be drawn between two types of publicly-funded employment programmes:  The labour-intensive employment programmes, which maximise short-term employment creation, usually as a response to crisis or as a self-targeting means of  identifying the poor for income transfers. The labour-based employment programmes, which focus as much attention on the secondary objective of asset creation – especially infrastructure creation or maintenance – as on the primary objective of employment creation. Though similar in many ways, these two approaches have very different implications for project design (Devereux, 2002, Edmonds and Howe, 1980, Edmonds and de Veen, 1992).7 6 The MDGs indicators are an integral part of the PRSP Monitoring Indicators (MoFNP, 2004a). Both approaches have an impact on poverty: The former is purely short-term, with immediate impact on poverty alleviation but having little effect on sustainability or capacity-building; whereas the latter has a more sustained impact, with a longer-term and structural effect on employment creation and poverty reduction through the substitution of capital by labour within mainstream PIPs. 7 6 The relation between poverty reduction and rural infrastructure provision has been discussed from a macro perspective by various authors. Some leading proponents (World Bank, 1994, Ahmed and Donovan, 1990) point out the existence of strong linkages between RTI, agricultural growth and poverty reduction. These studies draw evidence from South East Asian countries like Indonesia or Malaysia, where a massive increase of rural infrastructure was followed by a long period of economic growth and a reduction in rural poverty. Although the causal connection is not clearly established, Ahmed and Hossain (1990) suggest this would have happened as a result of the impact of infrastructure investment on the rise of agricultural productivity and the creation of new job opportunities. 1.2.3. The Role of Agriculture in Development and Structural Transformation In many developing countries, agriculture is still the dominant economic activity, accounting for large shares of employment and output (Gollin and Rogerson, 2010, ILO, 2004). In Zambia with its abundant unskilled labour, great endowment of natural resources, and relatively favourable climate, agriculture has provided, directly and indirectly, the bulk of employment whereas mining, directly and indirectly, has generated a major part of the country‘s income and wealth (Hill and McPherson, 2004, OECD, 2008). This has been a well-established pattern since modern economic growth began in the country (cf. notion of „a turning point‟ by Reynolds, 1985). Nevertheless, agricultural growth has been well below potential since the late 1970s, due to a number of constraints, including an accumulation of poor past policies (i.e. path dependency). Officially, the overall sector strategy in the Zambian 20022004 PRSP was to promote a self-sustaining, productive and competitiveexport-led agriculture sector, which ensures increased household income, food security, and creation of employment opportunities. This is the fundamental assumption behind the planned economic growth under the PRSP and the FNDP.8 8 The list of known agricultural products for export where Zambia has both comparative and competitive advantage includes e.g. coffee, cotton, groundnuts, flowers, paprika, etc., which large-scale and small-scale producers (under out-grower schemes) were encouraged to produce (GRZ, 2002b). 7 This is to be achieved through: Promotion of small-scale and large-scale commercial agriculture;9 land (tenure system) and infrastructure development;10 technological development and provision of agriculture extension services for food security; and targeted support system for food security (Ministry of Finance and National Planning, 2004, GRZ, 2002b, 2005).11 Moreover, with regards to Zambia‘s Eastern Province specifically, the FNDP states that in order to boost socio-economic development, the development of infrastructure in agriculture, education and health will receive the highest priority during the FNDP 2006-2011 period. The provincial strategic focus is to ‗increase the linkages between agriculture and related industries with a view to create jobs in order to reduce the high poverty levels in the Eastern Province (GoRZ, 2005). In addition, the literature reviewed in chapter 2 notes several supplementary reasons why so many individuals are involved in subsistence agriculture, including such things as various barriers which impede the growth of the nonagricultural sector (Gollin and Rogerson, 2010).As agricultural growth faltered, governments and donors looked to the rural non-farm economy (RNFE) to provide future growth in incomes and employment.12 It was hoped that policy reforms associated with Structural Adjustment Programmes (SAP) would create a more enabling environment for the growth in RNFE (Hazell and Hojjati, 1999). However, an IPFRI study by Hazell and Hojjati(1999) shows that the regional growth linkages are surprisingly strong in Zambia‘s Eastern Province. They find that Zambian farmers spend large shares of incremental income on non-tradable foods. Moreover, because farm production and investment linkages are still very weak, most of the income multiplier arises within the agricultural sector itself. Only 0.20 Zambian Kwacha (ZMK) of income (or 13% 9 Under the PRSP commercial farms, large-scale corporate agricultural estates or agro-processing industries in rural areas were encouraged and synergies were established for small-scale and medium producers to feed off the large-scale operators via out-grower schemes (GRZ, 2004b). 10 The 1975 Land (Conversion of Title) Act declared that all land has no commercial value and cannot be sold in itself unless there are developments/improvements on it. This was reversed in 1995. However, in most rural Zambia, close to 100% of land is still under traditional authority. 11 Since independence, Zambia has been implementing a number of poverty reducing programmes. These efforts include: the Intensive Development Zones (IDZ), which were later reformed into the Integrated Rural Development Programme (IRDP), the Rural Reconstruction Programme (RRP), the cooperative movement and Lima programmes; Programme Against Malnutrition (PAM); House Ownership Programme; and Micro-credit for the Poor (ILO, 2003a; cf. Kingombe, 2004a). 12 In the mid-1990s RNFE accounted for on average 10-20% of all full-time employment in rural SSA, and 25-30% of rural income (Hazell & Haggblade, 1993). 8 of the multiplier) is generated in the local non-farm economy. Their results imply thatagricultural growth will lead to only modest levels of diversification from agriculture in the regional economy. However, the farm/non-farm linkages might be strengthened e.g. by investments in rural infrastructure and transport systems that better link the villages and towns (Hazell and Hojjati, 1999). Figure 1.1 below illustrates the view that the provision of transport infrastructure should not be supply-driven; rather, it should be developed in response to demand. It shows that transport development should be seen as a necessary but not a driving factor in the development process. Transport is an important contributor in the development process, and investment in transport infrastructure is to be a response to demand from other economic and social sectors. An example is the requirement for improved infrastructure as a result of higher agricultural production. 9 Figure 1.1: The Transmission Mechanisms between Infrastructure, Sustainable Pro-poor Economic Growth, and Agricultural Trade Poverty Reduction in Rural Areas Sustained pro-poor agricultural growth and rural employment creation in agricultural sector and nonagricultural sector in rural areas combined with export of cash crops and agro-processed products Macroeconomic issues: Investment promotion in rural areas and agriculture; export promotion and level playing field for local industry; Governance Issues: Decentralization of management of rural infrastructure development Agricultural: Finance and investment; Agricultural business; Land and infrastructure development; Technology development; Food security. Industry: Export Promotion; Rural industrialization Infrastructure Development: Roads: Rehabilitation, Maintenance and construction of rural infrastructure Soft infrastructure: Credit markets for financing new investment in the agricultural sector Source: Authors‘ adaptation from Zambian θRSθ. 10 Social Sectors: Education and Training; Health and nutrition; Malnutrition prevention. 1.3. Analytical Research Approach and Scope of Research Transport infrastructure investment in developing countries has received renewed attention in the past decade with greater recognition of the links between high transport costs and poverty (Lebo and Schelling, 2005, World Bank, 2005a, UNCTAD, 1999b, SADC, 2005). Yet, despite large amounts spent on rural roads, remarkably little formal evidence exists on their benefits at the household level. What has been lacking is a general methodology for estimating these gains using micro-data (Jacoby and Minten, 2009). It is important to recognize the limitations of project analysis, especially given the time and effort that is required to undertake an economic analysis of a project. Moreover, the central purpose of project analysis of investments from the point of view of national economic objectives as expressed in the PRSP/FNDP, seems to be as important in the beginning of the 21st century as in the 1960s, if not more so (Curry and Weiss, 1993). It is, however, evident that – other than the highly criticised Cost-Benefit Analysis (CBA) – there is no standard methodology employed in these studies (DTZ, 2004).13 Hence, in line with Chung(2000), this thesis takes the position that all research paradigms are valid although they are distinct and philosophically irreconcilable (i.e. a paradigm of “pragmatism”). This philosophy suggests that quantitative and qualitative methods can logically be combined and that epistemological positions need not predetermine the choice of research method (cf. concept of triangulation).14 Only data can help resolve the research questions. Rigorous empirical evidence is needed to support the underlying economic analysis of the view that public investments in the rehabilitation and maintenance of rural feeder roads are beneficial to the broad-based growth of the rural (farm and non-farm) economy and rural 13 There is, as yet, no clear template to follow in terms of assessing wider economic impacts (DTZ, 2004). 14 Most researchers agree that the design of closed-ended questionnaires would be greatly improved if focused qualitative work could be carried out before the questionnaires are designed. Such qualitative work increases the researcher‘s understanding of how local people think about the research areas that will be covered by the survey questionnaires. 11 livelihoods. The approach to resolving this issue “outside” the CBA framework needs to be within realistic informational constraints (e.g. data availability and comparability across the potential roads‘ zones of influence (van de Walle, β00βν Banister and Berechman, 2000; DTZ, 2004).15 Banister & Berechman contend that economic growth from infrastructure improvement is predicated on the presence of allocative externalities. The key point is that if allocative externalities are not present in the local economy, then all of the benefits from an investment project are confined to travel or accessibility related benefits.16 These benefits are fully captured by the measured change in consumer surplus as welfare gains. In that case, growth effects cannot be expected, and attempts to regard some benefits as economic growth amounts to double counting (figure A1 in Annex 1). Moreover, most of the studies reviewed in chapter 2 below highlight the importance of roads in promoting economic growth and development. However, few of them provide information on the distributional and poverty impacts of road investments and in most of their specifications the studies also fail to take into account road quality. To contribute to filling this gap on how rural road investments affect rural growth, poverty reduction, equity and trade, we look for evidence by carrying out a series of 5 micro-level studies, which rely on the analysis of respectively post-harvest survey data, rural household survey data, community survey data and transport survey data. Thus, the aim of this thesis is to identify the medium and long-term effects of the Public Work created rural transport infrastructure on livelihoods and local economies by attempting to disentangle them from confounding influences as well as 15 Macroeconometric studies seem to be one of the least efficient approaches for determining infrastructure gaps. One might make more headway by looking at more disaggregated time series e.g. rate of return where there is some evidence that some types of infrastructure could have been in short supply (Gramlich, 1994). 16 Transportation services are input factors into a host of production, consumption and locational activities. Therefore, a reduction in transportation costs, from an infrastructure improvement, is likely to affect firm‘s and individuals‘ behaviour in other markets, some of which may be outside the price system, which are referred to as allocative externalities (Banister & Berechman, 2000:167). 12 initial conditions at respectively the district; community; and household level in Zambia‘s Eastern θrovince.17 According to the development argument rural transport and communication facilities (i.e. supply factors), particularly rural feeder roads, plays a critical role in the growth and development of the Zambian economy.18 As an essential input, it raises productivity and lowers transport and communication (i.e. production) costs (Howe and Richards, 1984), which literally translates into the competitiveness of all final goods produced in the country,19 and therefore has a significant bearing on the welfare of the people and poverty reduction (GRZ, 2002b). The objective of the thesis is to understand how and why the distribution of rural household production and consumption in Eastern Province in general (chapters 4-5) and respectively consumption and trade in Chipata and Lundazi districts in particular (chapters 7-8) has changed since the sequential implementation of the EPFRP from 1996 to 2001. The main research question is: What happens to the local development at the moment inaccessibility no longer is perceived to be a major problem to economic activities due to the rehabilitation and maintenance of feeder roads in the zones of influence? This applied microeconomic analysis aims to enhance our understanding as to whether providing the best incentives to the farmers is done via higher farm gate prices or via reduced costs (i.e. RTI investment),20 or by employing both strategies 17 In the 1990s developing country governments owned, operated, and financed nearly all infrastructure, primarily because its production characteristics and the public interest involved were thought to require monopoly – and hence government – provision (World Bank,(1994). 18 Infrastructure is an umbrella term for many activities referred to as ―social overhead capital‖. Neither term is precisely defined, but both encompass activities that share technical features (e.g. economies of scale) and economic features (e.g. spillovers from users to nonusers) (World Bank, 1994). 19 The key recognition is that transport cost is a composite of physical distance, travel time and transport fare. 20 An important theoretical question raised by Banister & Berechman(2000) is whether transport development constitutes a necessary and sufficient condition for local economic development. If transport can be regarded as a constraint on the attainment of economic opportunities in an area, then it can be regarded as a necessary condition. Howe & Richards(1984) suggest that the number of elements in the growth process that are designated as necessary but not sufficient is substantial: Capital in 13 together (i.e. the idea of complementarity between price and non-price strategies (Platteau, 1996; Deininger and Olinto, 2000).21 Trying to find answers to this puzzle should per se be considered as our contribution to the current state of knowledge. Thus, the thesis endeavours to explain the following unresolved specific hypotheses whose merits are to be evaluated in chapters four to eight.    Chapter four addresses a hypothesis test proposed in statistical terms: The mean response in agricultural production and productivity growth to labour-based investment in rural roads within the treatment areas is the same as the mean response in the control areas. Chapter five and seven likewise tests a null-hypothesis that states: The mean per adult equivalent real consumption expenditure in treatment districts and control districts are equal. Chapter eight investigates the following empirical generalization: The poor state of the rural feeder road network has constrained the private traders‟ response to and benefit from the liberalized agricultural marketing system in Chipata and Lundazi districts. Figure 1.2 below presents a schematic paradigm showing the causal relationships between infrastructure investment and economic development under certain not necessarily realistic model restrictions in a rural Least Developed Country (LDC) context. Notwithstanding this caveat, economic development from infrastructure investment can be measured either through the real effects22 or through the capitalized effects.23 It is also seen that the investment generates the so-called general and transport investment in particular; appropriate psychological attitudes towards economic activity and change; entrepreneurial abilities and education; the legal, social and political environment; the kind and amount of natural resources…the phrase „necessary but not sufficient‟…might well be expunged from the literature. (cf. World Bank, 1994; Banister&Berechman, 2000). 21 To show whether and to what extent, price reform and infrastructural investments should be viewed as complementary instruments or strategies: An increase in public goods designed to remove constraints on agricultural growth raises the impact of prices on output and vice versa. An increase in roads and other public goods is expected to increase the marginal productivity of privately provided inputs (labour, fertilizers, capital, land, etc.), and the converse is also true. 22 These include changes in factor productivity, changes in the location of households and firms, changes in production and in consumption decisions and changes in agglomeration economies. 23 The combined results (i.e. the accessibility effects) are changes in relative accessibility in terms of mode, network links, spatial location and time of day. Accessibility effects, in turn, stimulate the socalled real effects. Accessibility effects are further capitalized as land rent and consumer surplus. 14 multiplier effect, which results from the infusion of a large sum of capital into the local economy (Banister and Berechman, 2000).24 A principal element in the methodological framework is the level (i.e. geographical scale) at which the analysis should be undertaken (Banister and Berechman, 2000; New York State Department of Transportation, 2000). In line with Banister and Berechman(2000) our research approach is primarily at the microeconomic level, which makes it possible to allow for differences in the form of (unobservable) individual ―district effect,‖ which are not captured by studies conducted in the framework of single cross-country regressions (Islam, 1995). 24 Banister & Berechman do not regard this effect as part of the economic development from the transport investment. 15 Figure 1.2: The Basic causality paradigm of the relationship between transport infrastructure investment and economic development Trigger mechanism Performance and accessibility Networks Infrastructure investment Travel Markets Mode Connectivity O-D pattern Time Capacity Multiplier effect Urban form and structure Analysed in chapter 8 Discussed in chapters 4, 5, & 7 Changes in network accessibility These effects will be examined in chapters 4, 5, 6, 7 and 8, Economic development To be addressed in chapters 4, 5, 6, 7 & 8 Equilibrium Optimality Dynamics Output per capita Employment Private Investment Land rent (Producer surplus) Consumer Surplus Source: Banister & Berechman (2000:38). Location and real effect Productivity Location Agglomeration Production Consumption Employment Environment Social Equity Chapters 4, 5, 6 & 7 16 1.4. A Brief Summary of the Key Findings The main empirical findings, which constitute the contributions to the literature on the socio-economic impacts of rural feeder roads, are found in the core of the thesis comprising chapters 4 through 8. In chapter 4 we find some evidence indicating that cotton yields (MT/HA) followed different patterns in the treatment districts compared with the control districts of Zambia‘s Eastern θrovince. We also find that the EPFRP treatment had an effect that was significant, however our results are sensitive to the choice of model specification. The cotton yield improvements were not sustained beyond the short-term. In chapter 5 we find that there seem to be a linkage between household consumption growth and feeder roads improvements in rural areas in Eastern Province. However, both the results from the cross-section and pseudo-panel data analysis are sensitive to the model specifications. Cotton‘s share in income among the rural households rose by 183% for the total sample between 1998 and 2004 and paradoxically it rose respectively 157% and 294% for the treatment and control districts. On the other hand, the mean distance to services and community assets diminished significantly in the same period. In chapter 6 we e.g. find that the squared poverty gap for most districts in Eastern Province wasn't reduced, which suggests that the changes experienced haven't been relatively pro-poor. While the inequality was higher in the pre-treatment region in 1998 compared to the control region this is no longer the case in 2004 after the EPFRP treatment. Chapter 7 consists of both a quantitative analysis and a more qualitative analysis. In the former we find considerable heterogeneity in impacts across the 16 surveyed communities (i.e. primary sampling units (PSUs)). These impacts are highly context specific and to a large extent the hostage of the small sample size per community as well as overall. We again find another paradoxical outperformance of the treatment areas by 17 the control areas, which most likely is very context specific. Through the more qualitative analysis we find that only 42% of the PSUs had seen their quality of life go up because of the impacts associated with the EPFRP. However, in 12 (63%) of the reporting communities the life quality situation in the PSU was considered better than before the start of the EPFRP. Amongst these 12 communities, 5(45%) considered that the major determinant was directly associated with feeder road rehabilitation. Our regression results lend support to the latter qualitative finding, which leads us to conclude that: the EPFRP treatment does have an impact on poverty; the distance to the market negatively affects poverty; and access to education has a significant poverty effect. Rural households‘ asset endowments (i.e. bicycle and education) are significant determinants of poverty reduction. In chapter 8 we refer to the baseline study by Chiwele et al.,(1998), which find that the nature of the infrastructure and the road network connecting the surplus remote areas and the deficit regions in 1996 was one of the key reasons why the new grain marketing system at the end of the trade liberalisation period from 1991 to 1996 still remained undeveloped. From our own 2005 follow-up cross-sectional survey we find that in gross it is primarily small-scale private companies engaged in ‗agricultural marketing‘ which had made a positive relocation decision triggered by the EPFRP. 1.5. An Overview of the Structure of the Thesis Following figure 1.2 above the thesis is organized as follows. Various approaches and findings of road evaluation studies at both the macro and micro-level is discussed in chapter 2. In particular the macro-level review provides an inspiration for both the choice of covariates and the specification of the growth models at the micro-level conducted in the core chapters of the thesis. Chapter 3 consolidates key information about the case study region - Zambia‘s Eastern θrovince - and the Eastern Province feeder roads project (EPFRP). Then each of the successive five core chapters presents a different framework used to estimate the impact of rural road development, because reliance on any one technique in all circumstances is unlikely to be appropriate. Chapter 4 seeks to estimate the crop 18 production and productivity benefits between treatment and non-treatment districts by means of a secondary data analysis of a series of repeated independent cross-sectional post harvest surveys. Chapter 5 estimates the effect of rural road development on long-term household consumption outcomes between treatment and non-treatment districts by comparing a 1998 baseline Living Condition Monitoring Survey (LCMS) with a 2004 follow-up LCMS with no panel element. Chapter 6 describes the source and characteristics of the LCMS information. Then it shows how the effect of rural road rehabilitation on household consumption can be translated into poverty and inequality impacts for different types and percentiles of households. Chapter 7 qualitatively and quantitatively estimates the welfare outcomes between sub-district treatment and non-treatment standard enumeration areas by comparing a 1996 baseline LCMS with our own 2005 pseudo-panel follow-up survey. Chapter 8 carries out a before-after comparison of trade and marketing outcomes based on the findings from a 1996 existing Zambian study by Chiwele et al., and our own 2005 follow-up transport and agribusiness survey of the likely road users. Finally, chapter 9 summarizes the main results and discusses some of the derived policy implications of the statistical evidence. In the end we suggest some areas that future research need to address in order to have a more accurate idea of the impacts that rural road rehabilitation has on rural households‘ key monetary welfare indicatorsμ Income and consumption. In order to achieve this our comprehensive literature survey clearly indicates that more micro-based studies of the impact of rural transport infrastructure investments are needed to fill the gap. 19 Chapter 2: Theories of Economic Growth and Empirics of Growth: An Exposition and Assessment of the Macro-level Theoretical and Empirical foundations of the Dynamic Models of Production and Consumption Growth at the Micro-level 20 2.1. Introduction After three decades of zero or negative growth, Africa began a growth spurt around 1995 that has been sustained at least to the years before the global recession in 2009. It is believed that most of the recent African growth is due to rising oil and natural resource prices, which entails a redistribution of income from mineral-poor countries to mineral-rich countries (Collier, 2006; OECD et al., 2010). Contrary to this commonly held idea that African growth is largely based on natural resources and helps only the rich and well-connected, Pinkovskiy and Sala-i-Martin(2010) show that Africa‟s income distribution has become less rather than more unequal than it was in 1995, and therefore, that a great deal of the benefits of this growth has accrued to the poor. Notwithstanding this pro-poor growth controversy, bringing about sustained growth in Africa remains one of the biggest challenges (Collier and Gunning, 1999). Moreover, the broad growth literature has had difficulty in coming to grips with the particular character of the African continent. In virtually all cross-country growth regressions, the „Africa dummy‟, i.e. some unexplained factor that causes African economies to show significantly lower growth than the rest, shows up uncomfortably large. Attempts to reduce its size (Freeman and Lindauer, 1999; Gallup and Sachs, 2000; Gallup et al., 1998; Sachs and Warner, 1995) by changing the way in which specific variables are constructed or by introducing variables relating to institutional, physio-geographic and ethnic endowments have, in this context, had only limited success. As a result, even though there are large differences in recent growth performance between African countries (tables A1-A6), a large part of „Africa‟s growth paradox‟ persists. Deininger and Okidi(2003) propose that inability to explain this differential could reduce the applicability and acceptance of policy conclusions derived from such studies. Zambia in general and Eastern Province in particular shares many of the structural factors generally quoted as responsible for low growth in an African context. For example, it is ethnically diverse,25 subject to tropical diseases such as malaria (Gallup and Sachs, 2001, Weil, 2010), as well as being a landlocked country (Limao and Venables, 1999, Collier, 2007, Faye et al., 2004) in Southern Africa with a tropical climate. Moreover, Zambia has had to 25 The population comprises approximately 72 ethnic groups, most of which are Bantu-speaking. Almost 90% of Zambians belong to the nine main ethnolinguistic groups: the Nyanja-Chewa, Bemba, Tonga, Tumbuka, Lunda, Luvale, Kaonde, Nkoya and Lozi. In the rural areas, each ethnic group is concentrated in a particular geographic region of the country. 21 cope with the HIV/AIDS pandemic since the mid-1980s. After having experienced almost consistent decline in the real GDP level since 1980, 1996 marked the turning point, which with the exception of 1998 achieved a decade of economic growth as measured in GDP in constant prices. The constant GDP growth rate even went beyond 5 per cent from 2003 to 2009. The picture is less rosy seen through the lens of real GDP per capita, which in the positive growth period from 1999 to 2006 only recorded an average per capita growth rate of 2.70 per cent, albeit this measure showed steady incremental growth. Thus, Zambia still hasn‘t reached the minimum annual real GDP growth of 7 per cent needed to achieve the Millennium Development Goal (MDG) 1 of ‗eradicating extreme poverty and hunger‘ through achievingμ    Target 1a: “Reduce by half the proportion of people living on less than a dollar a day;”26 Target 1b: “Achieve full and productive employment and decent work for all”, and Target 1c: “Reduce by half the proportion of people who suffer from hunger.”27 Although the mainstream tendency has been for trade theory perspectives to dominate development thinking in recent years a number of researchers and policy analysts, including Rodrik(1995), have continued to start from the development end, rather than the trade end, of the relationship between trade and development. The work of UNCTAD has also been informed by an approach which starts by examining the sources of growth and development, and then considers how international trade fits into this process (UNCTAD, 2004). This idea is captured in figure 2.1, which shows the causal relationship between trade, the development of productive capacities, employment creation and poverty reduction. Since our thesis likewise seeks to build on that body of work concerned with international trade from a development perspective, extending it to the areas of trade and subregional integration, infrastructure development and poverty reduction, the organisation of our literature review will be done accordingly. In addition, the creation of indicators for our rural development analysis will require necessarily a theoretical and methodological framework. The theoretical clues for the creation of rural development indicators may also be found in: Spatial economics; classical theory of 26 1.1 Proportion of population below $1 (PPP) per day; 1.2 Poverty gap ratio; 1.3 Share of poorest quintile in national consumption. Source: UNDP. http://www.undp.org/mdg/goal1.shtml 27 1.9 Proportion of population below minimum level of dietary energy consumption (ibid). 22 location; and regional growth (Akder, 2002). These three themes are captured by the ‗new economics of geography‘ approach, which has developed a common ―grammar‖ to analyze industrial organization, international trade, and economic growth (Fujita et al., 1999). Fujita et al.,(1999) show how a common approach that emphasizes the three-way interaction among increasing returns, transportation costs, and the movement of productive factors can be applied to a wide range of issues in urban, regional, and international economics. We will draw on each of these themes throughout the core analysis in part II. Figure 2.1: The Relationship between Trade, the Development of Productive Capacities, Employment and Poverty Source: UNCTAD, 2004:77. Since the literature on the impact of road investments on economic growth and poverty reduction is extensive, we divide this literature into four broad groups. In the following two sections, we proceed by first extracting the insights gained in the growth theories and the empirical cross-country literature. Then in section four and five we focus on the micro-foundations of the rural growth process by looking respectively at the supply side and the demand side of the rural growth equation. This is done by first taking a closer look at the role played by public capital in the dynamic rural growth process. Then in section five we look at the role played by agricultural trade in terms of driving pro-poor rural growth. Finally, in section six we conclude and present a few policy implications based upon the review of the literature. 23 2.2. Theoretical Foundations: Mainstream Theories of Economic Growth UNCTAD(2004) argues that the essence of a development approach to trade and poverty is that it begins with an analysis of how development occurs, rather than an analysis of how trade occurs, examining the role of trade within processes of development and assessing the effects of trade on poverty from this perspective. By focusing on trade and poverty UNCTAD(2004) opens up the question of the quality of trade in terms of social outcomes of expanded international trade. The basic analytical framework proposed by UNCTAD(2004), which is illustrated in figure 2.1 above, has three components: (i) International trade; (ii) the development and utilization of productive capacities; and (iii) poverty.28 The development of productive capacities involves three basic processes: First, accumulation of physical, human and organisational capital; second, structural transformation; and third, technological progress. Investment in the acquisition of ever-increasing stocks of various forms of capital is the first and most basic component of increasing productive capacity (UNCTAD, 2004). In our case the process of capital accumulation entails investment in rural transport infrastructure. Kaldor(1963) listed a number of stylized facts that he thought typified the process of economic growth (i.e. Kaldor‟s growth laws): 1. 2. 3. 4. 5. 6. Per capita output grows over time, and its growth rate does not tend to diminish; Physical capital per worker grows over time; The rate of return to capital is nearly constant; The ratio of physical capital to output is nearly constant. The shares of labour and physical capital in national income are nearly constant. The growth rate of output per worker differs substantially across countries (Gillis et al., 1992, Hunt, 1989, Barro and Sala-i-Martin, 1995, Degnbol Martinussen, 1994, Basu, 1997, Reynolds, 1985, Hausmann et al., 2005). In parallel with the earliest period of construction of transport infrastructure a debate over the impact on economic development has reigned. What has been established so far is that during the early stage of development (cf. phase one in Dorward et al., 2004b), the socalled development argument is valid,29 because the existence of only a few links in the 28 The latter is defined in a multidimensional way to include low income and consumption, lack of human development, and vulnerabilities such as food insecurity. 29 The development argument, which has been central to some of the exponents of regional development theory (Hirschman, 1958; Hansen, 1965), suggests that if a region has all the economic factors present for growth, then its full potential will not be realized without further transport investment (Banister & Berechman, 2000; Gillis et al., 1992; Hunt, 1989). 24 network enables a clearer identification of the impacts and an inference of the causal relationships (Banister and Berechman, 2000). Both the neo-classical and the new endogenous growth theories regard investment as the fundamental source of improved productivity and economic growth, but the two views diverge on the exact transmission mechanism.30 Most importantly, the neo-classical framework focuses on internal returns to investors who appropriate the benefits of new investment, while new growth models emphasise external effects such as productivity gains spill-over to others. A seminal new growth theory paper by Lucas(1988) considers the prospects for constructing a neoclassical theory of growth and international trade that is consistent with some of the main features of economic growth by which he means accounting for the observed pattern, across countries and across time, in levels and rates of growth of per capita income.31 Lucas(1988) considers an application of the standard neo-classical model to the study of twentieth century U.S. growth. He asks whether this Solow(1956)-Denison(1961) model as it stands is an adequate model of economic development. Lucas concludes that as a useful theory of economic development it fails badly for two central reasons: Its apparent inability to account for observed diversity across countries and its strong and evidently counterfactual prediction that international trade should induce rapid movement towards equality in capital-labour ratios and factor prices. In turn, Lucas(1988) considers two adaptations of this standard model to include the effects of human capital accumulation. The first model retains the one-sector character of the original model and focuses on the interaction of physical and human capital accumulation. The second model examines a two-good system that admits specialized human capital of different kinds and offers interesting possibilities for the interaction of trade and development. Lucas(1988) first constructs an alternative model32 by adding what Schultz(1963) and Becker(1964) call „human capital‟ (i.e. general skill level) to the Solow model in order to consider a complementary engine of growth to the ‗technology change‘ that served the 30 In the Solow-Swan model or the Ramsey model the effects from changes in governmental activities would amount to shifts in the production function. Thus, these types of changes would affect the steady-state level of per capita output and would also affect the per capita growth rates during the transition to the steady state (Barro and Sala-i-Martin, 1995). 31 Lucas(1988) uses the term 'theory' in a very narrow sense, to refer to an explicit dynamic system, something that can be put on a computer and run. This is what he means by the' mechanics' of economic development. 32 It is identifical to the Solow Model in most other aspects: The system is closed, population grows at a fixed (exogenous) rate; and the typical household has the same preferences. 25 purpose in the Solow model. This approach is technically very close to the similarly motivated models of Arrow(1962), Uzawa(1965) and Romer(1986). The theory of human capital focuses on the fact that the way an individual allocates his time over various activities in the current period affects his own productivity (i.e. internal effect of human capital) in future periods. Lucas(1988) argues that by introducing human capital into the model, then, involves spelling out both the way human capital levels affect current production and the way the current time allocation affects the accumulation of human capital. The first model is a system with a given rate of population growth but which is acted on by no other exogenous forces. There are two kinds of capital, or state variables, in the system: physical capital that is accumulated and utilized in production under a familiar neoclassical technology, and human capital that enhances the productivity or both labour and physical capital, and that is accumulated according to a 'law' having the crucial property that a constant level of effort produces a constant growth rate of the stock, independent of the level already attained. The first Lucas Model emphasizing human capital accumulation through schooling is consistent with the permanent maintenance of per-capita income differentials of any size, and hence Lucas(1988) considers that it has contributed to the objective of obtaining a theoretical account of cross-country differences in income levels and growth rates. Though this model seems capable of accounting for average rates of growth, it contains no forces to account for diversity over countries or over time within a country (except for arbitrary shifts in tastes or technology). Lucas(1988) develops a two-commodity elaboration of the model in which human capital accumulation is taken to be specific to the production of particular goods. This second Lucas model thus admits the possibility of wide and sustained differences in growth rates across countries, differences that one would not expect to be systematically linked to each country's initial capital levels. The crucial dichotomy between the neo-classical and the new endogenous growth theories as illustrated by the Lucas(1988) lecture style paper leads to differences regarding the role of investment as a source of growth, policy prescriptions, and implications for long-run gains in productivity, and living standards (Stiroh, 2000).33 33 However, these two frameworks can be viewed as complements rather than substitutes, with neoclassical input accumulation explaining the majority of growth and the new growth theory providing a conceptual foundation for the remainder of productivity growth that falls outside the neoclassical framework. 26 Contributors such as Aschauer(1989) have used the neo-classical model (Ramsey, Harrod, Domar, Solow, Swan etc.) as a starting point (Stiroh, 2000, Banister and Berechman, 2000). However, most of the recent theoretical work on the role of public investment has been done within an endogenous growth framework, as the emphasis is on long-run effects. This literature has according to Barro and Sala-i-Martin(1995) provided some insights into why countries grow at different rates over long periods of time.34 Following the influential work of Barro(1990) a number of researchers for instance, (Barro and Sala-i-Martin, 1992, Baxter and King, 1993, Futagami et al., 1993, Turnovsky and Fisher, 1995) have developed models in which governmental activities, in the form of provision of infrastructural services, affect the long-run growth rate of the economy through the production function, as a factor along with private capital. The main theoretical prediction of this literature is that increases in government spending on infrastructure are associated with higher long-run growth rates; however, this rise in the growth rate is reversed after a point (i.e. the hump-shaped Barro curve), showing that there is an optimum value for public investment (Stiroh, 2000, Mourmouras and Lee, 1999, Barro and Sala-i-Martin, 1995). In this chapter, we will not dwell on the severe critique of the early economic growth and development theories, which I will confine to a remark by von Braun(1995), who concisely summarizes a widespread perception, namely that some of these theories due to their highly aggregate conceptualization and slight regard for the institutional complexities of rural and urban labour markets in low-income countries had only limited usefulness as instruments for guiding policy.35 Sala-i-Martin(1997) suggests that the problem faced by empirical growth economists is that growth theories are not explicit enough about what variables xj belong in the “true” regression. That is, even if it is known that the ―true‖ model looks like the linear regression model (3.1) below, one does not know exactly what particular variables xj should be used. Since the ―true‖ variables that should be included are not known, one is according to Sala-i34 Models with Public Services and taxes: Barro(1990) constructs a growth model that includes public services as a productive input for private producers. Three versions of this type of model is considered: Publicly-provided private goods, which are rival and excludable; Samuelson(1954) style publicly-provided public goods (G), which are non-rival and non-excludable; and publicly-provided goods that are subject to congestion. In the third category, public goods are rival but to some extent non-excludable (Barro and Sala-i-Martin, 1990). 35 In their analysis of effects of market liberalisation on cash crop production, see Dorward et al.(1998) and Stein(1994) for a critique of neo classical economics, which they characterised as being a-institutional. 27 Martin(1997) left with the question: What are the variables that are really correlated with growth? Moreover, Easterly and Levine(2002) point out that while “Total Factor Productivity (TFP)” refers to the “something else” (besides physical factor accumulation) that accounts for economic growth differences across countries both in the level of GDP per capita and the growth rate of GDP per capita, different theories provide very different conceptions of TFP. Because the different theories provide very different views of TFP, they don‘t provide very clear guidance to policymakers and to growth theorists. Additionally, Dercon and Hoddinott(2005) make two important points. First, some factors cause levels of household consumption to diverge across time or space. For example, exploiting insights from endogenous growth theory, it is possible to allow for growth rates to be increasing functions in some endowments of factors of production, while decreasing in other factors. For example, if infrastructure variables have positive growth effects, this would be a sign of external effects in infrastructure. Second, Dercon and Hoddinott(2005) warn that several critical reviews of this framework, such as those by Temple(1999) and Easterly and Levine(2002), highlight the importance of applying this framework with care in either a macro or micro context, given the theoretical and empirical assumptions implied by this model and a range of potential econometric concerns.36 We believe that some theoretical progress of the micro underpinnings of the conventional macroeconomic growth theories could be achieved by incorporating some of the recent findings from the New Institutional Economics (NIE) sub-field of economics. According to Doner and Schneider(2000) NIE helps to identify a series of obstacles, problems, imperfections, and failures, both in states and in markets that can or should be remedied by various institutional means. Or in the words of Dorward et al.,(1998) NIE has through its theoretical framework a distinctive contribution to make to thinking and policy on rural development issues.37 36 37 Of which endogeneity, omitted variable bias, and the presence of a lagged dependent variable are but three. See Dorward et al.(1998) for interesting list of factors that can influence policies and institutional change. 28 2.3. The Empirical Challenge: Cross-Country Patterns of Economic Growth By analysing ninety-eight countries in the period 1960-85 Barro(1991) found that the growth rate of real per capita GDP is positively related to initial human capital and negatively related to the initial (1960) level of real per capita GDP. Moreover, he also found that countries with higher human capital have lower fertility rates and higher ratios of physical investment to GDP, but at the same time growth is only insignificantly related to the share of public investment. The seminal works of Barro(1991) and Mankiw et al.,(1992) led to a vast empirical literature also using cross-country regressions to search for empirical linkages between long-run growth rates and a substantial variety of initial endowments, economic policy, and politico- institutional indicators (Levine and Renelt, 1992). The basic methodology consists of estimating the mean of a random variable conditional on one or more explanatory variables by running cross-sectional regressions of a parametric model: (3.1)     1 1  1 1  ...   n  n   Where  is the vector of rates of economic growth, and X1, …, Xn are vectors of explanatory variables, which according to Sala-I-Martin(1997) vary across researchers and across papers, sometimes due to mis-specification of the models.38 2.3.1. Identified Regressors by the Empirical Growth Literature Starting with the co-variates suggested by standard neoclassical growth models, the range of factors considered has expanded rapidly to range from purely economic ones like the initial level of income, the investment rate, R&D and/or technology (Keller, 2002, Bar-Shira et al., 2003), factor accumulation and/or TFP/productivity (Young, 1994, Bar-Shira et al., 2003, Kögel and Prskawetz, 2001, Young, 1998, Hsieh, 2002), the distribution of opportunities and assets, and human capital (De la Fuente and Domenech, 2001, van Leeuwen, 2004, Mankiw et al., 1992, Hanushek and Kimko, 2000, Cohen and Soto, 2001, Ciccone and Papaioannou, 2005). It is however difficult to find a correlation between education and economic growth at the macro level. Several methods have been used to examine the relationship between education and economic growth (Stevens and Weale, 2003) 38 The resulting estimates can be highly misleading if the assumed parametric model is incorrect. Horowitz and Lee(2002) reviews several semi-parametric methods for estimating conditional mean functions. They conclude that semi-parametric models can achieve their aim of providing flexible representation of conditional mean functions, but care is needed in choosing the semi-parametric specification. 29 and include growth accounting, factor of production models, and endogenous growth models. A common approach is the production function approach whereby output depends on a number of factors such as physical capital and human capital (unskilled and skilled labour). Human capital is measured by the percentage of working age population in secondary school. Mankiw et al. (1992) estimate an equation for output per person for 98 non-oil producing countries in 1985 and find that human capital raises output (te Velde, 2005). Other findings are summarized in te Velde(2005). Barro (1997) finds that one year of additional education raises growth by 1.2% per annum. He also suggests that education is important in catch-up of low-income countries in terms of growth and productivity. Other regressors include policy indicators (Easterly, 2001), socio-economic variables (e.g. electoral competition, checks and balances, the conflict trap, and bad governance) (Collier and Gunning, 1999, Collier, 2009), measures related to institutional infrastructure and the rule of law (e.g. property rights measures; effective government) (North, 1990, Rodrik et al., 2004, Hall and Jones, 1999, Acemoglu et al., 2001, Bockstette et al., 2002, Mitchener and McLean, 2003, Desdoigts, 1999, Tabellini, 2005), and physio-geographic and natural characteristics (e.g. initial mineral wealth and other factors such as distance to the equator and land-lockedness with bad neighbours) (Sachs et al., 2004, Collier, 2007). This rapid expansion of the empirical growth literature happened in tandem with a significant increase in the quality of some of the variables used (Deininger and Okidi, 2003).39 There is a very large literature on links between trade liberalization (openness) and economic growth (Baldwin et al., 2001, O'Rourke and Williamson, 2005, Vamvakidis, 2002, Falkinger and Grossmann, 2005, Ades and Glaeser, 1999, Frankel and Romer, 1999, Alesina et al., 2000, Alcalá and Ciccone, 2003, Levchenko, 2004, Bhagwati and Srinivasan, 2002). However, there are strong methodological objections to some of the key empirical findings, which indicate a positive relationship between openness and growth (Rodriguez and Rodrik, 2001). But recent objections have prompted further responses (Srinivasan and Bhagwati, 1999, Krueger and Berg, 2003) as well as amendments to the case for openness (Dollar and Kraay, 2001, UNCTAD, 2004). See Durlauf and Quah (1998) for a comprehensive overview of the ‗standard‘ growth models and Aron(β000) for a discussion of the more specific institutional variables incorporated. 39 30 The controversy about the effects of openness has seesawed between ―it is good‖ and ―it is bad‖ to reach the more nuanced position that “it is good if the right complementary policies are adopted.” UNCTAD(2004) argues that this common-sense proposition is, unfortunately, tautological and empirically irrefutable. Other identified regressors include: The role of financial intermediaries (Fisman and Love, 2003, Fisman and Love, 2004, Benhabib and Spiegel, 2000, Tressel, 2003, Claessens and Laeven, 2002, Raghuram and Zingales, 1998, Moyo, 2009); ethnic, linguistic and religious fractionalization ones (Fearon, 2003, Alesina et al., 2003, Easterly and Levine, 1997, Collier, 2009); to transaction costs (Nunn, 2005, Martimort and Verdier, 2000), ODA (Gomanee et al., 2005), inequality (Dagdeviren et al., 2001, Barro, 2000, Voitchovsky, 2005, Banerjee and Duflo, 2003, Sala-i-Martin, 2004, van der Hoeven, 2008, Aghion et al., 1999, Cornia, 2004),40 and many other variables that all have been found to be significantly correlated with growth in regressions such the above linear regression model.41 Limao and Venables(1999) use several sources of evidence to explain transport costs in terms of geography and a measure of the infrastructure of the trading countries, and of any countries through which their trade passes. From both data sets, they find that landlocked countries are disadvantaged, although they are able to overcome a substantial proportion of their disadvantage if their infrastructure, and the infrastructure of their transit countries, is of high standard. However, by looking at intra-African trade flows they find that these are somewhat lower than would be predicted by standard gravity modeling, and they show that most of this poor trade performance can be accounted for by poor infrastructure. A range of quantitative studies (Morrison and Schwartz, 1996, Vijverberg et al., 1997, Nadiri and Mamuneas, 1994, Cassou and Lansing, 1999; Stiroh, 2000, Baxter and King, 1993, Easterly and Rebelo, 1993, Mourmouras and Lee, 1999, Hulten and Schwab, 1991, Tatom, 1991, Holtz-Eakin, 1994; Wilson, 2004, Canning, 1998, Pedroni and Canning, 2008) attempting to measure the effect of public infrastructure on output growth followed a 40 There is vast literature on the direction of causality between growth and inequality. Lower levels of inequality may lead to higher growth, through various channels including political stability and greater human capital accumulation: e.g. Perotti, R.1992; Alesina, A. and D.Rodrik, 1994; Persson, T. and G.Tabellini, 1994. 41 Sala-i-Martin (1997) moves away from the extreme-bounds test used by Levine & Renelt(1992) and allows all models to include three fixed variables, so when he combine these three variables along with the tested variable and then with trios of the remaining 59 variables, he always have regressions with seven explanatory variables. Using this approach, Sala-i-Martin(1997) finds that of the 62 (59) variables found in the literature 22 appear to be “significant.” 31 seminal series of papers written by Aschauer in 1989,42 in which he econometrically put infrastructure investment together with the US aggregate productivity slowdown observed after 1973 (Gramlich, 1994, Canning and Pedroni, 1999). Aschauer(1989) includes a flow of productive services from government capital, G, into the neoclassical model as: (3.2) Y = Af(K, L, G), to illustrate that public investment stimulates private investment by increasing the rate of return to private capital and concludes that a core infrastructure of e.g. highways has most explanatory power for productivity (Stiroh, 2000, Barro and Sala-i-Martin, 1995, Banister and Berechman, 2000). However, empirical estimates of the productivity impact of infrastructure investment are either inconclusive ranging from no effect to rates of return in excess of 100% per annum or difficult to pin down (Gramlich, 1994, Stiroh, 2000, World Bank, 1994). Moreover, one should also be aware that infrastructure variables estimates, such as the density of the road network, may be biased insofar as they may capture other country-level fixed effects,43 more careful construction of the stock of infrastructure available finds a smaller though still significant impact (Canning, 1998). Finally through new panel-based unit-root and co-integration tests aimed at isolating the sign and direction of long-run effects in a manner that is robust to the presence of unknown heterogeneous short-run causal relationships Pedroni and Canning(2008) find a great deal of variation in the results across a panel of countries from 1950 to 1992. Taken as a whole, the results demonstrate that e.g. paved roads are provided at close to the growth maximizing level on average, but are under-supplied in some countries and over-supplied in others. Their simple co-integrated panel results help to explain why cross section and time series studies have in the past found contradictory results regarding a causal link between infrastructure provision and long run growth. 42 However, it was the influential study by Barro(1990) that triggered an avalanche of macroeconomic studies of the long-run effects of public investment on macroeconomic performance. Barro and Sala-i-Martin(1992) developed a model in which governmental activities, in the form of provision of infrastructure services, affect the long-run growth rate of the economy through the production function, as a factor along with private capital. 43 For example distance to equator is a factor that does not vary over time and is likely to proxy for other unobserved characteristics, such as transport costs (Deininger and Okidi, 2003). 32 2.3.2. Other Issues in the Study of Growth Brock and Durlauf(2000) questions the empirical practice in the study of growth of many of the contributions referred to above. They argue that much of the modern empirical growth literature is based on assumptions concerning regressors, residuals, and parameters which are implausible both from the perspective of economic theory as well as from the perspective of the historical experiences of the countries under study. This leaves open a number of questions. First, the limited number of country observations available and the open-endedness of the underlying model set limits to the ability to test more rigorously the robustness of the underlying hypotheses and the parameters obtained. This, together with the fact that many of the explanatory variables are correlated, implies that any specific result may be highly dependent on the particular specification adopted. Also, given the need to use data from the national level, aggregation bias, together with measurement error, may pose problems. This would not only result in losing most of the information specific to gender but also lump together all policy changes in a single time dummy. Also, many of the variables chosen are at best imperfect representations of what the model intends to measure. Differences or sudden shifts in standards and underlying definitions can conceal considerable heterogeneity within and across countries. This problem is aggravated by unobservable differences in the policy regime, which, in addition to increasing measurement errors, may be related to countryspecific unobservable attributes (Brock and Durlauf, 2000, Deininger and Okidi, 2003). Second, looking only at aggregate country-level data also makes it more difficult to deal with issues of poverty and inequality in addition to growth, and in an integrated framework. The debate from the 1960s and early 1970s, whether policies that aim to increase growth will at the same time also help the poor, experienced a resurgence of interest following the general revival of growth theory sparked by the work of (Romer, 1986, 1990) and (Barro, 1990, 1991) and has since been widely debated in the literature (Dollar and Kraay, 2001, Alesina and Rodrik, 1994, Perotti, 1992, Persson and Tabellini, 1994). Existing distributional data across countries are too noisy to make specific inferences on the issue (Banerjee and Duflo, 2000, 2003), and it is unlikely that, barring a significant improvement in the databases available, cross-country data and approaches will allow us to resolve the issue (Bourguignon, 2000 referred to in Deininger and Okidi, 2003). 33 In the latest addition to the debate Pinkovskiy and Sala-i-Martin(2010) claim that “the conventional wisdom that Africa is not reducing poverty is wrong.” Using the methodology of Pinkovskiy and Sala‐i‐Martin(2009), they estimate income distributions, poverty rates, and inequality and welfare indices for African countries for the period 1970‐2006. They show that: African poverty is falling rapidly and that the growth spurt that began in 1995 decreased African income inequality instead of increasing it. All classes of countries experienced reductions in poverty. In particular, poverty fell for both landlocked as well as coastal countries; for mineral‐rich as well as mineral‐poor countries; for countries with favorable or with unfavorable agriculture; for countries regardless of colonial origin; and for countries with below‐ or above median slave exports per capita during the African slave trade. Chen and Ravallion(2008) came to exactly the same conclusion in their research using a different methodology. However, they also point out that the decline in the aggregate poverty rate has not been sufficient to reduce the number of poor, given population growth. Pinkovskiy and Sala-i-Martin(2010) obtain lower poverty rates than Chen and Ravallion(2008) and they appear to have a somewhat steeper decline. These differences reflect the different methods. Ravallion(2010) notes two points: (i) Chen and Ravallion(2008) show that the poverty decline in SSA tends to be larger for lower poverty lines (in the region $1-$2.50 a day) and (ii) Pinkovskiy and Sala-i-Martin(2010)‘s method attributes the entire difference between GDP and household consumption to the current consumption of households, and they assume that its distribution is the same as in the surveys. These assumptions are very unlikely to hold, and they give an overly optimistic picture. In effect, Pinkovskiy and Sala-i-Martin(2010) are using a lower poverty line than Chen and Ravallion(2008). Another important difference is that Chen and Ravallion(2008) are more cautious about the data limitations. There are not enough good household surveys available yet to be confident that this is a robust new trend of a falling poverty rate for SSA. Their main graph has 30 annual data points since 1995. But these are not real data points in any obvious sense; rather they are synthetic (model-based) extrapolations based on national accounts and growth forecasts according to Ravallion(2010). In fact from the data catalogues of the International Household Survey Network (IHSN) we have national household surveys for all but 10 of the 34 48 countries in SSA since 1995. However, for only 18 countries, including Zambia, do we have more than one survey since 1995.44 Finally, to the extent that differences in growth are ‗explained‘ with reference to immutable country-specific factors, the relevance of cross-country evidence for actual policy formulation is limited (Deininger and Okidi, 2003). In the same vein, Banister and Berechman(2000) argue that the production function type macroeconomic models (3.2), used in the literature to measure the relationships between the country‘s level of infrastructure supply and GDP growth, are largely inadequate for evaluating the effect of infrastructure development on the economic growth of cities and regions.45 They find that it is difficult to construct causal relationships that support the data as the effects of external factors, time and stage of development will all influence the direction and strength of those relationships. This suggests that a micro level analysis is more appropriate. The same conclusion is e.g. reached by Pedroni and Canning(2008) who argue that for policy purposes their results point to the need for detailed country studies in order to find appropriate rates of return to infrastructure. 44 See ωSη Zambia‘s Survey Data Catalogue at http://www.zamstats.gov.zm/nada/?page=catalog. The International Household Survey Network provides the tools and guidelines used to establish the National Data Archive. 45 The level and quality of services available to users from public infrastructure are affected by myriad factors. The type of the infrastructure facility (e.g. rail or road), its location, particular design, level of maintenance and form of governance. Therefore, the lumping together of all forms of public infrastructure facilities to compute a single aggregate measure of public capital stock while disregarding the particular attributes of specific types of capital is likely to generate biased estimates of the rate of economic growth from this capital according to Banister & Berechman(2000). 35 2.4. Supply Side: Productive Government Expenditures‟ Impact on Economic Growth Poor physical quality roads network increases transport costs that, in turn, constrain investment and economic activity, such as crop production; employment opportunities in the farm and non-farm economy. Approximately 900 million people in rural areas live without access to all-season-roads. This gap constitutes a serious hindrance to economic and social opportunities of those living in these areas. Governments in developing countries devote significant efforts to expand and improve rural roads, often with international funding. Previous studies on public expenditure focused on performance of budget implementation. Deininger and Okidi(2003) conclude that it has proven difficult to compare relative returns to investments such as infrastructure, which are required to make informed allocative decisions for public expenditures. Fan et al.,(2000a) and Fan et al.,(2002) constructed an econometric model to estimate the effects of government spending on poverty reduction through various channels, using secondary data from government statistical agencies in India and China. Building on previous IFPRI studies in Asia and Ugandan data availability, a study by Fan et al.,(2004) develops and adapts a simultaneous equations model to estimate the effects of government expenditure on agricultural production and on rural poverty through different channels. These sub-regional studies find government spending on infrastructure highly significant (Fan and Hazell, 2000, Fan et al., 2000a). This section consists of two sub-sections. The first takes a brief look at some of the theoretical approaches to rural development, whereas the second presents the results obtained in the various empirical studies reviewed in terms of the socio-economic impacts identified. 2.4.1. Theoretical Approach to Rurality The question on basic geographical unit for data observation is closely related to the approach to rurality. The spatial approach has its roots in optimal location theory. The twin phenomenaμ Distance and area begin in ―The Isolated State‖ by J. H. Thünen. The role of distance is exemplified by the fact that transport costs affect not just market prices, but also the location of production facilities. The role of area, on the other hand, implies that the markets for specific goods are subject to definite geographic limits (Akder, 2002, Venables and Limao, 2002). 36 Models built on this understanding have a hierarchical vision of space (concentric rings) and an interaction of agglomeration and dispersion forces. This relationship of agglomeration and dispersion forces will structure: The space functionally and the geographic specialization that emerges will assign different functions to the center and periphery. In this spatial approach, rural coincides with periphery. The economic and social formation on the periphery is considered as traditional and dependent on the urban center. Within this context rural area refers to an area that falls into statistical threshold values of certain indicators. These might be agricultural census statistics on land use that also indicate intensity; size of population, and population density (Karlsson and Berkeley, 2005).46 Akder(2002) suggests that the spatial approach might be relevant for developing countries when there are strong differences between rural and urban, and when rural consists more of homogeneous agricultural areas. Then the choice of indicators for both classification and analysis might be made according to the spatial approach. Like the economic development and growth theories, regional economic growth theories may also help to justify the rural development indicators used in our regional growth analysis (Akder, 2002). 2.4.2. Productive Government Expenditures and Rural Development Much of the debate during the 1960s, the 1970s and 1980s revolved around freight transport. It was consistently argued that a reduction in freight transport costs would lead to the exploitation of scale economies, and this in turn would lead to reductions in commodity prices. The underlying argument was that these benefits exceeded the direct reductions in transport costs (Tinbergen, 1957). Banister & Berechman(2000:15) highlight several weaknesses in the argument, particularly as it relates to depresses areas:  ―A reduction in transport costs to a depressed area may make it easier to supply other areas from the area in question, but at the same time it will make it easier to supply that If the typology of regions were distinguished by ―remote rural‖, ―accessible rural‖ and ―urban‖, this would correspond more to the spatial approach. Almost all studies carried out by OECD have relied more on the territorial approach. The typology chosen by these studies isμ ―predominantly rural‖ where more than η0% of the regions‘ population lives in rural communitiesν ―significantly rural‖ where 1η-50% of the population lives in rural communities and ―predominantly urbanized‖ where less than 1η% of the regions‘ population lives in rural communities with a threshold value of less than 150 per square kilometer (Karlsson and Berkeley, 2005). 46 37 area from elsewhere. It is analogous to the international trade theory argument of the  impacts of a reduction in the tariff.47 Regions that will benefit from investment are those that already have non-transport advantages, such as a buoyant local economy, new industries, a heavy inflow of investment, available sites, a high quality labour market, agro-climatic endowments, and agricultural opportunities (i.e. observed and unobserved characteristics that affect e.g.  public investments in roads, and public spending decisions)(Khandker et al., 2006). The problems of depressed areas can be traced to non-transport factors (e.g. the need for industrial restructuring and increases in labour productivity). Improved communications may do more harm than good.‖ 2.4.2.1. Poverty Reducing Employment Intensive Investment Programmes The 1980s and 1990s saw a tremendous proliferation of poverty reducing employment programmes, especially in Asia and Africa. Reviews of these public works programmes have been undertaken on several occasions (Burki et al., 1976, Carapetis et al., 1984, Howe and Richards, 1984, Clay, 1986, Gaude et al., 1987, Ravallion, 1991, Gaude and Miller, 1992, Hirway and Terhal, 1995, von Braun et al., 1992, von Braun, 1995, Keddeman, 1998, Banister and Berechman, 2000, DTZ Pieda Consulting, 2004; Chipika, 2005; McCord, A and Slater, R. 2009). 47 The provision of improved transport infrastructure may open an area to external competition which would reduce prices for businesses and consumers in that area. This may be regarded as net benefit at the local level but will create losses for some local businesses (DTZ, 2004). 38 Table 2.1: Long-term effect of rural roads identified by literature Authors Invest. remains in Creation of Rural non-farm Area / employment / Availabilit Income increasing job y of generated opportunities / schooling reinvested rural non-agr. / school in Agr. Activities enrolment 1 Edmonds and Howe, 1980 2 3 X X Commit more land, Less land fertiliser, devoted to machinery major to farm staple food production crops 4 5 X X Positive Positive effects on effect on output Negative the (farm- effect on adoption of gate) input improved prices prices technology Labour use per unit of cropped land 6 7 8 9 X X X 12% ↑ LB/FFW Facilitated Increasing Improved Wages gorwth in Inequality Marketing compared Agr. / bt. Opportuni to Impetus Accessible ties and Female for and non- Reduced participati Agricultu Food on in work- ral wages Security economic accessible Transactio growth areas n Costs force (land-less) level 10 11 12 13 ↑ Noticeable ↑ X 14 15 >50% El Hawary and El Reedy, 1984 Obare et al., 2003 Ahmed et al., 1995 X ↑↑ Zohir, 1990 ↑ Mitra and Associates, 1991 Ling and Zhongyi, 1995 X Jones, 1984 X X Dev, 1995 X X Sathe, 1991 X X Binswanger et al., 1989 X Gaviria et al., 1989 Webb and Kumar, 1995; Webb, 1995; Teklu, 1995 X X Lebo and Schelling, 2005 Source: Findings reviewed by author. Stabilized ↑ ↑ ↑ ↑ 39 ↑ ↑ ↑ The work, since the early 1970s, of the International Labour Organization (ILO) and of the World Bank has shown conclusively superior benefits accruing to the rural poor and higher financial and economic (i.e. actual resource) benefits by the adoption of labour-based (LB) rather than equipment-based (EB) methods of road construction and maintenance (Howe and Richards, 1984; ASIST, 2003b; Taylor and Bekabye, 1999;48 Riverson et al., 1991); Kenya Roads Board, 2004;49 Gaviria et al., 1989; Kapsel and Daima, 2004; Taylor et al., 2008)50 (table 2.1). Taylor et al.(2008), the most recent of these comparative ILO studies, confirms the central hypothesis raised by all the previous studies, namely that labour-based construction methods are viable, significantly cheaper in both financial and economic terms, and offer higher employment potential for unskilled workers, as well as greater indirect benefits to the national economy than the conventional, equipment-based technology. This is done by testing the hypothesis on empirical evidence from rural gravel road construction and maintenance projects carried out in Tigray and Oromiya Regions of Ethiopia between 1997 and 2007. The overall implications of their study is that the macro-economic framework would be improved if Ethiopia increased the use of labour-based methods in infrastructure development due to the stronger multiplier effects on the economy in terms of more income to the household involved and stronger stimulus on local private investments. Lipton(1996) and Keddeman(1998) in their literature reviews distinguish the following categories of impact on the economy: Short-term direct effects; Short-term indirect effects; Opportunity costs; Backward linkages; Forward linkages; and Long-term direct effects, which together constitute the total impact for the assessment and evaluation of projects. Keddeman(1998) points out that information about the long-term effects of employment ―δabour-based road works methods have been proven to be technically and economically comparable or superior to equipment-based methods when labour costs are US$4 or less per day than the minimum wage rate for most countries in SSA, implying that this approach is efficient (Banjo, 1999; Taylor&Bekaye, 1999). 49 The Government of Kenya with support from development partners, embarked on an ambitious Rural Access Roads Programme (RARP) in 1974, which by its conclusion in 1985 had constructed over 8,000 km of road by LB methods and created several thousands of jobs. This highly successful programme provided the foundation extensive pioneer work into the development of the LB techniques and management practice. 50 The Bank's transportation department had not been engaged in any significant feeder road development and had concentrated only on the main roads. Between 1960 and 1988 the Bank committed 8 loans amounting to US$470 million of the highway sector. In contrast, the rural road development had largely been relegated to the Bank's agricultural staffs, which had allocated approximately US$380 million for rural road components of Agricultural Development Projects (ADP) in 16 states. 48 40 intensive investment programmes (EIIPs) are both limited and difficult to compare across programmes (table 2.1). Notwithstanding these measured benefits, since early 1980s there has been a great deal of evidence to suggest that the major investment programmes in rural roads have not achieved the hoped for increases in agricultural production and in living standards of rural population (Ali-Nejadfard, 2000; World Bank, 2005c; McCord, A and Slater, R. 2009; McCord, A. and Wilkinson, C. 2009). Banjo(1999) suggests that if the use of intermediate-means of transport (IMT) such as bicycles, animals, and animal-drawn vehicles were facilitated, the productivity and wellbeing of rural households would improve significantly. Ali-Nejadfard(2000) highlights some of the results and findings from several studies carried out in Africa and Asia to show how rural travel and transport could catalyse rural development. Nevertheless, Lipton(1996) argues that long-term effects of EIIPs are diverse and complex and doubts do persist about the benefits to the poor of the assets created through PWPs. This doubt could according to Van de Walle(2002) be explained by the existing methods of project appraisal for rural roads don‘t properly reflect the potential benefits to the poor. In fact, a majority of studies on rural roads lack a proper counterfactual analysis, limiting the availability of evidence on impact of these investments on key development outcomes. A study by Chipika(2005) assess the evidence base and methodologies for poverty reduction impact assessment in the EIIP/ASIST- Africa Programme. He concludes based upon a review of more than 90 baseline and impact studies, which are available in the ASIST Information Service, that they are deficient in failing to classify the potential beneficiaries of EIIPs by their poverty categories. This limitation makes it difficult to assess changes that have occurred from the period the baseline studies were conducted up to the period when the impact assessment studies were conducted. Chipaka(2005) therefore warns that the EIIPs outreach to the poor households will remain weak unless substantial efforts are made to target society‘s poorest segments. In many cases the evidence of such targeting is not available which partly explains the perceived poor links between EIIPs and poverty dimensions. 41 2.4.2.2. Structural Models of Road Impacts Structural form models identify impacts on the basis of economic assumptions about how the world works. They rely heavily on a specific model of behavior, usually with clear economic assumptions and are internally consistent. Van de Walle(2008, 2009) classifies the models as follows: (1) CGE models in which roads are modeled as exogenous transport cost savings; (β) ―εacro-style‖ simultaneous-equation econometric models of the economy. Impacts of past roads spending estimated by region with time series data (Fan et al., 2000, 2004; Fan and Chan-Kang, 2005); (3) “Micro” partial equilibrium work calibrating farm-household models for estimating the impacts of reduced transport costs through road provision (Jacoby, 2000; Jacoby and Minten, 2009). The downside of these structural approaches is that assumptions have to be plausible and they may not be empirically testable. The upside is that they gain on what we can learn as long as the assumptions are valid. In contrast impact evaluation is highly a-theoretical and basically reduced form. As mentioned by van de Walle(2008, 2009) often there is very little that we can understand about why it is that we have impacts in general and specifically what the channels are through which roads are having impacts. Nevertheless, the two types of approaches are considered to be complementary by van de Walle(2008, 2009), because they are asking different questions and are looking at different things. 2.4.2.3. Impact Evaluation of Rural Road Infrastructure Impact evaluations of road infrastructure are complex because of the economy-wide effects that roads create. Roads influence a wide array of economic and social activities. Acting through lowered transport costs, roads might promote market activities, the availability and use of social services; affect the division of labour inside and outside the household; etc. A thorough evaluation of all these effects is necessary in order to assess the contribution of this type of investment on the welfare of the population (World Bank, 2010).51 The central empirical obstacle to estimating road impact is reverse causation (Fan, 2004). Roads are not randomly placed and people do not randomly settle next to roads once 51 Source: World Bank Development Impact Evaluation Initiative Website consulted February 2010. 42 they have been constructed. The causal link between better road access and the benefits of such access may thus be hopefully obscured. Longitudinal (i.e. micro-level panel) data spanning a period of road construction can alleviate the endogenous road placement problem insofar as the unobservables determining such placement are fixed over time (Jacoby and Minten, 2009), while overcoming the limitations of the aggregate cross-country approach. Most recent impact evaluation studies that take into account the endogeneity issues use double-difference (DD) combined with other methods to deal with the initial conditions that affect the trajectory of impacts. For instance DD combined with propensity score matching (Mu and van de Walle, 2007). DD focuses on difference in outcomes over time between project and non-project communities. This approach purges additive time-invariant observables and unobservables (and deals with time invariant selection bias). But, initial conditions may also influence subsequent changes and trajectories (time varying selection bias). Propensity-score matching (PSM) is used to select ideal comparison communities (van de Walle, 2008, 2009; Jalan and Ravallion, 2003; McCord, and Wilkinson, 2009). The results from a long-term evaluation of a World Bank-funded roads rehabilitation and improvement project in Vietnam by Mu and van de Walle(2007) show heterogeneous impacts across regions and socio-economic groups. For example, it finds that markets are more likely to develop as a result of road improvements where communities have access to extended networks of transport infrastructure. Another method used by Gibson & Rozelle(2003) has been to combine DD with instrumental variables in Papua New Guinea. Khandker et al.,(2006) use DD and controls for initial conditions through OLS for Bangladesh. A different approach applied by respectively Jalan & Ravallion(2002) and Dercon et al.,(2006) uses dynamic panel data models to look at either initial assignment or changes in road assignments. Here the approach is to look at the impact on consumption and poverty reduction. The key is to look at at least three waves of a panel to be able to use this approach adequately. However, it is very rare to find this kind of data on access to roads. 43 A panel covering four periods in six provinces in China permits exploration of households‘ ways of smoothing consumption and coping with risk (Jalan and Ravallion, 2002). A 4-year panel from Vietnam not only illustrates the regionally differentiated nature of growth but also identifies factors that helped households escape from, or caused them to fall into, poverty (Glewwe et al,. 2000; Deininger & Okidi, 2003). If such panel data span broad policy changes at the macro-level, such as liberalisation of input or output prices, modification of subsidy schemes related to social services and education, or a large contraction due to balance-of-payment difficulties, the ability to use information about the same household in two-periods will produce more precise and less biased estimates. A 4-year panel from Peru enables exploration of how differences in initial endowments affected households‘ ability to deal with a macroeconomic crisis (Glewwe and Hall, 1998; Deininger & Okidi, 2003). A number of panel data sets allow an analysis of similar questions in Africa. A shorter panel from Ethiopia demonstrates the importance of price variables as well as exogenous shocks (rainfall) for analysing growth at the household level (Dercon, 2001). Micro-level survey and a longer panel data evidence from Uganda spanning 1992-2000 by Deininger and Okidi(2003) confirms the benefits from Uganda‘s decisive liberalisation of output markets. It demonstrates the importance of improving access to basic education and health care emerges more clearly than in cross-country analysis, but benefits depend e.g. on complementary investments in electricity and other infrastructure. An impact evaluation of rural roads investments in Bangladesh by Khandker et al.,(2006) finds that roads improvements led to lower input and transportation costs, higher production, higher wages, and higher output prices as well as to increases in both girls‘ and boys‘ schooling. A study of rural roads in Nepal by Jacoby(1999) finds that access to roads improves the productive capacity of poor households. However, the study also concluded that the impact of roads on poverty reduction was limited and had no effect on inequality. 44 An evaluation of impacts of road rehabilitation in Georgia by Lokshin and Yemtsov(2005) find that opportunities for off-farm and female wage employment increased, but that most of the increase occurred in non-poor households. An impact evaluation by Escobal and Ponce(2003) find that the rehabilitation and maintenance of roads in Peru improved some measures (access and attendance to schools and child health centers) or had no significant impact on others (agricultural production, income, poverty). In order to impose further structure on the microeconometric approach Ravallion and Jalan(1996), Jalan and Ravallion(1997, 1998) and Dercon and Hoddinott(2005) borrow from the conceptual framework used to understand growth at the national or cross-national country level reviewed in section 3 above. Ravallion and Jalan(1996) point out that it is not implausible that community capital has positive external effects in the micro-level growth process. Analogous to the role of firmspecific knowledge and economy-wide knowledge in Romer(1986)‘s model, they conjecture that on fully accounting for all private inputs, there will be constant returns to scale to the privately provided inputs, but increasing returns to scale to overall inputs, including community capital. One limitation to the studies, that specify an aggregate production function which includes transportation infrastructure among the set of explanatory variables, is the failure to take road quality into account. Road quality can vary greatly within a country and different quality roads can act in different ways. Failure to discriminate amongst types of roads can also lead to biased estimates according to Fan(2004). Dercon and Hoddinott(2005) estimate a series of probit regressions through which they find that an increase of 10 km in the distance from the rural village to the closest market town has a dramatic effect on the likelihood that the household purchases inputs, controlling for the effect of other factors. However, they get mixed results in terms of the likelihood of engaging in various activities when roads of poor quality (accessible only to carts, animals, or people) were replaced by good quality roads (reasonable access to any vehicle). 45 Dercon and Hoddinott(2005) also find from their Fixed effect IV regression that increases in road quality have strong positive growth effects: Improvement in roads leading to local towns, from a road poorly accessible to buses and trucks to one reasonably accessible for buses and trucks in the rainy season results in 3.5 percent higher growth. Furthermore, there is a persistent and divergent effect linked to road quality: The better level of past road quality increases growth. This finding is corroborated by a number of other studies e.g. (Fan, 2004, Deichmann et al., 2002). Fan(2004) for example finds that low quality (mostly rural) roads have benefit/cost ratios for national GDP in China that are about four times larger than the benefit/cost ratios for high quality roads. As far as agricultural GDP in China is concerned, high quality roads do not have a statistically significant impact while low quality roads generate 1.57 yuan of agricultural GDP for every yuan invested. Investment in low quality roads also generates high returns in rural nonfarm GDP. In terms of poverty reduction, Fan(2004) also finds that low quality roads raise far more rural and urban poor above the poverty line per yuan invested than do high quality roads. The presence or absence of road infrastructure is perceived to be one of the main determinants of food price variation. An analysis by Minten and Kyle(1999) show that in the case of the Democratic Republic of Congo (DRC), food price dispersion is significant both across products and across regions. They demonstrate that transportation costs explain most of the differences in food prices between producer regions and that road quality is an important factor in the transportation costs. Building on these evaluation studies of rural roads our thesis seeks to contribute to the micro-level supply-side literature by seeking to answer the key research question: Do rural assets created through the EPFRP build sustainable livelihoods and reduce poverty in the treatment areas? Our main contribution is the focus on physical quality of the assets created in chapter 7 and whether the assets are still functional 4 to 7 years after the completion of the various segments of the network. In other words, this thesis fills a gap by looking at the medium-to long-term effects on livelihoods and local economies in terms of crop production and yield in chapter 4; household consumption and income in chapters 5 to 7. 46 2.5. Demand Side: Rural Growth Through Agricultural Trade Early studies by economists opened up a debate as to whether a reduction in transport costs brought new areas and products into the market. Rostow(1960) argued that this was the case and that transport investment also contributed to a major new export sector and was instrumental in the development of the modern industries (Banister and Berechman, 2000). This take-off notion is often advanced as justification for the need to end the isolation of hitherto inaccessible areas. The building of a road, the propagation of means for moving about, makes it possible to export the produce of local production operations to other markets, within either a close or distant radius, such as the district centre or the international market. These outlets allow the initiation of a 'virtuous circle', characterised by intensified production, increased incomes for the population, diversification of consumption and the possibility of accumulation (Peguy, 1998). Until the early 1970s transport costs remained among the most important explaining factors in economic geography and theories of regional and industrial development, but during the 1970s transportation and transport costs disappeared almost completely out of mainstream economic geography. However, during the 1990s transport and communication made their way back into the mainstream again, but now transformed into the much broader concept of logistics, which has become an increasingly important element in the organisation and restructuring of the globalizing economy. From being an external factor of location transport and communication have become an integrated part of the production and distribution system also in Africa (Pedersen, 2000). Thus, the success of Africa's exports, as well as its spatial development, depends on lowering transport costs. Naudé and Matthee(2007) present a case study of the firm location decisions of exporters in South Africa to illustrate the significance in particular of domestic transport costs for manufactured exports. Their is that Africa's international transport costs are significantly higher than that of other regions, and its domestic transport costs could be just as significant. Moreover they show how domestic transport costs influence the location, the quantity, and the diversity of manufactured exports. Their findings confirm those of studies undertaken by the Transport Economics Laboratory (LET, Lyon), the National Institute for Transport Research and Safety (INRETS, Arcueil) and the SITRASS teams, which have shown that, in 1996, transport prices per tonne/km in SSA could be as high as ECU 0.30 for journeys of less than 50 kms, reflecting 47 reduced-tonnage consignments in rural areas, as compared with prices of the order of ECU 0.05 per t/km for longer journeys. An international comparison between the countries of Asia and French-speaking Africa found that long-haul trucking was twice as expensive in Africa as in Asia (Peguy, 1998). The general conclusion from surveys of firms has been that transport is a second order consideration for location or relocation decisions, provided that there is a good quality road network available. Transport is a background variable that has to be present, but comes behind factors that more directly affect the efficiency, productivity and profitability of a firm. It is also considered a factor that is beyond the control of an individual firm (Banister and Berechman, 2000). There are many arguments from Traditional Location Theory and more recent Trade Theory (Krugman, 1991a, Krugman, 1991b; Fujita et al., 1999, Krugman, 1998) about where new development is likely to take place. Geographical concentration relies on the interaction between increasing returns, transport costs and demand. With sufficiently strong returns, each manufacturer will serve a national market from a single location. Krugman(1991b:41) argues that ―it is the interaction of increasing returns and uncertainty which ‗makes sense of εarshall‘s labour pooling argument for localization.‘52 Labour market pooling, together with the supply of intermediate goods and knowledge spillovers, ensures economies of localization at the regional levels‖ (quoted in ψanister & ψerechman, 2000:51). The low density and poor quality of rural roads, which results in prohibitive transport costs, also leads Wiggins(2001) to present arguments from this „New Economic Geography (NEG)‟. He suggests that rural areas, particularly more remote rural areas, suffer from a wide range of economic disadvantages as regards information and access to input and output markets, with the result that urban areas have a strong comparative advantage for many economic activities. He argues that in ―the deep countryside‖ only activities with a strong natural resource base (e.g. agriculture and, in some areas, tourism), local processing of agricultural products and non-tradable services for the rural population will survive. Exceptions to this may arise where rural labour costs are much lower than urban costs, and 52 A substantial share of rural manufacturing involves agro-processing and the production, repair and supply of farm inputs (Hazell and Hojjati, 1999). 48 there are good communications and road networks in rural areas encouraging labourintensive industries to locate in rural areas (Kydd et al., 2004). This NEG argument is supported by Jacoby and Minten(2009)‘s hypothetical rural road project that reduces the transport costs of the most remote households by around 75USD/ton, which in turn raises their incomes by about 50%. However, they emphasize that this gain due to the reduction in the costs of goods transport, both exports and imports, is small compared to that from improved access to non-farm earning opportunities in town. This suggests that there may be potentially important complementarities between rural road construction and urban economic development (Jacoby and Minten, 2009) and that the rural non-farm economy (RNFE) will play an important role in determining future prospects for employment growth and poverty alleviation in Africa (Hazell and Hojjati, 1999). At low per capita income levels, agricultural growth will according to Hazell and Hojjati(1999) only lead to modest levels of diversification from agriculture in a labour short regional economy like Eastern Province, Zambia. However, the farm/non-farm linkages might be strengthened by investments in rural infrastructure and transport systems that better link the villages and towns, and continued policy reform to create a more enabling economic environment for the region's farmers and non-farm entrepreneurs. However, Jalan and Ravallion(1997) find that there are publicly provided goods in the so-called spatial poverty trap setting, such as rural roads, which only generate non-negligible gains in living standards. The prospects for growth in this kind of poor areas will then depend on the ability of governments and community organizations to overcome the tendency for under-investment that such geographic externalities are likely to generate. A few studies have quantified the effects of infrastructure provision on trade and growth; all find a positive correlation. Francois and Manchin(2007) estimate a large panel of bilateral trade flows over the period 1988-2002 for many countries and focus on the effects of communications and transport infrastructure. They estimate an increase of one standard deviation (from the mean) in the communications infrastructure raises the volume of trade by roughly 11 percent, compared to a 7 percent effect on transport infrastructure and a 2 percent effect on trade for tariffs. For the LDCs, a category containing most SSA countries, transport is more important than communications. The effects of communications infrastructure on trade grows as a country reaches the middle income range (Te Velde and Meyn, 2008). 49 Using econometric evidence and spatial network analysis techniques to explore the relationship between road transport quality and overland trade in SSA Buys et al.,(2006) find that upgrading a primary road network connecting the major 83 urban areas in SSA would expand overland trade within SSA by around US$250 billion over 15 years. The Sub-Saharan Africa Transport Policy Program (SSATP)53 evaluated the Northern Corridor (covering Kenya, Uganda, Rwanda, Burundi, North Tanzania, South Sudan and Eastern DRC) performance considering the quality of physical infrastructure and the quality of services (transport time). Delayed deliveries due to slow procedures at ports and borders increase the costs for African companies. As result of delayed deliveries, African countries tend to increase their inventory holding. Firm surveys among nine African countries found that firms hold, on average, the equivalent of three months of input needs (Fafchamps and Hill, 2004). In case intra-regional frameworks build-up effective transportation networks, the cost for stockholding could be decreased, which in turn would contribute positively to improve the business climate (te Velde and Meyn, 2008). Gaviria et al.,(1989) refer to Lele, who elsewhere has argued that there is a need for greater coordination between the development of agricultural potential and that of rural road infrastructure. The absence of such coordination in many instances results from a lack of appreciation of the fundamental importance of rural roads in the early stages of rural development (Dorward et al., 2002, Dorward et al., 2004c).54 Thus, chapter 8 of the thesis builds on the location theory and demand side literature. It endeavours to make a contribution by disentangling the medium to long-term observable impact of the EPFRP rural asset creation on transport, market access, journey time and vehicle operating costs in two districts of Zambia‘s Eastern θrovince. 53 SSATP covers 35 African countries forming 8 economic communities headed by UNECA, AU/NEPAD and AfDB. SSATP is now in the process to implement its second Long-Term Development Plan (LTDP) (2008-11). The overarching theme is the creation of responsive transport strategies including road management and finance, transport services and regional transportation. 54 Most countries of SSA are stuck at various places within stage γ (called ―going to market‖), and few have progressed to stage 4 (―making markets work,‖ and ―making decentralization work‖) (Townsend, 1999μ1β4 quoted in Donavan, 1999:102). 50 2.6. Conclusions and Policy Implications An influential study by Rodriguez and Rodrik(2001) conclude, based upon detailed analysis of probably the four best known papers in the field: Sachs and Warner(1995); Dollar(1992); Ben-David(1993); and Edwards(1998) in addition to analysing Frankel and Romer(1999), and discussing e.g. Harrison(1996), and Wacziarg(1998), that the nature of the relationship between trade policy and economic growth is far from having been settled on empirical grounds. Rodriguez and Rodrik(2001) are sceptical that there is a general, unambiguous relationship between trade openness and growth, which they suspect is dependent on a host of country and external characteristics. Mindful of the critical evaluation of the Rodrik-style arguments by Srinivasan and Bhagwati(1999), we still believe that it is worth pursuing the Rodriguez and Rodrik(2001) recommendation that research would be more productive if it aims at ascertaining the circumstances under which open trade policies are conducive to growth and at scrutinizing the channels through which trade policies influence economic performance. The cross-country regressions don‘t seem to be the best tools for analyzing the problem of understanding the linkage between trade and growth (Srinivasan and Bhagwati, 1999). Nevertheless, the traditional cross-country literature is a rich source of inspiration for our micro level approach. Jalan & Ravallion(1998) argue the macro-level theories of economic growth offer some clues as to why poor areas exist and why they remain persistently poor over a long time, which we will explore in chapters 5-7. The standard Solow-Swan „„exogenous‟‟ growth model suggests that the answer lies in persistent spatial differences in technologies or the policy environment (such as a well-managed decentralized labour-based maintenance programme at the District Council level), in the absence of which the lagging areas will eventually catch up even without factor mobility according to Jalan & Ravallion(1998). They also argue that a somewhat richer set of explanations for poor areas can be found in „„endogenous growth models‟‟ 51 (e.g. Romer, 1986; Lucas, 1988).55 These predict that growth rates will also vary with initial conditions; factor mobility will then be crucial to the prospects for poor areas. Persistently poor areas can thus arise from inequalities in community endowments between treatment and control areas (Jalan & Ravallion, 1998). Jalan & Ravallion(1998) takes living in a designated poor area as exogenous to household choices. However, the existence of spatial externalities may well entail that the growth path of future household living standards is dependent on the community characteristics. The problem is essentially one of omitted-variable bias when there is initial state-dependence in the growth process.56 They use the example that a low endowment of local public goods (e.g. non well-maintained rural feeder roads) may simultaneously induce a lower rate of (agricultural) growth (chapter 4) and a higher probability of the community being declared poor (chapter 5-7). They argue that unless this is accounted for, the value to households of living in an area which is targeted under the sizeable EPFRP will be underestimated. The method that Jalan & Ravallion(1998) propose entails consistently estimating a dynamic model of consumption growth at the household level using (pseudo/synthetic) panel data. The model allows them to test the dynamic impact on households of whether or not the area of residence is covered by the poor-area programme, controlling for initial conditions at both the household and community level. Models in which the first difference of a variable depends on initial conditions have been popular in the literature testing endogenous growth models on country- and regional-level data sets (Jalan & Ravallion, 1998). By using similar variables, exclusive reliance on within-country household-level analysis allows us to enhance the precision of the coefficients and the policy relevance of the recommendations than those derived from a macro level analysis and/or aggregate regressions across countries, in a number of ways (see Deininger and Okidi, 2003, Dercon and Hoddinott, 2005). 55 Jalan & Ravallion(1997) theoretical model for example extends the Cass-Koopmans-Ramsey model of optimal consumption growth in a straightforward way to allow geographical effects on the marginal product of own capital, analoguos to the role of knowledge externalities in the models of Romer(1986) and Lucas (1988). Their model uses longitudinal observations of growth rates at the micro level. 56 The initial conditions – as represented by a dynamic system‘s initial state vector – influence the future evolution of that system (Jalan & Ravallion, 1998). 52 Another advantage of using the rural household as the unit of observation is that one can identify external effects on production processes at the household level which are otherwise lost in aggregation (Ravallion and Jalan, 1996). Another implication is that standard growth-theoretic models derive testable propositions about the steady state of an economy; in our setting one can allow for deviations between the current year‘s data and the underlying model‘s steady-state solution, due to shocks and/or adjustment costs. Ravallaion & Jalan(1998) suggest that we thus obtain a more powerful test of the impact of initial conditions on the evolution of living standards than is possible by only modelling the long-run average growth rates. Although, the other more nuanced literature of in-depth analyses of country experiences discussed in section 4 and section 5 largely suggest a positive correlation between transport infrastructure investment and production, productivity as well as poverty alleviation, there are also a number of unsettled theoretical, methodological and empirical questions that need to be properly investigated before one can conclude with certainty that a given transport infrastructure in a small open land-locked economy such as Zambia has measurable and significant effects on economic development (Banister and Berechman, 2000). Regarding available studies on the effects of rural roads infrastructure investment, most specialized literature has just documented the different impacts that such investment could have on accessibility to product and factor markets and key public (social) services, without controlling the effects of other covariates that could be increasing or reducing the positive impacts resulting from this investment. Moreover, the review study by Chipika(2005) concludes that the linkages between the EIIP/ASIST programme with the other key sectors are variable in Botswana, Kenya, Tanzania, Zimbabwe, Mozambique, Namibia, and South Africa. In many cases this is not acknowledged well in the baseline and impact assessment studies carried out. Yet it is clear that the EIIP/ASIST programme cannot make a great deal of mileage in reducing poverty (as an isolated programme) without forging strong linkages with these sectors 53 and others depending on country specific circumstances. From ωhipika‘s summary table of selected countries with some basic data we see that there consistently is a clear distinction between the macro and micro level of impact of road maintenance; upgrading and construction of projects which have been implemented from a period ranging from 10 to 20 years in these countries (see table A7 in Annex: Chapter 2). He consistently finds that the impact is high at the micro level but low at the macro level with the exception of South Africa.57 The flawed methodological framework used in these types of public projects evaluation has been rehabilitated considerably thanks to the introduction of propensity score matching techniques developed by Rosenbaum and Rubin(1983) and extended by Heckman et al., (1997b), which allows the construction of counterfactual scenarios, sufficiently robust to enable researchers to claim causal relations (Blundell and Costa Dias, 2005). However, this methodological alternative has only recently been incorporated to the analysis of social and economic impact derived from rural roads construction, rehabilitation and maintenance projects by Escobal and Ponce(2003). Based on these recent microeconometric policy evaluation approaches in the subsequent core chapters of the thesis we will focus on considering how the use of household and firm level data could improve the exploration of the determinants of sustained rural growth and agricultural trade. Through the literature review we have shown that there is relative little empirical research on the quality, usage and sustainability of PWP rural roads, nor on its impact on livelihoods over time. The evaluation of such rural assets created through labour-based PWP is considered to be a gap in current policy evaluation literature. Thus, the thesis‘ endeavour to address this knowledge gap is our contribution to the literature, which seeks to provide insights on the medium – to long – term effects of rural roads investment based on the analysis of micro data within one country. 57 See McCord and van Seventer(2004) who argue that the link between EIIP and poverty is week unless well targeted. 54 Chapter γ: Zambia‟s Eastern Province and Eastern Province Feeder Road Project 55 γ.1. Key contextual information about Zambia‟s Eastern Province Overall 47 percent of Zambia‟s land area is defined as agricultural land. However, from 199η to β00β of Zambia‘s η,260,000 hectares of arable land the percentage share under cereal production fluctuated between 10 and 15 percent. The irrigated land as a percentage of cropland only increased slightly from 1.33% in 1994/1995 to 2.95% in 2002/2003 despite the plentiful water supply from lakes and rivers dissecting the country.58 εoreover, it wasn‘t until the period 1998 to β008 that agriculture‘s value added share consistently exceeded 20% of GDP (WDI, 2010) due to the fact that agriculture has been one of the faster growing sectors of the economy (FAO, 2009). Zambia‘s Eastern Province covers an area of 69,106 square kilometres and today has 8 districts namely Chadiza, Chama, Chipata, Katete, Lundazi, Nyimba, Petauke and Mambwe. In 2000(2010) Eastern Province had a population of 1,300,973(1,707,731). Of this population, 49.4 (48.96) per cent were male and 50.6 (51.04) per cent were female. Eastern Province was growing at an average annual population growth rate of 2.72 (2.66) per cent between 1990 and 2000 (2000 and 2010) (CSO, 2001, 2011). There are several unique features to the Eastern Province. Despite experiencing the least percentage growth among Zambia‘s nine provinces, the agricultural households in Eastern θrovince still constitutes Zambia‘s largest population.59 The 2000 census of population also found that the whole province had the highest number (231,120) and the second highest percentage of female headed households (19.8%) (CSO, 2001). Of the 221,703 agricultural households in rural Eastern Province in 2000, the majority (38.6 percent of the total) were engaged in the three major agricultural activities: Crop growing, livestock- and poultry rearing while 28.8% were involved in crop growing and poultry rearing only. Only 20.6% were exclusively engaged in crop growing. 58 ηne estimate shows that Zambia‘s water potential could enable it to irrigate up to η00,000 hectares of land. Currently only 13 percent of this potential is utilized, mainly by medium- and large-scale farmers. 59 At the same time in 2000 Eastern Province had the lowest proportion of urban population at only 9 percent. Moreover, 4.1% of the agricultural households had an urban residence in 2000. 56 Eastern Province‘s economy is agro-based and depends entirely on the soil with maize, cotton and tobacco being the major cash crops most of which are intended for the export market. The small scale farmers remain the key players of the local economy (Lungu, 2006; cf. chapters 4-8). Zambia has been classified into 4 broad agro-ecological region zones on the basis of the average precipitation pattern and the quality of the soils.60 Eastern Province has two distinct agro-ecological regions—the Eastern Plateau and the Luangwa Valley (see Map A2 in appendix). Central, Southern and Eastern Plateau known as agro-ecological region II covers the Central, Southern and Eastern fertile plateau of Zambia (CSO, 1994). It is characterised by: Moderate rainfall ranging from 800 and 1,000 mm of annual rainfall.61 Zambia Meteorological Service datasets include monthly rainfall from 1993 collected from weather stations in the four districts of Chipata (Chipata and Msekere stations); Lundazi; Petauke; and Mambwe (Mfuwe station). The Eastern Province timeseries represents the provincial average, given our assumption (see table A3.1, in Annex 3). The years 1996/1997 and 2003/2004 were periods of above average rainfall levels, whereas the agricultural seasons 1997/98 and 2004/2005 were respectively below and above the average rainfall levels in Eastern Province. In Eastern Province the rainfall is concentrated between October/November and April/May, during the other months there is no rainfall at all. There has been considerable volatility in agricultural growth driven by high variations in rainfall (see figure 3.1) and the low share of irrigated land. Crop production was negatively affected by the severe 1992 and 1995 draught. Both short-term fluctuations in rainfall as well as the long-term effects of climate change have made rural farm households vulnerable to successive periods of famine (Chapters 5-7). 60 Region I The Luangwa - Zambezi River valley zone. Region IIA The Central, Southern and Eastern Plateau. Region IIB The Western, semi-arid plains. Region III The Northern, North western high rainfall zone (Siacinji-Musiwa, 1999). 61 In the valley areas, the rainy season tends to begin and end earlier than elsewhere. 57 Figure 3.1: Rainfall pattern in Zambia‟s Eastern Province, 1994-2005 1400,00 1200,00 Eastern 1000,00 Rainfall (mm) Chadiza (301) (i) Chama (302) (iii) 800,00 Chipata (303) (i) Katete (304) 600,00 Lundazi (305) Mambwe (306) (iii) 400,00 Nyimba (307) (ii) Petauke (308) (ii) 200,00 Long-term Mean 0,00 Notes: 1994 and 2002 were modest drought years in Zambia. Source: Authors‘ based on Zambia Meteorological Service data. 3.2. The Eastern Province Feeder Road Project During Zambia‘s early post-independence years there was some initial investment in rural infrastructure. However, the reallocation of funds to the mines of the Copperbelt and the industries of Lusaka led to the deterioration of existing roads in more remote rural areas (World Bank, 2004; Thurlow and Wobst, 2005). Concerning the selection process of the Eastern Province chosen for project placement, in the first half of the 1990s the road network in Eastern Province was in bad shape due to lack of maintenance and repair.62 Generally the road accessibility within the province was very poor especially in the rainy season. As a result most private transporters were not willing to put their vehicles on the neglected routes (Eastern Province Chamber of Commerce and Industry). So, when the UN General Assembly classified Zambia as a Least Developed Country (LDC) in 1991, thereby rendering Zambia eligible to receive UNCDF assistance, 62 Eastern Province lies between Latitude 10 and 15 degrees South and Longitude 30 and 33 degrees East. The Province lies between two international boundaries Malawi in the east and Mozambique in the South. The North Western boundary is marked by the Luangwa River, which separates Eastern Province from Lusaka, Central and Northern Provinces. 58 UNCDF fielded an identification mission in 1993, which resulted in the formulation of two projects in the Eastern Province:  The Rehabilitation and Maintenance of Feeder Roads (FRP) project; and  The District Development Planning and Implementation (DDP) project. The FRP started 12 June 1996 as a pilot for the introduction of LBT for road construction and the establishment of maintenance systems for the feeder roads. It was executed based on the θroject documentμ ‗Addendum to Project Agreement‘ signed β1st of October 1996. The FRP ended 31 December 2001. The FRP was a project within the Ministry of Local Government and Housing (MLGH) and implemented by the District Councils of Eastern Province,63 which was designed to build and strengthen capacities in the local authorities and local private sector to rehabilitate and maintain feeder and urban roads through contracting systems. The FRP was enhanced through the linkages to the DDP, which supported participatory planning and strengthening of the service delivery capacity of District Councils. In principle, FRP operates within the framework of the DDP, which aims at developing the capacity within the districts to plan and manage public works, and involving communities in all development processes (Rwampororo et al., 2002; Clifton et al., 2001).64 The project had four immediate objectives: i. ii. iii. iv. Develop capacity of district councils works departments to plan, design, implement and manage road rehabilitation and maintenance works by establishing Contract Management Units (CMU). Develop a private sector construction industry capable of rehabilitating and maintaining feeder roads using labour based methods by training and equipping small-scale road rehabilitation and maintenance contractors. Improve access to highly productive agricultural areas. Create direct employment in the rural communities by encouraging participation of local communities in road works (Clifton et al., 2001; Rwampororo et al., 2002). 63 The District Councils being the feeder roads authority in Zambia, act as client organizations whose responsibilities include: to select and prioritise roads, to prepare and sign rehabilitation and maintenance contracts and tender documents, to supervise and certify the works, to pay the contractors, etc. The District Works Departments are trained to act as contract managers in all these aspects. 64 The FRP operated alongside the DDP, also funded by UNDP and UNCDF, and had the same Project Manager. In principle, FRP operated within the framework of the DDP, which aimed at developing the capacity within the districts to plan and manage public works, and involving communities in all development processes (Rwampororo et al., 2002; Kalinda, 2001). 59 Table 3.1: Eastern Province Road Sector Network, (km) i. ii. iii. iv. v. vi. Trunk Main District Feeder (*) (Primary Feeder) Total (i+ii+iii+iv) 415 179 1,516 3,862 -2,359 5,972 Source: Road Sector Investment Programme (ROADSIP), Bankable Document, August 2001. Note: (*) Feeder roads have a further internal classification of primary, secondary and tertiary. The EPFRP was redesigned in May 1996 to rehabilitate 580 km of feeder roads and to place 700 km of roads under regular routine maintenance.65 The target of 580 km rehabilitation was revised to 450 km by the mid-term evaluation, as a result of the increase in the amount of gravelling required and the resulting increase in costs per km (US$9,000 rather than US$7,000). Table 3.2: Eastern Province Feeder Road Network EPFRP Km Rehabilitated Roads Rehabilitated and Maintaned Roads Total KM Percentage of total feeder road network Percentage of primary feeder road network 21 404/3,862 = 10% 404/2,359 = 17% Enhanced Maintenance Roads 383 429 404 429/3,862 = 11% 429/2,359 = 18% Total Maintained Road 833 833/3862 = 22% 833/2,359 = 35% Sourceμ Authors‘ calculations based on Rwampororo et al. β00β. What is not clear in the document is whether the 450km of rehabilitated roads are included in the 700km to be maintained or are additional to them. If however, the maintainable kilometres were based on an expected capacity of 25 contractors capable of maintaining 30 km each, then this would give a maximum of 750 km of feeder roads under maintenance. It is therefore assumed that the 580 km (77%) were to be part of the 750 km, especially as all rehabilitated roads should be immediately placed under a maintenance regime. In relation to the feeder road network in the district this improvement to roads would account for 750 / 3,862 = 19% of the entire feeder road network in Eastern Province shown in table 3.1. However, given that the roads were prioritised and the most important links were identified to be rehabilitated and maintained, the impact in the Province would be greater than the proportion of the network addressed (see last column in table 3.2).66 65 Costs continued to rise over the remaining project period. The ‗appraisal‘ stage used to select the road links to be rehabilitated can be thought of as ‗ex-ante evaluation (Van de Walle, 2009).‘ 66 60 The number of kilometres of improved feeder roads under routine maintenance was zero for a period of almost two years upon completion of the EPFRP. There were practically no funds entering the districts in Eastern Province for routine maintenance and there were no funds allocated from National Road Board (NRB) and/or Ministry of Local Government and Housing (MLGH) for this activity since the 1999. Some of the activities funded under the 1999 budget were carried out in 2000, but no separate allocation for 2000 was made. Therefore the figure of 450km achieved for maintenance can only be viewed as a previous maximum. It is also worth noting that only 21 km of the 450 km under maintenance were on roads rehabilitated under the FRP. This means that (404-21 = 383 km) of rehabilitated feeder roads have had no maintenance in 2002 since they had been improved. The remaining (450-21 = 429km) of maintained feeder roads were improved and maintained through enhanced maintenance contracts. This brings the total to 833 km of road addressed by the FRP (table 3.2) (Rwampororo et al., 2002). These maintained feeder roads are illustrated in Map 3.1. 61 Map 3.1: Illustration of the Eastern Province Feeder Roads REPUBLIC OF ZAMBIA EASTERN PROVINCE CHA M A DISTRICT C ha m a Lunda zi NORTHERN PROVINCE LUNDA ZI DISTRICT M A M B WE DISTRICT M 12 M ALAWI C H IP A T A P ETA UKE DISTRICT T4 C ha diza Ka t e t e P e t a uk e NYIM B A DISTRICT T6 T4 M OZ AM BIQUE N yim ba LEGEND UNDP Feeder Roads Project Int ernat ional B oundary Dist rict B oundary SCALE (A pprox.) 0 25 50 Trunk / M ain Roads Km 75 100 125 150 Source: UNDP EPFRP Document. 62 Dist rict Roads 3.2.1. The Implementation of the Feeder Road Programme In the original project design, it was intended that the project would respond to the need to transport a bumper crop of maize from the 1993/94 agricultural season. Road rehabilitation was expected to be done speedily using equipment-based methods. A lack of locally available contractors to undertake the works posed a serious problem for the project and was the basis for the redesign carried out in May 1996 (Rwampororo et al., 2002; Clifton et al., 2001). A well-documented system for identification and selection of potential contractors was developed and has been established within the district technical departments. The project developed objective criteria that were used in the selection of contractors sent for training at the Road Training School (RTS) in 1996. 13 out of 21 routine maintenance contractors who received training at RTS in 1996/7 were operational by the end of the year 2000. 7 out of 10 small scale contracting firms who qualified for continuing support after the second trial contracts were still operational in 2005 (Clifton et al., 2001). Thus, the EPFRP carried out three main areas of activity, including:   Training of 8 small-scale labour based rehabilitation contractors and rehabilitation of some 450km of roads during the life of the project; Training of 25 small-scale maintenance contractors with 700km of feeder roads coming under a routine maintenance regime by the end of the project Until the first half of 1998 the EPFRP concentrated activities on training of contractors (and local consultants) to carry out rehabilitation and maintenance of roads with the Project Management Unit (PMU) carrying a largely line management and execution function in place of the district councils (Clifton, 1998). The 8(7) small-scale contracting firms and 25 maintenance contractors, who were identified and trained at RTS for the project,67 did not have suitable equipment needed for use in labour-based road rehabilitation works. UNCDF and MLGH decided to procure seven sets of light equipment and hand tools using part of the Feeder Road Rehabilitation 67 Initially, 10 labour-based rehabilitation contractors and their staff were trained. 8 of them were selected for a continued training. 63 Fund. However, since all funds under this Feeder Road Fund were to be spent on road rehabilitation works, MLGH decided to set up a credit scheme to facilitate the loan repayments by contractors. The total equipment loan package to each contractor amounting to USD 134,000 was designed to be liquidated in four years (table A3.3 in Annex 3) (June 1997 - June 2001) (Clifton et al., 2001). The maintenance programme for example comprised 50 km of maintenance in Chipata district, which included the following types of works: Road formation; Culvert construction; Drainage works; Spot gravelling; and Vegetation control. In Chadiza the 50 km implied: Bush clearing; Culvert construction; Drainage works; and Heavy grading, whereas in Katete the 50 km maintenance work entailed: Bush clearing; Reshaping; Spot gravelling; Pothole patching; and Drainage works with similar types of activities in the other districts. In 1998 this level of maintenance activity should be compared with the estimated average length of roads in each district of around 500 - 600km of gazetted district and rural roads plus ungazetted rural roads. However, until July 1998 no definitive list of district roads had been identified for Eastern Province and despite requests, only a few districts were able to supply a list of district roads or a sketch road map (Clifton, 1998). This changed in July when GITEC Consult on behalf of the Department of Infrastructure and Support Services (DISS) of the MoLH published two volumes of the District Feeder Road Maps and Lists. By May/June 1998 15 labour-based maintenance contractors had been trained and had completed 2 trial contracts. They then were engaged on their first full contract. The project experienced delays in the provision of funds for maintenance and the contractors did therefore had a long break between their first contracts. The funding issue should now be under control with maintenance funds for the Districts pledged by the Government and the NRB. It was however expected that they still needed intensive guidance and support from the project to pick up their duties in an effective manner. The 8 rehabilitation contractors subsequently completed their first trial contracts (FTC) at the Tamanda Loop in Chipata District. Following a comprehensive performance 64 evaluation by the RTS and the FRP, 7 contractors were finally selected to continue with their second trial contracts (STC). These rehabilitation contracts were located in 5 different districts. The 7 contractors started the work on their second contracts during the last week of November 1997, and it was expected that the contracts would be completed by the end of April 1998 (Clifton, 1998). To facilitate an objective transparent process for selection of roads for rehabilitation and maintenance the project was to develop and install a road selection and priority ranking criteria based on road condition inventories, traffic counts and key socio-economic data such as agricultural production. FRP developed formats for road condition inventories and traffic counts, which have been used by District Councils for collection of field data prior to contract preparation. However, a comprehensive database was not developed because DISS, with technical assistance from GITEC, carried out road condition and inventory surveys and developed a ranking system for all feeder and urban roads in Zambia. The data was processed into maps and booklets using the GIS system and this was being used by District Councils in Eastern Province for selection of roads for road maintenance interventions (Clifton et al., 2001). A baseline study was carried out in 1996 in Eastern province at the beginning of the EPFRP that provides the socio-economic profile per district. Unfortunately, it was not followed up, which would have helped in assessing changes that may have taken place over time in order to ascertain the increased access to social services such as health, education, water supply and sanitation (Sakwiya, 2005; Rwampororo et al., 2002).68 Some traffic counting had also been done prior to the selection of the feeder roads to be included in the EPFRP. Thus, traffic, its economic value and population size in the area was taken into account, along with the presence of schools and health clinics. At the time of the documentation of each road the project sent an assessment teams, which carried out the traffic counts and an inventory picking up all the key features of each particular road: The villages; estimating the population in the villages; the location of schools and 68 The National Consultative Forum comprising: The district and provincial administration; UNDP; MoLH; and the ILO, met once every quarter to review of progress of the EPFRP and it produced quarterly progress reports (Sakwiya, 2005). In addition to two 1998 Mid-Term Evaluations of the FRP and DDP, and the final evaluation of the EPFRP, a study by Kalinda(2001) was done on the lessons learnt on the DDP and another one by Clifton et al.,(2001) on the lessons learnt from the FRP. 65 health clinics; water points and gravel pits. Based upon this information the project was able to prepare the documentation for either maintenance or rehabilitation, which provided the foundation on where the road had to be rehabilitated and where it had to be maintained.69 Moreover, the EPFRP was targeting roads that had already been listed in the road document, so the PMU assumed that an environmental impact assessment had already been done. Initially the PMU, which consisted solely of engineers, was supposed to have benefited from the planning side of the DDP, but due to delay, the project preceeded without a proper ex-ante project appraisal. This was not addressed since there wasn‘t any social scientist included in the project design to identify the socio-economic impacts of the project (Simon Tembo, 2005). „Kawaye Chataya‟ company owned by Isaac Manda was one of the 7 rehabilitation contractors, which contributed 3 km to the rehabilitation of the 20.4 km Tamanda Loop (RD118) through the FTC in end-1997 (table A3.12, Annex 3). Kawaye Chataya was then awarded a contract (CHI/98/01) for full rehabilitation of the feeder road “Eastern Dairy to Madzimoyo” (U33) in Chipata district, which was running from March 1998 until end 1999. In total the company rehabilitated 26.1 km, which corresponded to around 1.6 km of rehabilitation per month or around 2.3 workerday per km. The level of supervision had improved when the contractor was rehabilitating the at Tamanda Loop as the 2nd full contract (FTC-1), compared to the first trials in Chadiza). After that contract the Kawaye Chataya contractor received another contract (θET/99/0γ) covering a β1.7 km segment of the road (RD41γ) from ‗T4 to ωhataika‘ in Petauke District, where the productivity increased to 2.2 km per month on average or to 1.7 workerday per km (Table A3.12, Annex 3). Despite the fact that the terrain – clay – was tougher in Chipata than in Peauke district, the contractors used the same equipment (Table A3.3, Annex 3). On the other hand, the gravel pits were adjactent - within a 100 meter distance - from almost all the feeder roads covered. In order to keep to the workplan, the company also started to manufacture its own culverts, because the original local supplier either didn‘t keep the deadlines or the standards weren‘t high enough. The 69 This pre-project implementation documentation of the baseline is found in the εoδH‘s project archieves. 66 company also manufactured a number of major structures, for example one drift of 30 meters on the Chiparamba road (B). Of the 8 trained small-scale LBT contractors one contractor failed. The equipment of that contractor went to another „Justed Development‟, which also was owned by Isaac Manda. Thus, after the trial at Tamanda Loop Isaac εanda‘s subsidiary ‗Justed Development‘ went to Chiparamba road (A) likewise in Chipata district afterwhich the company started work on ‗T4 – ωhikhombe‘ road (Uβγ) in Katete District. Finally, the company went to Mambwe District (table A3.12, Annex 3) (Isaac Manda, 2005). Another LBT contractor was named „Rapid Construction‟. This company was owned by Agrippa Mwanza. This contractor was also, in addition to working on 3 km of Tamanda Loop as a FTC, awarded two full rehabilitation contracts each of one year duration in Katete District. They covered respectively 17.η km of the ‗εbinga-tθ‘ road (Uβ9) and θ.1 km of the ‗Kavulanungu-Tθ‘ road (unclassified). It took around one year to complete the work. Moreover, the company was also contracted to rehabilitate feeder roads in Petauke Districtμ Two segments of respectively 7.7 km and 10 km of the ‗T4εumbi‘ road (RD41η) as well as one segment of 11.γ km of the ‗Sichilima-εawanda‘ road (RD135). It took around 3 months to to complete the 7.7 km (STC-7/7) contract and 8 months to complete the 11.3 km (PET/99/01) contract (table A3.12, Annex 3). After the grading had been completed the contractor experienced a lot of problems with labour recruitment for the ‗Kavulamungu - T6‘ road, which was built in a scarcely populated area in the bush. The contractor had to import labour from the urban areas to complete the contract. Moreover, the contractor experienced a lot of troubles with the urban workers who e.g. complained about the allocated tasks. Their presence also created a lot of tensions with the other labourers recruited from the rural areas. This contractors also manufactured two major structures, namely two drifts of a total length of 60 meters on the ‗Sichilima – Mawanda‘ road. The company was able to rehabilitate β.η km per month on average. It seems as though drinking had a negative impact on the labourers reliability and productivity (Agrippa Mwanza, 2005). 67 A fourth of the seven LBT contractors was Henry Dane Zule, the Managing Director of „Mtondo Construction‟ company. This contractor rehabilitated like the rest 3 km of the Tamanda Loop; 7,4 km of Chiparamba road (B) from the Sub-Centre to the Mfuwe road, which also included a 30 meter drift lasted less than 6 months; and a 3 km Kenneth Kaunda GER-Stadium road. All these roads were located in Chipata District. The roads were impassable prior to the beginning of the project due to the sheer size of the potholes. In Lundazi district ‗εtondo ωonstruction‘ under an one-year contract (LUN/98/02) rehabilitated 20.1 km of the completely delabidated ‗Mwase – Lundazi‘ road (R110), which was considered to be in an even worse state prior to the project start than the Chiparamba road. And in Petauke districts he rehabilitated two sections on the ‗T4 – Chikalawa‘ road (R1γ) of respectively β0 km and η km. The contractor was able to rehabilitate on average 1.5 to 2 km per months when all resources were available on time (Table A3.12, Annex 3). For example, the recruitment of labourer in certain areas such as the Chiparamba Loop were not as easy as in other more densely populated areas. If the local labour force already was occupied in other income generating activities mainly farming, then the salary offered by the EPFRP (i.e. ZMK2,500 per task) wasn‘t attractive enough to give them an incentive to join the project. In fact there was a general tendency not to make the issuing of the contracts coincide with the slack season of the agricultural sector as would have been expected. As a consequence the contractor had to recruit from Chipata town. But these town dwellers also brought a lot of problems such as demanding wages every second day (instead of receiving a salary at the end of the month of ZMK27,500), which in turn was being misued for buying beer. Hence the urban labourers were brought back to town, once the local communities started to respond positively to the vacancy announcements with the aim to earn a non-farm income in addition to the farm income. On average the company would employ not less than 50 workers at any moment during the workplan, sometimes increasing the numbers of workers to as many as 130. The trained supervisors were assisted by so-called ‗gang‘ leaders. After the contract had been awarded the company would carry out a sight visit, during which it would meet with the local (traditional and/or political) leaders. The contractor would inform them about the forthcoming road rehabilitation and the need for the local community‘s assistance 68 along the roads in the recruitment of labour for a given pre-specified day. The workers from Chiparamba were considered the most productive and hard working (Henry Dane Zule, 2005). The last contractor interviewed was Ephraim Kamoto who owned „Wheeltrax Contractors‟. In addition to the Tamanda Loop (RD118) this contractor also was contracted (CHI/01/01) to rehabilitate 5 km segment of ‗Kapata - Chizongwe road‘ (U01) of which only 0.5 km was rehabilitated. In Lundazi district the company was responsible for three sections of the ‗Lundazi – Mwase‘ road (RD110) of respectively θ.1 km (STC3/7); 15.3 km (LUN/98/01); and 9.7 km (LUN/99/01) from late 1997 to late 1999. The contractor recalled that the loop was in such a bad state before 1997 that motorized vehicle couldn‘t pass. Thus, the cash crops prior to the EPFRP were transported via oxcards or bicycles only. The inputs would only be brought to the part of the roads, which was somewhat passable. The traders who ventured into the areas at the time only were small-scale traders (chapter 8). Finally, the company was awarded a contract (CHA/99/01) to rehabilitate 20 km of the ‗Chadiza – Tafelanson‘ road (RD40η) in Chadiza district from late 1999 to 2000 (Table A3.12, Annex 3). Afterwards, the company went to Chipata district again to rehabilitate the ‗Chipata-Chizongwe‘ road for another feeder road project between 2002 and 2003. The company also won a maintence contract in Wbuwi – Zozwe road in Chadize district. Moreover, the company also participated in a joint-venture under the Smallholder Enterprise and Marketing Programme (SHEMP) project maintaining the ‗εacha – ωhongo‘ road from β004 to β00η in Chipata district, which was extended for further gravelling on certain parts that hadn‘t been gravelled. During the rainy season the project advanced only 1.5 km per month on average. The project got slowed down because the workers would tend to their farm fields for cultivation purposes during the rainy season, which meant that the project would have less manpower. Moreover, in Lundazi districts the Tumbuka tribe didn‘t allow their married women to take part in the project as labourers. Moreover, the project would stand still for up to 3-5 days during heavy rainfall (Ephraim Kamoto, 2005). 69 This statement was qualified by the project‘s national road engineer, Simon Tembo(2005), who mentioned that „seasonality‟ was incorporated into the planning, and in that sense it was not considered as a big problem. During planning the PMU set targets for the rainy season and the dry season. The targets for the months coinciding with the farming season was reduced to 1 km completed per month, including structures. For the dry season during and the slack period where labourers were expected to be free of other duties the targets were set up to 1.8 km per month. Contractors who took advantage of working in the right period were in some cases producing more than 2 km per month. During the rainy season some of the contractors would even close their sites allowing their labourers to attend to their fields (Simon Tembo, 2005). Concerning to whether poor areas were targeted by the EPFRP and on what basis. According to the observations by the interviewed LBT contractors, all the feeder roads included in the EPFRP were impassable before the project was launched in 1998 because the roads sector lacked funds and was poorly managed. This resulted in severe neglect and large portions of the network, which were in need of rehabilitation and maintenance. The deterioration of the road network meant that productive areas were inaccessible during the rainy season, and some roads impassable even in the dry season. This hampered the introduction of farm inputs and the as well as the timely evacuation of harvested agricultural produce (Rwampororo et al., 2002). According to the project documents the EPFRP was designed to stimulate the agricultural production. It was thought that rural household income and quality of life would improve through rehabilitation, maintenance and sustainable use of the roads. Thus, the attainment of these development objectives would lead to the contribution to sustainable development of Eastern Province through the establishment of a comprehensive integrated strategy for rural infrastructure for rural development, relying to the extent possible on locally available private sector resources and the capacity of District Councils‘ (ωlifton, 1998). Concerning political influence in the choice of areas of intervention. The possibility of rural roadwork in both Northern and Eastern Provinces was originally 70 considered as neither area had attracted major assistance in the sector. In consultation with the Government of Zambia, the Eastern Province was identified as an area of concentration for UNCDF assistance, based on its development needs and as it was the next most productive province after Central, Lusaka and Southern provinces (Rwampororo et al., 2002). In other words, there is evidence of selectivity bias in terms of the EθFRθ‘s systematic selection of the location of the feeder roads in potentially more favourable agricultural areas. Finally, the PεU didn‘t considered that carrying out a cost-benefit-analysis (CBA) made any sense, because it is comparing usage in terms of how many motorized vehicles were using one particular feeder road against another road. First of all, because the traffic count would already give an indication of the economic value of that segment of road. Secondly, some roads weren‘t trafficked because of bottlenecks, e.g. a collapsed bridge or heavy deterioration of road surface so that the local communities only use the road by alternative means of transportation such as oxcarts. ψut that doesn‘t mean that the area doesn‘t have an economic potential. A CBA based on traffic count would continue to disadvantage potential rural areas simply because they are not accessible to road traffic at the time of the counting. Hence, the EPFRP relied on agricultural potential as a selection indicator rather than traffic volume.70 For most feeder roads the assessments were based on traffic count; inventory and surveys carried out within 1-2 months. The PεU didn‘t have enough time to carry out detailed analysis on average speed; passability etc., because there was a lot of pressure to have the contractors operating continuously (Tembo, 2005). 3.3. Justification of Analytical Approach The feeder roads were clearly not randomly placed, and it is likely that the factors that mattered to the road placements also affected the outcomes. Van de Walle(2009) shows that there are three distinct ways in which the problem of „endogenous road placement‟ (i.e. selection bias) can arise in this context. Table 3.3 provides a summary of the generic types of endogeneity. In each case there is some third relevant variable X that 70 The SHEMP project ran into problems using the CBA approach to prioritize the roads. 71 is jointly correlated with the road‘s placement and the outcome variable. Van de Walle(2009) emphasises that failure to properly control for X can generate large biases in estimates of the impacts of rural roads. Table 3.3: Types of endogeneity of programme placement associated with the existence of third variables (X) Source: Van de Walle, 2009:31. The first type of endogeneity is when there is an initial condition that is correlated with road placement and is also correlated with the level of the outcome variable. For example, if places within the eight districts in Eastern Province with agricultural cash crop potential are selected to get better feeder roads, then a bias can be expected when assessing road impacts on agricultural incomes (see chapter 4), given that the measured values of those incomes will partly reflect the differences in initial potential according to van de Walle(2009). Van de Walle(2009) mentions a second type, in which the initial condition that influences placement also influences the changes in outcomes as would be the case if roads were targeted to poor areas with attributes that determined both their poverty and their subsequent growth path. This is likely to be an important source of endogeneity in the rural road context (Jalan and Ravallion, 1998) (see chapters 5-7). Finally, Van de Walle(2009) also cautions that the endogeneity problem can arise when changes in programme placement are a function of time-varying factors, such as when road expansions accord with changing economic or political conditions that are 72 themselves correlated with the changes in outcomes. For example, if new feeder roads are rehabilitated or maintained in places that are growing faster due to the presence of export oriented cash crops such as Tobacco and Cotton in various parts of Eastern Province or due to the proximity of the θrovince‘s district centres, there will clearly be a problem in disentangling the rural road‘s impact on incomes. Van de Walle(2009) notes that in all three cases the problematic initial condition or time-varying factor may be observed or unobserved; this difference has important implications for the methods used to try to address the endogeneity problem. Van de Walle(2009) does allude to certain academic positions that argues that endogeneity is not a concern when, as is the case in this thesis, one is studying the impacts of rural roads at the micro (farm (chapter 4); household (chapters 5-7); farm, or the road users (chapter 8)) level, since assignment is to communities or geographic areas. However, she qualifies this by warning that there could still be a problem in this case because there are likely to be individual characteristics that are unobserved but geographically determined ─ or geographically correlated (such as through location choice (see chapter 8)) ─ and are also correlated with the things that influence programme placement. Thus, important generic sources of endogeneity bias are associated with EPFRP interventions. Initial conditions, such as poverty levels or pre-existing rural transport infrastructure networks in Eastern Province, are likely to determine road placement, as well as to influence the levels of relevant outcome variables and also the subsequent growth paths and prospects of the village communities, farms and rural households within them. Van de Walle(2009) argues that a credible evaluation methodology needs to correct for these potential sources of selection bias through a careful construction of the counterfactual. Furthermore, given that roads may have dispersed effects on many factors, information on conditions prior to the intervention must naturally be available. In Table 3.3 van de Walle(2009) provides a summary of specific solutions associated with the three types of endogeneity. 73 Single difference comparisons can be either reflexive (before and after) comparisons that track gains solely in project areas, or with and without comparisons that take single differences in mean outcomes between participants and non-participants using cross-sectional data as in chapter 5 and chapter 7. Van de Walle(2009) believes that the reflexive comparison is unable to separate the road impact from the general economic and other changes that would have happened without the EPFRP project. The with and without comparison can only identify the impact of the roads if one can deal with the selection bias. If it is believed that the data do not include some of the relevant X variables that jointly influenced programme placement and the level of the outcome measure then Van de Walle(2009) considers a double difference (DD) (or „difference-in-difference‟) design where a first difference is taken between outcomes in the project areas after the program and before it, minus the (second) corresponding difference in comparison nonproject areas. This can be obtained from a simple OLS regression where the dependent variable is the change in the measured outcome variable between the pre-intervention and the post-project data and the only explanatory variable is a dummy for whether the areas participated in the EPFRP. Used alone, this method applied in chapter 4 assumes that placement is a function of initial conditions (observed or unobserved) that take the form of an additive time-invariant error term (type 1 endogeneity, Table 3.3). Once this is differenced out, any remaining differences in outcomes between the treated and nontreated can be attributed to the EPFRP (Van de Walle, 2009; Wooldridge, 2002; Deaton, 1997). A conventional DD does not allow for initial conditions (type 2, Table 3.3) or time-varying factors (type 3, Table 3.3) that also influence subsequent changes in outcomes over time. To deal with these sources of endogeneity bias, Van de Walle(2009) suggests the importance of controlling for the initial conditions and any time variant factors that simultaneously influence the placement of roads and subsequent growth rates. One approach used in chapter 4 for correcting for these biases combines double 74 differencing with propensity score matching (PSM) to select ideal comparison areas from among the sampled non-project areas (Ravallion and Chen 2005; Lokshin and Yemtsov, 2005; Mu and van de Walle, 2007 and 2008; Chen et al. 2007; reviewed in chapter 2). Van de Walle(2009) proposes that the better we measure initial conditions (such as political clout, local leadership, social capital and empowerment) and relevant time variant attributes, the less concern there will be that these latent factors might influence changes both in road placement and outcomes over time. However, one can never completely rule out the possibility of omitted initial conditions or changes over time in those variables that are correlated with placement or outcomes. The idea behind the Double difference combined with Instrumental Variables is to find a variable ─ known as the Instrumental Variable (IV) ─ that determines the road placement (or changes in placement) but does not determine outcomes conditional on placement. In other words, the IV only affects outcomes through the programme placement, thus allowing identification of impacts robustly to selection bias. In a first step the IV is used to explain the road access variable. The predicted value from that equation is then used instead of the road variable in the DD outcomes regression (Van de Walle, 2009). A good IV purges the evaluation of endogeneity bias. It is the only option for dealing with an unobserved X that is thought to be correlated with placement or changes in placement and changes in outcomes. (The latter requires an IV that is correlated with the changes in placement but only affects outcomes via changes in placement.) (ibid.). In a situation where longitudinal data on households or communities is available for at least three periods, Van de Walle(2009) propose that it may be possible to implement a dynamic panel data growth model to estimate the impacts of roads on income or consumption growth. Jalan and Ravallion ( 2002) use a micro model of consumption growth with Chinese household panel data covering 5 years. Dercon et al. (2007) try something similar to assess the impacts of changes in road access on consumption in Ethiopia (chapter 2). Unfortunately, it was not possible to find 75 longitudinal data in Zambia with at least three points in time and information on road access for a particular road project area that one wants to evaluate. The potential for randomization to evaluate roads and deal with selection bias seems limited according to van de Walle(2009), although she does mention that she could imagine a situation whereby a government could be convinced to randomize road interventions through a phase in of the road project. However this common approach to randomizing does not appear promising given that the full impacts of a road intervention may take a long time to emerge. Experience suggests that feeder road project phases do not typically get implemented in a short period of time but often over the course of a few years as different road links take longer to get implemented than others as was also the case with the 1998-2001 EPFRP. Van de Walle(2009) suggests that it will be necessary to control for the heterogeneity of factors that jointly influence outcomes and road placement using the above mentioned the non-experimental methods instead. The core chapters of our thesis is primarily concerned with the simple binary case where some units are assigned the rural road and others are not. The treatment variable is then a dummy which takes the value 1 for being in the project area and 0 for non-project areas, from which an appropriate comparison group can be drawn. This is reasonable in the case of small rural feeder roads where the benefits go to a selfcontained area and one can identify non-beneficiaries (Van de Walle, 2009). Finally, one approach used in chapter 7 would be to use the distance to the feeder road as the treatment variable under the assumption that accessibility to the road as measured by distance is still an important aspect of the benefits. Regressing outcomes on distance to the intervention, van de Walle(2009) warns that one must still worry about the same endogeneity problems as before. For example, the possibility that poor people live in the more inaccessible places at further distance from the road. She argues that in this case one must control for initial conditions in a regression context since matching is no longer feasible. 76 The foregoing discussion based on Van de Walle(2009) suggests that a credible rural roads impact evaluation will require panel data, including pre-intervention baseline data. This will need to be available for both project and non-project areas that resemble the project areas to enable creation of an appropriate counterfactual. In order to further control ─ whether by means of a regression or a matching model ─ for initial conditions that may have led to placement as well as outcomes, one will need detailed information of baseline attributes that potentially influenced selection into the project as well as appropriate controls for exogenous time varying factors. Unfortunately, such an ideal dataset is not available. The data should also contain a set of outcome indicators of interest, controls for heterogeneity, and data that allow one to differentiate between welfare groups. There are tremendous gains from data that is specifically designed and collected for the evaluation (Van de Walle, 2009). Unfortunately, such a dataset is not available either for our purpose. Concerning outcome variables, these can be numerous in this thesis. Chapters 57 focus on a single welfare or poverty indicator, rural household consumption, whereas chapter 4 focuses on farm production/productivity, assumed to encompass and reflect the net effect of a myriad of factors. Collection of such data can be difficult to do well and needs a carefully crafted household survey and well-trained interviewers. In chapter 5 we work with a tested consumption module from the existing Living Condition Monitoring household survey questionnaire, which relies on CSO interview teams with experience in such data collection which ensures data quality. Using an indicator of welfare as the outcome indicator has the advantages that it allows a distributional analysis of welfare impacts in chapter 6, and facilitates analysis since the focus is on one all-encompassing indicator only. By allowing time to elapse between project completion in end 2001 and the post-project data collection in respectively 2004 (for chapters 5-6) and 2005 (for chapter 7) so that impacts have had a chance to be reflected in the welfare measures. However, focusing on a single outcome variable does not necessarily reduce the amount of data that has to be collected according to Van de Walle(2009). 77 Alternatively in chapters 7 and 8 we focus on a larger variety of outcome indicators without attempting to aggregate into a single welfare measure. This allows an analysis and understanding of the underlying linkages and potential heterogeneity across impacts. We included “intermediate” or “output” outcomes ─ such as access, distance, vehical operation (transport) costs savings and quality of road surface ─ and more “final” outcome indicators such as choice of relocation (chapter 8) that one expects to be affected by improved mobility/access. The post-project follow-up data collection in 2005 allowed for a sufficiently long time interval for impacts to emerge. Finally, detailed information on the project ─ including when and where the road works were implemented and what was done is provided in section 3.2.1 above in order to determine the EθFRθ‘s impact. Van de Walle(2009) finds that there is surprisingly little explicit discussion of the issue the road‟s zone of influence in recent attempts to implement rural road impact evaluations. Most studies typically test for impacts in well-defined communities such as villages serviced by an improved road or located in its vicinity. In chapters 4-6 the project area is defined as the district/constituency that the road passes through, whereas in chapter 7 it is defined as the village/primary sampling unit (PSU). This approach will not always be ideal, but we felt that it makes sense given the constraints faced. Alternatively in chapter 7 we also set a maximum distance of 5 km on either side of the road link and confine the search for impacts to this area.71 During our frield work in 2005 we used a Global Positioning System (GPS) receiver to help identify the spatial coordinates of the area desired. However, for the identification of the project area it was not feasible to list all communities and households located in the area so that a sample of PSUs and households within them could be randomly selected.72 71 Other studies typically use anywhere from 2 to 15 km. Van de Walle(2009) recommends that to ensure similarity as well as for practical and logistical reasons, a common strategy is to pick potential comparison areas in the vicinity of the treatment areas (though outside the catchment area of the project‘s impact). 72 78 Part Two Empirical Investigations of the Investment-led Rural Development Approach through Rural Transport Infrastructure Development in Zambia‟s Eastern Province 79 Chapter 4: The Impact of Rural Roads on Cash Crop Production in Zambia‟s Eastern Province between 1997 and 2002 80 “The challenge for us all is how to stimulate the small-scale farmers and private partnerships.” Peter Daka, Zambia‘s εinister of Agriculture, 2010.73 4.1. Introduction In Zambia‘s Eastern θrovince poverty is largely a rural phenomenon (chapter 5). Agriculture is an important part of the livelihoods of many poor people, and it is frequently argued that agricultural growth is a fundamental pre-requisite for widespread poverty reduction. θaradoxically, in Zambia agriculture‘s share of GDθ in terms of value added is higher today, than what it was at independence in 1964. This, together with increasing recognition of the diversity of poor rural people‘s livelihoods and the difficulties in ―getting agriculture moving‖ in areas where most poor people live today, has led to questions about the importance of agriculture for pro-poor rural growth, about the benefits of attempts to promote directly agricultural growth and development, and about the best means to promote such growth (Dorward et al., 2004c). Fafchamps, Teal and Toye(β001μ1γ) argue that ―while higher rates of growth achievable in export manufacturing may make it theoretically the best sector to support poverty reducing growth, in practice ―only a handful‖ of African countries will be able to achieve this, so that ―the 4η or so other African countries that do not become export platforms,‖ e.g. through ωhina‘s special economic zones (Brautigam, 2009) or the African Growth and Opportunity Act(AGOA) approved by the U.S. Congress in May β000, ―must rely on other engines of growthμ Agriculture, mining, tourism or a combination of them (quoted in Dorward et al., β004μ7ηf).‖ Despite the strong arguments for agriculture having provided the main engine of growth for rural poverty reduction in the past, reliance on pro-poor agricultural growth as the main weapon against rural poverty today may not be appropriate if the areas where today‘s rural poor are concentrated face severe difficulties in raising agricultural productivity or in accessing wider agricultural markets (Dorward et al., 2004c). 73 Monday, 24 May 2010 18:09 UK. BBC World Service, Lusaka, Zambia. 81 Although reforms in Zambia have led to promising signs of agricultural growth in recent years, Thurlow and Wobst(2005) argue that the persistence of poverty suggests that there remain significant constraints to poor Zambian households‘ participation in this growth process. One of the key constraints is market access created by poor rural infrastructure such that around 40 percent of agricultural households are still engaged solely in subsistence agriculture Transport infrastructure investments in rural areas are hypothesized to affect poverty through various channels. It increases agricultural productivity, which in turn directly increases farm incomes and helps reduce rural poverty (Fan, 2004). Thus, hypothetically speaking, improved provision of rural feeder roads should lead to lower transaction and farm production costs, while facilitating trading of cash crops and fostering long-run economic growth that contribute to the expansion of the economy of Eastern Province. Hence, we want to test the following hypothesis:74 The mean response in agricultural production and productivity growth to labour-based investment in rural roads within the treatment areas is the same as the mean response in the control areas. In the case of Zambia‘s Eastern θrovince, the main agricultural activities are cotton, tobacco, maize, vegetables, and groundnuts. Since, income earners in the rural areas are not going to benefit much from price increases alone, so quantity responses are going to be critical for poverty reduction (Balat et al., 2004). We use data from the Zambian Post-Harvey Surveys (PHS) covering all the districts of Eastern Province in the period from 1996/1997 to 2001/2002, allowing us to measure the short-term and medium-term gains from the EPFRP covering five districts in Eastern Province (Chadiza; Chipata; Lundazi; Katete; and Petauke districts), which was implemented during this period. These data are used to set up an empirical model of cash crop choice and cotton productivity. 74 In the words of Lofgren et al.(2004), spending on feeder roads in rural areas leads to the strongest reduction in national and rural poverty. 82 Our comparative inter-district analysis build on two studies Brambilla and Porto(2005) and Brambilla and Porto(2007) which propose a dynamic approach accounting for entry and exit into the agricultural cotton sector to avoid biases in the estimates of aggregate productivity, when measuring productivity in agriculture and which can be applied to our observational data for Zambia. The dataset is constituted by a repeated annual (i.e. equal spaced) sequence of independent cross-sections of farm-level data based on a large random sample of the population. The objective of this chapter is to quantify the direct and indirect rural transport infrastructure investment impacts of the EPFRP. Although, the estimation of this supply response has proved difficult in the preceding literature, we will nevertheless explore the impacts on the production of the main cash crop in Zambia‘s Eastern θrovince. The aim is to estimate whether the differential cotton yield generated by increased market agricultural activities mainly is due to the EPFRP treatment, with rural roads understood as a non-traditional production factor input in the production function.75 The remainder of this chapter is organized as follows. Section 2 offers a discussion of agricultural cash and food crop production trends in Zambia‘s Eastern θrovince. Section 3 presents the equations to be estimated, and discusses the way in which the impact of the EPFRP on productivity is identified. Section 4 briefly presents the PHS data covering the period from 1996/1997 to 2001/2002. Section 5 presents empirical results for cotton productivity along with a number of robustness checks. Finally, section 6 concludes and discusses a few policy implications of the empirical findings. 75 Only a total of 34,329 worker days were generated in Mambwe by Rehabilitation works which is less than 20% of the average workers days of the catchment districts. Moreover no workers days were created by Maintenance Road Works, therefore Mambwe is categorised as a control district. 83 4.2. Background and Setting This recent positive trend in agriculture is confirmed by national crop production data, which shows a slight upward trend between 1996 and 2003 for e.g.: Barley; cassava; groundnuts; seed cotton; and tobacco, whereas maize; millet and sunflower seed had decreased. Figure 4.1 shows the changing levels of yield (Hg/Ha) for the main food and cash crops in Zambia. National maize yield fell dramatically both in absolute terms and relative to other crops in the latter part of the 1990s after which the maize yield incrementally converged towards its earlier level. The fluctuations were driven both by shifting area size devoted to harvesting maize as well as production levels. The national yield of seed cotton almost experienced a reversed trend, in the sense that the yield increased significantly towards 1997/98 after which it gradually declined until 2005. As seen from figure 4.1 there was a wealth of diverging growth experiences amongst the other non-maize crops, some of which such as groundnuts and tobacco have performed well over the decade, whereas the yield of millet and sunflower seed declined. However, despite its declining importance the more-drought susceptible crops maize has remained one of the dominant staple crops in Zambia together with cassava. Figure 4.1: Yield of Selected Cash Crops in Zambia, 1996-2005 25000 20000 Yield (Hg/Ha) Barley Groundnuts, with shell 15000 Maize Millet 10000 Seed cotton Sunflower seed 5000 Tobacco, unmanufactured 0 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 Source: Author's calculation based on Food and Agriculture Organization (FAOSTAT, 2009). Note: Figure does not show floricultural production, which has been one of the fastest growing cash crops in recent years (World Bank, 2004). 84 Apart from changes in the level of crop production, there have also been substantial changes in its composition. Much of this has been driven by the agricultural policies that were implemented by the new MMD government (Smale & Jayne, 2002; Pletcher, 2000). Cotton is one of the key agricultural activities in rural Zambia and the cotton sector has been a success story in Zambia since a process of liberalization in cotton production and marketing began in 1994 (Balat and Porto, 2005b). Significant percentages of cotton farmers (due to soil characteristics) are observed only in the Southern Province, Mumbwa in Central Province, and the Eastern province, where it is the most relevant cash crop activity. In these three cotton-growing provinces, a large share of the cash income of rural farmers comes from the sale of cotton seeds. Eastern Province is the most important area for cotton production and it share of Zambia‘s total output increased from 1η% in 1994 to βγ% in 1997 (table 4.1).76 Table 4.1: Share of Cotton Area in Total Cropped Area, 1993-1998 Harvest Year 1993 1994 1995 1996 1997 1998 Province (%) Central Eastern Southern Zambia 9% 6% 3% 3% 7% 5% 5% 3% 10% 7% 1% 4% 9% 14% 4% 6% 16% 15% 6% 6% 13% 17% 7% 7% Province (km2) Central Eastern Southern 8495,55 4146,36 2558,49 6607,65 3455,3 4264,15 9439,5 4837,42 852,83 8495,55 9674,84 3411,32 15103,2 10365,9 5116,98 12271,35 11748,02 5969,81 Zambia 22578,48 22578,48 30104,64 45156,96 45156,96 52683,12 Source: Zambia Food Security Research Project, 2000 based upon Post-Harvest Surveys, Ministry of Agriculture, Food & Fisheries, Database Management Unit, Central Statistical Office. From table 4.1 in the agricultural season 1995/1996 we notice a 100 per cent jump in the share of cotton area in Eastern Province. However, it was the year where the outgrower programmes were offered by two private companies Lonrho Cotton and Clark Cotton to provide participating small-to medium scale farmers with inputs and extension services on loan (chapter 8). From 1977 to 1994 the Lint Company of Zambia (Lintco) had a monopsony in seed cotton markets, and a monopoly in inputs sales and credit loans to farmers.77 The initial reforms of the mid-1990s eliminated most of these interventions and markets were 76 As much as 45% in 1996 if we only include these three provinces in the total area sum. Cotton was not a significant crop in Zambia until the late 1970s when the Lint Company of Zambia (Lintco) was set up by the GoRZ to develop cotton in Zambia. 77 85 liberalized. The new MMD government decided in 1995 to privatize and sell Lintco to Clark Cotton and Lonrho Cotton.78 However, it did not succeed in introducing much competition in the sector. This is because the three large-scale private firms segmented the market geographically. In consequence, liberalization in 1996 gave rise to geographical monopsonies (i.e. the initial phase of regional private monopsonies) rather than national oligopsonies (Balat, 2005; Brambilla, 2005; Balat and Porto, 2006). In particular, instead of the localized monopsonies, entrants and incumbents started competing in cotton trade in many districts of Eastern Province, which was dominated by two businesses Clark Cotton, a South African firm which took over the Chipata Ginnery, and Sable Limited a completely new entrant into the market, which diversified into cotton trading from other trading activities in 1992/93 (Chiwele et al., 1996, Poulton et al., 2004). In addition, some entrants that were not using out-grower schemes started offering higher prices for cotton seeds to farmers who had already signed contracts with other firms. This caused repayment problems and increased the rate of loan defaults (Balat and Porto, 2005a, Brambilla and Porto, 2005). Table 4.2: Percentage of Farmers Growing Cotton in Eastern Province, 1997 – 2002 District Chadiza (301) Chipata (303) Katete (304) Lundazi (305) Petauke (308) Total Catchment Districts Chama (302) Mambwe (306) Nyimba (307) Total Control Districts Total 1996/97 44,79% 40,92% 50,51% 24,11% 24,34% 35,48% 18,92% 59,62% 6,25% 29,93% 34,86% 1997/98 27,27% 33,90% 53,03% 25,78% 16,79% 30,99% 33,33% 61,82% 8,11% 38,28% 31,75% 1998/99 11,24% 34,54% 35,18% 43,23% 20,30% 31,04% 30,26% 50,00% 11,32% 28,22% 30,68% 1999/2000 10,00% 25,74% 39,55% 11,92% 8,44% 19,55% 11,25% 49,15% 6,67% 21,11% 19,76% 2000/2001 26,14% 36,16% 52,17% 27,04% 21,37% 32,50% 10,00% 54,90% 22,22% 26,86% 31,71% 2001/2002 27,00% 36,97% 52,36% 24,52% 21,31% 32,20% 16,88% 54,24% 23,73% 30,26% 31,93% Sourceμ Authors‘ calculations based on ωSη‘s θost Harvest Surveys 1997-2002. On top of all this, world prices in the 1998/99 agricultural season began to decline, and farm-gate prices declined as a result. After many years of high farm-gate prices, and with limited information on world market conditions, felt that out-growers‘ contracts 78 Before taking over the operations of Lintco in Mumbwa through privatization, Lonrho went into cotton trading in the 1992/93 marketing season following liberalization and traded alongside Lintco for some-time (Chiwele et al., 1998). 86 were being breached, and the participation rates in the whole Eastern province declined from 32% percent in 1997/1998 to 20% in 1999/2000 (table 4.2). From 2000/2001 to 2004/2005 the percentage of cotton growers returned to the previous level of more than 30% (table 4.2) and is correlated with entry into cotton. In 2004, there were three companies based in Eastern θrovince‘s capital ωhipataμ ωlark Cotton Zambia Ltd; Dunavent Zambia Ltd. Cotton;79 and Zambia-China Mulungushi Textiles Joint Venture (ZCMT). Table 4.3: Fraction of Land Allocated to Cotton, 1997 - 2002 District 1996/97 1997/98 1998/99 1999/2000 2000/2001 2001/2002 Chadiza (301) 19,80% 10,76% 8,77% 5,09% 15,38% 17,12% Chipata (303) 24,97% 21,03% 18,81% 14,24% 12,66% 13,07% Katete (304) 21,19% 26,77% 20,02% 14,79% 18,24% 19,14% Lundazi (305) 11,96% 15,38% 23,67% 6,77% 8,71% 8,89% Petauke (308) 7,41% 9,66% 9,19% 3,23% 8,39% 7,35% 17,07% 17,02% 9,77% 11,50% 11,50% Total Catchment Districts 16,37% Chama (302) 8,38% 15,55% 14,66% 3,59% 3,48% 6,01% Mambwe (306) 33,45% 21,22% 30,74% 23,89% 18,13% 22,21% Nyimba (307) 1,56% 1,60% 7,87% 4,84% 9,72% 9,87% Total Control Districts 21,84% 13,97% 14,54% 9,99% 13,73% 16,37% Total 17,06% 16,67% 16,69% 9,79% 12,76% 12,16% Source: Authors‘ calculations based on the Post Harvest Surveys 1996/1997-2001/2002. Cotton farming was also affected by a decline of the land area devoted to cotton in 1998-1999 (when the outgrowing scheme was failing). This, however, was followed by a significant increase in area planted in 2000-2001, when the out-grower scheme was perfected with the entrance of the private company Dunavant, which in 1999 had initiated a distributor system (Tschirley et al., 2006; Poulton et al., 2004).80 The fraction of land allocated to cotton by the average farmer in the Eastern province‘s catchment districts plummeted from 17% in 1998/99 to less than 10% in 1999/2000, and subsequently stayed below the percentages reached in the control districts (table 4.3). On the other hand, the average cotton yield of the catchment districts not only increased continuously from 1996/1997 to 1999/2000, but from 1998/1999 when the 79 Its predecessor was Lonrho Cotton. Partly as a result of the failure of the outgrower scheme, Lonrho announced its sale in 1999 and Dunavant Zambia Limited entered the market. 80 87 EPFRP was implemented exceeded mean of the control districts until the trend changed in 2000/2001, where the overall Eastern Province cotton productivity dropped to less than 1 MT/HA from 1,64MT in the previous season equivalent to a percentage fall of more than 40 percent (table 4.4). Table 4.4: Yields per Hectare in Cotton (MT/HA), 1997 - 2002 District Chadiza (301) Chipata (303) Katete (304) Lundazi (305) Petauke (308) Total Catchment Districts Sub-total Observations Chama (302) Mambwe (306) Nyimba (307) Total Control Districts Sub-total Observations Total Eastern Province Total Observations 1996/97 0,739 1,935 1,069 0,925 1,137 1,304 380 0,988 1,841 0,271 1,580 41 1,331 421 1997/98 2,154 1,316 1,648 1,219 1,430 1,476 329 1,452 1,651 0,651 1,541 49 1,485 378 1998/99 1,869 1,848 2,241 1,204 1,507 1,689 337 1,320 1,056 0,774 1,151 46 1,625 383 1999/2000 1,477 1,470 1,817 1,078 2,686 1,677 242 1,863 1,358 0,434 1,381 37 1,638 279 2000/2001 0,777 0,870 1,103 0,783 1,065 0,943 397 1,016 1,164 1,081 1,105 70 0,967 467 2001/2002 0,770 0,869 1,099 0,773 1,090 0,944 417 1,017 1,130 1,081 1,088 75 0,966 492 Source: Authors‘ calculations based on the Post Harvest Surveys 1996/1997-2001/2002. Evidently these results begs the question as to why rural roads improvements are considered an important explanatory variable of the Yields per Hectare in Cotton given the higher average yield in the control districts. The most important indicator is the fact that the average cotton yield in the catchment districts exceeded the mean yield in the control districts exactly in the period from 1998 to 2000 when most of the feeder roads in the treatment zones had just been rehabilitated. Hence, in the first instant these descriptive statistics do seem to have some bearing on the question. 88 4.3. Framework The linkage between rural transport infrastructure and economic development can be identified can be expressed using economic development measures such as cash crop productivity or production.81 In other words, transport infrastructure improvements which influence travel behaviour and transport markets (chapter 8) must eventually be transferred into these measurable economic benefits, which also include improved factor productivity, increased demand for inputs, and greater demand for consumer goods. Moreover, the degree to which infrastructure improvements will affect economic development is not independent of the level and performance of the in-place capital infrastructure (Banister and Berechman, 2000). The impact of a transport infrastructure project on a regional economy varies depending on the phase of the project, because the interrelationships are not instantaneous and, in general, require considerable periods of time to transpire. Transportation spending for maintenance and rehabilitation of rural feeder roads affects current economic activity but also represents an investment in future growth. The main reasons for this are the long period necessary for investment implementation (1998-2001) as well as the time needed for the demand side adjustment. 82 The longer-term effect fosters economic growth that contributes to the expansion of a regional economy (New York State Department of Transportation, 2000).83 Underlying these time lags are market imperfections including incomplete information concerning infrastructure development, uncertainty regarding the behaviour of public authorities and private entities, high transaction costs emanating from imperfect land market and general market externalities (see e.g. Dorward et al., 1998; Kydd et al., 2003). All of these make the transformation of transport improvements into economic benefits highly time dependent. The overall result is a dynamic process whose evolution 81 In our context agricultural productivity is defined as output per hectare. As the effects of a transport project reverberate through the economy, increasing income levels, consumer spending, etc., government coffers will increase, allowing for an expansion and / or improvement of public services. 83 Cost related indirect economic benefits of transportation investment do not materialize instantaneously because they involve long-term business and household location decisions. In fact, a prevalent view is that economic effects are realized after lags between 4 and 7 years in the case of highway developments. 82 89 depends on the initial conditions of local transport and activity systems and on the local transport and economic policies (Banister and Berechman, 2000).84 The next two sub-sections present the equations to be estimated as well as the identification strategy for agricultural productivity. 4.3.1. Agricultural Productivity: Estimation Strategy The counterfactual framework is where each individual has an outcome with and without treatment. This approach allows us to define various treatment effects and it was pioneered by Rubin(1974) and since adopted by Rosenbaum and Rubin(1983); Heckman(1992, 1997); Imbens and Angrist(1994), Angrist et al.,(1996); Heckman et al., (1997b); Angrist(1998) and (Wooldridge, 2002). 4.3.1.1. A Counterfactual Setting A cause is viewed as a manipulation or treatment that brings about a change in the variable of interest, compared to some baseline, called the control. The basic problem in identifying a causal effect is that the variable of interest is observed either under the treatment or control regimes, but never both (Dehejia and Wahba, 2002). Framework We want to evaluate the causal effect of the binary rural transport infrastructure (RTI = EPFRP) treatment variable on a continuous „logarithm of cotton productivity (or production)‟ outcome Y experienced by units in the population of interest. For individual i, i = 1, . . . , N, with all units exchangeable, let (Y0i, Y1i) denote the two potential outcomes, i.e.: Y1i the outcome of unit i if i were exposed to the treatment: Di = 1. Y0i the outcome of unit i if i were not exposed to the treatment: Di = 0, where Di ϵ {0, 1} the indicator of the treatment actually received by unit i. Yi Y01 = X i + Di(Y1i – Y0i)the actually observed outcome of unit i. the set of pre-treatment characteristics. = Y1i – Y0i the causal (treatment) effect for a single unit i. 84 There is an alleged complementarity between transport and telecommunication technologies. The ability to use telecommunications (e.g. Agricultural Extension Services through radio programmes) may affect travel needs of the agricultural extension service officers. 90 The fundamental problem of causal inference‘ is that it is impossible to observe the individual treatment effect. It is impossible to make causal inference without making generally untestable assumptions (Sianesi, 2001, Abadie et al., 2004, Dehejia and Wahba, 2002). Under some assumptions the average treatment effect of the sampled Eastern Province agricultural household population can be estimated by:85  Average Treatment Effect (ATE) = 1 N  (Y1i  Y0i ) N i 1 = E(y1 – y0);  Average Treatment Effect on the Untreated (ATU) = E(y1 – y0 | D = 0);  Average Treatment Effect for the sub-population of the Treated (ATT) = E(y1 – y0 | D = 1). The primary treatment effect of interest in non-experimental settings is the expected treatment effect for the treated population; hence |D=1 = E(Y1i – Y0i | D = 1) = E(Y1i| D = 1) – E(Y0i| D = 1) = (4.1a) (4.1b) |D=0 = E(Y1i – Y0i | D = 0) = E(Y1i| D = 0) – E(Y0i| D = 0) = 1  (Y1i  Y0i ) , N1 i| Di 1 1  (Y1i  Y0i ) , N 0 i| Di  0 where N1 = ∑iDi and N0 = ∑i(1 – Di) are the number of treated and control units respectively (Sianesi, 2001; Abadie et al., 2001). The problem of unobservability is summarized by the fact that we can estimate E(Y1i| D = 1) but not E(Y0i| D = 1). Thus, we need to construct the counterfactual E(Y0i| D = 1) the outcome participants would have experienced, on average, had they not participated. The difference, (4.2) e = E(Y1i | D = 1) – E(Y0i | D = 1), can be estimated, but it is potentially a biased estimator of the difference in the outcomes with and without treatment, . Intuitively, if Y0i for the treated and comparison units systematically differ, then in observing only Yi0 for the comparison group we do not 85 Whether one is interested in the average treatment effect in the population (PATE) or the sample (SATE) does not affect the choice of estimator: the sample matching estimator will estimate both. However, in general the variance for SATE is smaller than for the PATE (Abadie et al., 2001; cf. Imbens, 2002, 2003). 91 correctly estimate Yi0 for the treated group. The role of randomization is to prevent such bias (Dehejia and Wahba, 2002).86 Thus, in our observational study of the EθFRθ‘s impact on cotton productivity (logyield), by definition there are no experimental controls. Therefore, there is no direct counterpart of the ATE. In other words, the counterfactual is not identified. As a substitute we may obtain data from a set of potential comparison units that are not necessarily drawn from the same population as the treated units, but for whom the observable characteristics, x, match those of the treated units up to some selected degree of closeness. The average outcome for the untreated matched group identifies the mean counterfactual outcome for the treated group in the absence of the treatment. This approach solves the evaluation problem by assuming that selection is unrelated to the untreated outcome, conditional on x (Cameron and Trivedi, 2005). Propensity Score Matching Rosenbaum and Rubin (1983, 1985) suggest the use of the propensity score—the probability of receiving treatment conditional on covariates—to reduce the dimensionality of the matching problem, by allowing us to condition on a scalar variable rather than in a general n-space (Dehejia and Wahba, 2002).87 The concept of propensity score is a conditional probability measure of treatment participation given x and is denoted p(x) (i.e. the probability of unit i having been assigned to treatment),88 where (4.3) p(x) = Pr{D = 1 | X = x} = E(Di|Xi) An assumption that plays an important role in treatment evaluation is the balancing condition, which states that (4.4) D ┴ x | p(x). 86 In a non-experimental setting, the treatment and comparison samples are either drawn from distinct groups or are nonrandom samples from a common population. In contrast, in a randomized experiment, the treatment and control samples are randomly drawn from the same population, and thus the treatment effect for the treated group is identical to the treatment effect for the untreated group (Dehejia and Wahba, 2002). 87 As the number of variables increases, the number of cells increases exponentially, increasing the difficulty of finding exact matches for each of the treated units. 88 Estimate the propensity score on the X‘s e.g. via probit or logit. 92 Type of Matching Estimators Matching on the propensity score is essentially a weighting scheme, which determines what weights are placed on comparison units when computing the estimated treatment effect: ˆ |D 1  (4.5) 1 N   Y  J Y  ,  iN   1 i i j J i j  where N is the treatment group, |N| the number of units in the treatment group, Ji is the set of comparison units matched to treatment unit i (see Heckman et al., 1998), and |Ji| is the number of comparison units in J1i. Expectations are replaced by sample means, and we condition on p(Xi) by matching each treatment unit i to a set of comparison units, Ji, with a similar propensity score. Our objective is to match treated units to comparison units whose propensity scores are sufficiently close to consider the conditioning on p(Xi) in the following proposition: |D=1 = Ep(X)[( |D=1, p(X))|Di = 1], (4.6) to be approximately valid (Dehejia and Wahba, 2002, Becker and Ichino, 2002). Three issues arise in implementing matching: Whether or not to match with replacement, how many comparison units to match to each treated unit, and finally which matching method to choose (ibid). The unit level treatment effect is Y1i – Y0i. However, only one of the potential outcomes Y1i or Y0i is observed for each individual and the other is unobserved or missing. The matching estimators we consider imputing the missing potential outcome by using average outcomes for individuals with ―similar‖ values for the covariates. Pair to each treated individual i some group of ‗comparable‘ non-treated individuals and then associate to the outcome of the treated individual i, yi, the (weighted) outcomes of his ‗neighbours‘ j in the comparison group: (4.7) yˆi  w y ij jC 0 ( p i ) j Where: 93 C0(pi) is the set of neighbours of treated i in the control group wij ϵ [0, 1] with w ij jC 0 ( p i ) 1 is the weight on control j in forming a comparison with treated i. We associate to the outcome yi of treated unit i a ‗matched‘ outcome given by the  j :| p  p | min {| p  p |}, outcome of the most observably similar control unit (‗traditional matching estimators‘) one-to-one matching: (4.8) C0(pi) = i j k { D  0} i k wik = 1(k=j). A weighted average of the outcomes of more (possibly all) non-treated units where the weight given to non-treated unit j is in proportion to the closeness of the observables of i and j (‗smoothed weighted matching estimators‘) kernel-based matching (4.9) C0(pi) = {D = 0} wij  p  pj   ∞ K  i  h  (for Gaussian kernel) (Sianesi, 2001).89 4.3.1.2. Estimating Model of Cotton Productivity In our second approach, our regression analysis seek to estimate whether the agricultural productivity (or production) outcomes are better in the treatment districts, where the existing feeder road network was partially improved, with similar outcomes from “control” districts where the EθFRθ wasn‘t implemented while recognizing the difficulty of ―controlling‖ for the myriad of other factors impacting on the comparative economic performance of these eight districts in Zambia‘s Eastern Province.90 We estimate a simple model of cash crop productivity. This is done by modifying the empirical model in Brambilla and Porto(2005, 2007), in which the dependent variable agricultural productivity (i.e. yield of cash crop) is defined in physical units (i.e. quantity per hectare of cultivated land). Let yhtc denote the volume of cotton production (in Metric 89 Non-negative; symmetric and unimodal. Thus, if the case study districts in which the EPFRP has been implemented can be shown to have experienced faster agricultural growth than would have been predicted on the basis of trends in the wider regional economy or with the three control group districts (Chadiza; Chama; and Mambwe), then this ―additional‖ growth may be deemed to be due to the EθFRθ (DTZ, 2004). 90 94 Tonnes) per hectare produced by household h in period t. The log of output per hectare is given by: (4.10) ln yhtc = xhtc c  It  ht  b0ht   htc  1RTI t Here, xhtc is a vector of household determinants of cotton yields including the age (i.e. experience) and sex of the household head, the size of the household, household demographics (i.e. share of male household members), input use (i.e. fertilisers), assets (i.e. livestocks), the size of the land allocated to cotton, farm size (i.e. stratum), and district dummies.91 The rural roads dummy variable RTIt captures whether the district has been „treated‟ by rural transport infrastructure improvements (i.e. this indicates the presence of the EPFRP) in the period from 1998/1999 to 2001/2002. Thus, we model the productivity effects of the EPFRP with one dummy variables (or alternatively the percentage share of the district feeder roads, which have been treated), RTIt. The impacts of these EPFRP treatments are measured relatively to the excluded category, which is the introductory phase 1996-1998 (with RTIt = 0) and the three control districts in the subsequent implementation phase of 1998-2002. Following Brambilla and Porto(2005, 2007) the model includes a number of fixed effects, such as districts effects (included in x), year effects, It, and idiosyncratic household level fixed effects ht and ht and  ht . The district effects include market access, local infrastructure (road density), local knowledge and access to credit; they are controlled for with district dummies. The year effects, It, capture aggregate agricultural effects and other shocks that are common to all farmers in a given period t. In equation (4.10), these effects cannot be separately identified from the infrastructure provision dummy RTI. To deal with this, Brambilla and Porto(2005) propose to model productivity in other crops (mainly maize) to difference out time varying factors that affect productivity in agriculture. 91 It is a production function, not a supply function, since prices are not included in 95 xhtc The household level fixed effect has two components: A farm effect, , and a cotton-specific effect, . The farm effect  captures all idiosyncratic factors affecting general agricultural productivity in farm h that are not observed by the econometrician and are thus not included in x. It includes soil quality, de jure (i.e., titles) and de factor land rights (Bellemare, 2009, Goldstein and Udry, 2008), know-how, and other factors that affect productivity in all crops. The cotton-(idiosynctratic) specific effect  is a combination of unobserved factors that affect productivity in cotton, including ability and expertise in cotton husbandry and suitability of the land for cotton (Brambilla and Porto, 2005). Although, cotton in Zambia‘s Eastern θrovince is intercropped with groundnut, we assume that the farmer‘s cotton characteristics are somewhat different than the farmer‘s characteristics in x e.g. given the different cotton cultivation practices (i.e. sowing, ploughing, harrowing, etc.) and marketing practices. According to Brambilla and Porto(2005) there are two problems with the household fixed effects. First, both  and  are unobserved by the econometrician but observed by the farmer when making decisions on input and on land allocation to different crops. Hence, some of the variables included in x may be correlated with these unobservables. In addition, entry and exit into cotton farming depend on these unobservables as well since farmers‘ decisions on land allocation to different crops may be based on  and . More importantly for our purposes, this entry/exit component affects the estimates of the RTI dummy by altering the composition of farmers that produce cotton in each time period. 4.3.2. Agricultural Productivity: Identification Strategy We use the random sample statistics from these target areas, which the CSO collected in the six year period from 1996/1997 to 2001/2002. This pseudo-panel dataset ideally should present us with an opportunity to use panel data analysis to test which factors that determine the variation of the productivity of cash crops. A panel data set 96 would allow us to account for both idiosyncratic effects.92 However, the post-harvest survey (PHS) dataset is a repeated independent cross section of farmers, which makes it impossible to track the same household over time as required in a genuine panel, because the sample design does not attempt to retain the same units in the sample (Baltagi, 2001). We thus need additional modelling to deal with the fixed effects. Brambilla and Porto(2005) propose to model agricultural productivity in maize to control for  (and the year effects, It) and to model the share of land devoted to cotton to control for . Another method to overcome the problem of the lack of panel data is by creating a pseudo-panel at a geographical scale by aggregation. This has proven to be quite useful for estimating structural relationships (Glewwe and Jacoby, 2000) to capture the short-to medium- run effects.93 Since the pioneering papers by Kuh (1959), Mundlak (1961), Hoch (1962), and Balestra and Nerlove (1966), the pooling of cross-section and time series data has become a popular way of quantifying economic relationships. Each series provides information lacking in the other, so a combination of both leads to more accurate and reliable results than would be achievable by one type of series alone (Laszlo, 1996). In this method groups of "like" households are created and changes in their income over time are analysed.94 The advantages of this method is that it allows us to make statements about changes that occur to different types of similar households over time but it involves loss of information on the variation within "like" groups (McCulloch et al., 2001).95 It is highly likely that the factors that attract better roads in certain areas also affect the agricultural productivity outcomes. Unless the comparison areas – the counterfactual – have the same factors, it will leave biased estimates. Selection bias occurs if for some The unobservables  and  are indexed by ht because, given the cross-sectional nature of the data, the unit of observation is a household-time period combination. However, if the data were a panel,  and  would be indexed by h only (Brambilla and Porto, 2006). 93 Banister & Berechman consider 10 years as the time it takes for land use and travel markets to converge to a state of equilibrium following an external change. Thus, medium to long terms effects are to be over 10 years. Bourguignon, Ferreira, and Lustig (2001) recommend at least a ten-year interval. 94 The method is adopted by cohort studies where individuals are grouped by age (or other attributes) and the cohort is compared with other cohorts over time (chapter 5). 95 Other possible partitions include: The strata used by the sampling frame; and main agricultural output (rural areas). 92 97 reason roads are poor in participating area and being compared with places that don‘t have these factors. The identification strategy relies on a modified difference-in-differences (DD) approach to get rid of endogeneity (Chapter 3). First, again following Brambilla and Porto(2005) we take differences of outcomes (i.e., productivity or production) across the different phases. Second, we use maize productivity to difference out unobserved household and aggregate agricultural year effects. Finally, since more productive cash crop farmers are also more likely to allocate a larger fraction of their land to cash crop production, we use cash crop shares, purged of observed covariates, as a proxy for unobserved cash crop-idiosyncratic productivity. Failure to adequately control for time-varying initial conditions that lead to the road placement can lead to very large biases in estimates of impacts (van de Walle, 2008, 2009). Yields per hectare in maize, yhtm , are given by: (4.11) ln yhtm = xhtm ' m  I t  ht   htm , Maize productivity depends on covariates, xhtm , including district effects, the agricultural year effects, It, and the farm effects  ht . By taking difference, we get (4.12) ln yht  ln( yhtc / yhtm )  ln( yhtc )  ln( yhtm )  x'ht   1RTI t  b0ht   ht . Here, the observed household covariates xht included in the estimation are based on the same determinants of productivity as mentioned above. The district EPFRP dummy capture local market access effects, we therefore allow marketing conditions to affect cotton (a cash crop activity) and maize (a mostly subsistence crop) differently. The coefficient α1 measure the impact of the implementation phase of the EPFRP on cotton productivity. There are two important identification assumptions. First, we assume that the agricultural effects, It, have the same effect on growth of cotton and maize output per hectare. This according to Brambilla and Porto(2005) means that we can use the trend in 98 maize productivity to predict the counterfactual productivity in cotton in the absence of the EPFRP. The assumption implies that we could use productivity in other crops to difference out the agricultural effects. Under the maintained hypothesis, the trend in maize productivity and the trend in the productivity of other crops should be similar. In the regression analysis, we use maize as control because, as opposed to the other crops, virtually all households produce it (table 4.5). Table 4.5: Percentage of Households that Grow Maize, 1997 – 2002 District Chadiza (301) Chipata (303) Katete (304) Lundazi (305) Petauke (308) Total Catchment Districts Chama (302) Mambwe (306) Nyimba (307) Total Control Districts Total Eastern Province 1996/97 100,00% 99,34% 99,49% 99,55% 99,25% 99,45% 100,00% 100,00% 100,00% 100,00% 99,51% 1997/98 100,00% 96,95% 98,48% 95,56% 98,09% 97,47% 94,44% 90,91% 100,00% 94,53% 97,16% 1998/99 98,88% 95,39% 98,49% 95,63% 97,42% 96,79% 97,37% 97,06% 100,00% 98,16% 96,97% 1999/2000 2000/2001 99,00% 97,73% 99,11% 99,35% 98,64% 98,91% 94,62% 94,85% 98,44% 99,24% 97,90% 98,14% 97,50% 100,00% 94,92% 98,04% 98,33% 100,00% 96,98% 99,43% 97,77% 98,32% 2001/2002 97,00% 97,88% 98,58% 96,55% 99,34% 98,01% 100,00% 98,31% 100,00% 99,49% 98,22% Sourceμ Author‘s calculations based upon θHS 199θ/1997 – 2001/2002. The second critical assumption of our difference-in-difference model is that the EθFRθ implementation didn‘t affect maize productivity. ψy including measures of labour, fertilizers and land allocation in the observed covariates x of the regression, these direct and indirect effects will be accounted for. There is the possibility that the EPFRP affected maize productivity and that α1 is a measure of the impacts of the EPFRP on cotton productivity relative to maize productivity. This possibility is ruled out from the plot (figure A2a) from where it can be seen that the parallel trends in maize productivity holds in our case between those districts that were affected versus those districts that were not affected by the EPFRP. Overall, this indicated that the differencing will identify the impact of the EPFRP on cotton productivity only. The heterogeneity of the suitability of the land for cotton production and know-how of cotton husbandry leads to different entry-exit decisions regarding cotton production, which alters the composition of the group of farmers that produce cotton in each of the EPFRP phases. The estimates of the changes in productivity at the aggregate level 99 comprise both the changes in productivity at the farm level and the changes in the composition of the farmers that produce cotton in each time period consistent estimation of the changes in productivity at the farm level requires that we control for entry and exit (Brambilla and Porto, 2005). If there are fixed costs in cotton production, then cotton will only be profitable if productivity is high enough. This means that there is a cut-off (which depends on prices, market conditions, infrastructure) such that farmers with productivity above this cut-off will enter the market and farmers below the cut-off will not enter (or exit, if they were in the market already). In consequence, measures of productivity that do not control for these dynamic effects may be artificially high (thus leading to downward biases in the estimates of productivity declines) (Brambilla and Porto, 2005). Brambilla and Porto(2005) extend the Industrial Productivity Analysis literature by developing a method to deal with entry and exit in the estimation of agricultural production functions and crop choices. Furthermore, whereas this literature relies on longitudinal surveys, our method in line with that used by Brambilla and Porto(2005) can be used in repeated cross-sections. ψrambilla and θorto‘s solution to this problem is to construct proxies for the unobserved productivity parameter. The method exploits the idea that since households with high unobserved cotton-specific effects – ϕht – (see above) are more productive in cotton, they are also more likely to devote a larger share of their land to cotton production. This means that we could use land cotton shares as a proxy for the unobservable ϕht in (4.12). In practice, consistent estimation requires that we purge these shares of the part explained by observed determinants of cotton choice.96 Let ahtc be the fraction of land allocated to cotton. A general model of these shares is Notice that omitting ϕ not only leads to inconsistencies because of the entry-exit effects, but also may induce correlation between some variables in the vector x and the error term in the difference-in-differences model. For example, the choice of inputs, such as labour or fertiliser use, will depend on ϕ (so that higher levels of unobserved productivity may be positively correlated with input use). The model in (3.14) takes care of these biases (ibid.). 96 100 (4.13) ahtc = mt(zht, ϕht), where z is a vector of regressors which includes EPFRP‟s district effects that affect selection into cotton production. We use the district EPFRP dummy to capture access to market and local feeder road network that facilitates farmer participation in market cash agriculture. The function m allows regressors z and unobservables ϕ to affect the shares a non-linearly. We begin by considering the simplest model with a linear functional form (4.14) ahtc = z‘ht t + ϕht, Estimation of (4.14) is straightforward, except for the fact that the share of land devoted to cotton is censored at zero. This means that OLS may be inconsistent. A simple solution is to implement a Tobit procedure. More generally, we explore a more semiparametric estimation of (4.14) by using a censored least absolute deviation (CLAD) model. We note that provided the right specification for the model is used, consistency follows because the regressors z are exogenous to ϕ. This requires that family composition or farm size does not depend on unobservables such as cotton-specific ability or land quality. Importantly, since we use data on all households to estimate (4.14), this equation does not suffer from a selection problem like the one we are attempting to control for in the productivity model (Brambilla and Porto, 2005). The allocation of land to cotton depends on several factors that we need to account for. In particular, the selection into cotton depends on the EPFRP. This means that we should include RTI in (4.14). Cotton choices depend on output and input prices, too. Unfortunately, we do not have information on prices at the farm level. To the extent that prices vary by time, or by district, however, we can account for them with year or district dummies. In practice, we estimate a different model like (4.14) in each of the six years from 1996/1997 to 2001/2002 (notice that t is indexed by t in (4.14)). This means that we will not be able to separate the effects of the implementation of the EPFRP from the effects of changes in international prices on land allocation, but we will be able to control effectively for ϕ in the productivity model. 101 Finally, we note that identification of ϕ requires that the selection into cotton is affected by the same unobservables that affect cotton productivity. In principle, it would be possible to argue that there are additional unobservable factors that affect the selection into cotton. Plugging in the estimates of ϕ in (4.12), the productivity model is: (4.15) ln yht = x‘ht + α1RTIht + b0  ht  ~ht  This modified difference-in-differences approach is consistent with entry and exit into cotton farming according to Brambilla and Porto(2005). 102 4.4. Data Agriculture censuses and farm structure surveys are useful sources for rural development analysis. While the former is carried out every 10 years in Zambia, the latter is done annually. Agriculture censuses have the same advantage as the population censuses in that they provide exhaustive results with detailed territorial breakdown (Karlsson and Berkeley, 2005). The Agricultural statistical system in Zambia has been producing both structural97 and performance data.98 After the census of 1971/72, CSO extended the surveys to cover the subsistence or smallholder sub-sector of agriculture. In 1985/86 the two types of surveys were renamed the Crop Forecasting Survey (CFS) and Post- Harvest Survey (PHS), respectively.99 These surveys are conducted in an integrated manner and as the core of the National Household Survey Capability Programme (NHSCP), which has been implemented since 1983. However, The Agriculture and Environment Department of Zambia‘s ωSη only have agricultural production data at the district level going back until 1995. 4.4.1. The 1990 & 2000 Censuses of Population, Housing and Agriculture The 1990 Census of Population and Housing provide such information as was required to create a sampling frame for inter-censal agricultural surveys.100 At the time of the 1990 census there were 57 districts in the country. The Census Supervisory Areas (CSAs) were demarcated within each district while the Standard Enumeration Areas (SEAs) were demarcated within a CSA. A geo-coding system was, hence, developed with each SEA having a unique 6-digit code (Kasali, 2002).101 The sampling frame comprised 4,193 CSAs out of which 3,231 are rural and 962 are urban. Each CSA 97 Structural data or basic agricultural statistics relate to characteristics of agricultural holdings that vary slowly over time (are normally collected in a Census of Agriculture). 98 Performance data or current agricultural statistics relate to: prices, quantities of inputs and outputs; enterprise costs and returns; and net farm incomes are collected mainly from current (annual) agricultural surveys. CSO and MAFF have been collecting current agricultural statistics since 1964. 99 Up to 1978/79 agricultural season, the survey was called the Agricultural and Pastoral Production Survey, later renamed in 1982/83 as the Early Warning and Agricultural Survey to encompass the Crop Forecasting and Post-Harvest stages of the agricultural season during which period the two different types of surveys were conducted. 100 Although the statistical unit in the Census was the agricultural holding, the agricultural household was used to identify the holding. All the data was collected by interviewing head of households or responsible adults. 101 Provinces were identified by a 1-digit code, districts by a 2-digit code, and both CSAs and SEAs by a 5digit code. 103 is made up of about 3 SEAs. Out of a total of 12,999 SEAs, a sample of 610 SEAs had been selected. The rural stratum had been allocated 349 SEAs.102 Following the 1990 Census of Population, Housing and Agriculture a Master Sample of agricultural SEAs was set up and this sample was used to collect Census of Agriculture data during the period 1990/91 to 1991/92 and during the PHSs of 1992/93 to 1999/2000. The CSO conducted the 2000 Census Mapping exercise from 1998 to 2000. At the time of this mapping exercise a number of new districts had been created. The total number of districts had as a result risen to 72. Parliamentary Constituency and Ward boundaries were also taken into account in demarcating the 2000 Census CSAs and SEAs (CSO, 2002; Kasali, 2002). 4.4.2. Review of Sample Design for 1996/1997 - 1999/2000 Post-Harvest Survey A stratified multi-stage sample design was used for the Zambia PHS. The sampling frame was based on the data and cartography from the 1990 Census described above. The primary sampling units (PSUs) were defined as the CSAs delineated for the census. The CSAs were stratified by district within province and ordered geographically within district. A total sample of 405 CSAs was allocated to each province and district proportionally to its size (in terms of households). A master sample of CSAs was selected systematically with probability proportional to size (PPS) within each district at the first sampling stage; the measure of size for each PSU was based on the number of households listed in the 1990 Census. The secondary sampling unit (SSU) is the SEA, that is, the sampling areas defined as the segment covered by one enumerator during the census. One SEA was 102 The "modified equal allocation method" has been used to allocate the SEAs to provinces. The method allocates Units equally across all the provinces by dividing the sample size by the number of provinces. Then, considering the population size, heterogeneity and homogeneity of the province, the probability proportional to size method yielded additions and subtractions to some provinces. The final results are somewhere between equal and proportional to size allocation (IB Thomsen, 1996). This has been done at all levels and it increases the probability of including even the remote areas in the sample (CSO, 1996e). 104 selected within each sample CSA with PPS for the survey. A new listing of households was conducted within each sample SEA, and the farm size was obtained for each farm household. The listed households within each sample SEA were then divided into two groups based on farm size: Category A for households with less than 5 hectares (ha.) and Category B for households with 5 or more has (table 4.1).103 Table 4.6: Frequency of Holdings in Eastern Province, 1996-2002 PHS 1996/97 PHS 1997/98 PHS 1998/99 PHS 1999/2000 PHS 2000/01 PHS 2001/2002 Frequency Percent Frequency Percent Frequency Percent Frequency Percent Frequency Percent Frequency Percent A-Small scale holding 956 78 1052 88 1111 88.5 1233 85.8 1060 84.9 1128 87.3 B-Medium scale holding 256 22 144 12 144 11.5 204 14.2 189 15.1 164 12.7 Total 1225 100 1196 100 1255 100 1427 100 1249 100 1292 100 Source: Author's calculation. The original sampling plan was to select 10 households from each category within the sample SEA, for a total sample of 20 households per SEA. However, it was found that most sample SEAs had less than 10 households in Category B. In order to ensure a sample of 20 households within each sample SEA, the remaining households were selected from Category A (Megill, 2000).104 Given that a large majority of the rural households in Zambia are involved in agriculture, the sample of farm households is effective for most types of agricultural characteristics being measured by the PHS. In fact, the purpose of the PHS has been to capture relevant data from, and keep abreast with the changes occurring in the agricultural sector (figure A1). Specifically, the objectives of the PHS include provision of actual figures pertaining to: Area planted to individual crops (land usage - allocation); Realised 103 Farm size for the small and medium farmers is difficult to obtain because the claims individual households have to land are not exclusive given that land is customarily owned. The problem is more acute in locations where the Chitemene farming system is practiced. Chitemene System involves the cutting of tree branches in a field to be used for crop production. Typically, CSO uses the land under crops to determine the extent of farm size. 104 Following the data collection for the 1997/98 PHS, it was found that more than 60 percent of the households selected in category B actually had less than 5 HAs, according to the survey data. This is due to changes in the plans of individual households in the amount of land planted in crops, as well as nonsampling error in the listing data. Even with the current level of misclassification, this farm size stratification increases the sampling efficiency for producing estimates of total crop area and production. 105 Production quantities (output in physical units); Sales of produce and income realized; Numbers of livestock and poultry; Purchase and use of agricultural inputs; Capital formation and other operational expenses; Demographic characteristics of heads of rural households; Farming practices and soil conservation methods used; Access to agricultural loans; and, access to market prices information and agricultural extension services in general. The reference period for this information is the agricultural season starting 1st October ending 30th September. However, the PHS estimates for some crops which are rare or limited to particular geographic areas have relatively high sampling errors.105 In order to evaluate the effectiveness of the PHS sample design in meeting these survey objectives Megill(2000) use a CENVAR (Census Variance Calculation System) software to tabulate the standard errors for key survey estimates from the 1997/98 PHS data. The CENVAR results found by Megill(2000) illustrate that the main limitation of the sample design was that it didn't not provide reliable results for rare crops. Moreover, over the period during which the PHSs have been conducted, the survey questionnaire has undergone several major revisions and differences in questions asked. 4.4.3. Review of Sample Design for 2000/2001 - 2001/2002 Post-Harvest Survey The PHS 2001/2002 also covered the whole country and was conducted in a sample of areas numbering 407 SEAs drawn using PPS sampling scheme, representing a sample proportion of about 5%. The survey was conducted in the same CSA and SEAs selected over the previous 4-5 years. The survey relied on the previous listing of household populations in 1999/2000 PHS but with a new sample drawn from this listing. Drawing on the experiences from the Census of Agriculture and the three PHSs that followed, it was realized that estimates for minor crops such as rice, sorghum, cotton, and tobacco were far from being satisfactory. Because of this, it became necessary to revisit the area frame in order to address the situation. In order to try and improve on the 105 The definition of in-scope farm households for the survey should also be examined. Megill recommends certain modifications to the sample design for improving the sampling efficiency for future surveys. 106 estimates for minor crops it was decided to create Crop Zones for these crops. In doing so a number of strata (Zones) were created in order to improve precision and accuracy in the estimates for minor crops. In each district, the allocated sample size was shared proportionately among the crop strata, i.e., the more SEAs a crop stratum had the larger its share of the sample. This was done whilst ensuring that a minimum of two SEAs was selected from each stratum to facilitate computation of sampling error of the estimates. Since the selection of participants in the PHS 2001/02 survey was not done with a simple random sample, a weight variable is used for our analysis.106 We use the overall household weight.107 The District level weight is developed using: The proportion of households who produce the crop in the district, the number of sampled SEA within each CSA, the number of households in the CSA, the number of households in SEA (all from Agricultural Census Survey Data). The district level weight is simply the probability that the number of households in a SEA will be selected as a primary unit from within a CSA within a particular District. After obtaining a complete list of the households in the SEA categorized as small or medium scale and the number of households to be sampled in each SEA, the SEA level weight is estimated. So with the District Level and SEA level weights, these two are multiplied and the product is the boosting factor. Table 4.7: Post Harvest Survey, Sample sizes by District, 1997-2002 District 1996/1997 1997/1998 1998/1999 1999/2000 2000/2001 2001/2002 Chadiza (301) 96 88 89 100 88 100 Chipata (303) 303 295 304 338 307 330 Katete (304) 198 198 199 220 184 212 Lundazi (305) 224 225 229 260 233 261 Petauke (308) 267 262 271 320 262 305 Total Catchment Districts 1088 1068 1092 1238 1074 1208 Chama (302) 37 36 76 80 70 77 Mambwe (306) 52 55 34 59 51 59 Nyimba (307) 48 37 53 60 54 59 Total Control Districts 137 128 163 199 175 195 Total 1225 1196 1255 1437 1249 1403 Sourceμ Authors‘ calculations based on ωSη‘s θost Harvest Surveys 1997-2002. 106 The WGT variable in the ID.dta file is the appropriate weight to use. Another file has been created that contains the weighting value for specific crops. That file is called cropwgt.dta. 107 The Weights (Boosting Factors) are the inverse of the probability that a given household has of being included in the sample. These factors are developed at the SEA level for each category of farmer. 107 The number of sample household in Eastern Province selected during the period 1996/97 – 2001/2002 was on average 1,274 households. They were interviewed during the period December and January using personal interviews with qualified respondents in sample households in sample areas (table 4.7). All PHSs were independent farm surveys and thus interviewed different households in each year. Consequently it is not possible to construct a panel of households using PHSs surveys in order to examine the correlates and causes of changes in the agricultural productivity of individual households over time (McCulloch et al., 2001, UNECA, 2005). 108 4.5. Estimation Results and Discussion This section focuses on the extent to which the productivity of cotton production in Zambia‘s Eastern θrovince from 199θ/1997 to β001/β00β is a result of the combined effects of the 1992 radical agricultural market liberalization and the subsequent rehabilitation of the feeder road network in Eastern Province in the period from 1996 to 2001 (chapter 6).108 4.5.1. Descriptive Statistics We are interested in measuring the impact of rural road interventions (i.e. access to local infrastructure and public goods and capital) on cotton yields per hectare (i.e. farm productivity). Table 4.8 Descriptive Statistics, 1996/1997 – 2001/2002 Variable Dependent variable Household determinants Variable Input use Assets EPFRP Geographic Variables 1998/1999 1999/2000 2000/2001 2001/2002 Full Sample Standard Mean Deviation Full Sample Standard Deviation Mean Full Sample Standard Deviation Mean Full Sample Standard Deviation Mean 1,33 2,31 1,48 2,09 1,62 3,06 1,64 3,02 0,97 0,68 0,97 0,68 Log of cotton output (in kg) per hectare 6,54 1,10 6,83 0,96 6,75 1,17 6,55 1,40 6,65 0,71 6,64 0,71 Age of the household head 46,7 15,0 44,4 15,2 45,5 15,3 43,0 14,3 45,7 14,7 45,3 14,7 2404,0 1506,1 2205,4 1537,9 2307,8 1535,0 2056,0 1371,5 2309,7 1465,6 2270,4 1459,2 Size of the household 5,8 3,2 5,7 3,0 5,94 3,20 6,17 3,43 5,97 2,95 6,34 2,93 Log of Size of the household 1,61 0,59 1,59 0,56 1,63 0,59 1,67 0,56 1,66 0,54 1,73 0,50 Household category (stratum) 1,22 0,41 1,12 0,33 1,11 0,32 1,14 0,35 1,14 0,35 1,13 0,33 Number of males in household 2,79 1,82 2,74 1,85 2,94 1,99 3,08 2,24 2,98 1,93 3,18 1,86 Number of females in household 3,03 1,97 2,91 1,79 2,99 1,85 3,09 1,92 2,98 1,73 3,16 1,81 Sex of head of household 1,23 0,42 1,23 0,42 1,24 0,43 1,24 0,43 1,25 0,43 1,25 0,44 Basal Quantity used (kg) 29,93 123,90 30,88 121,42 39,63 145,91 47,77 129,59 32,81 149,51 34,79 149,91 Topdressing Quantity used (kg) 27,18 104,57 30,50 122,82 38,71 127,18 45,80 118,37 31,98 145,77 33,69 147,14 Basal Fertilizers Used per cultiv. Area (kg per ha) 11,53 36,76 13,05 38,21 16,56 42,32 22,00 53,14 17,17 50,91 16,09 41,98 Top Dressing Fertilizers Used per cultiv. Area (kg per ha) 10,43 28,63 13,32 41,86 16,74 38,74 21,01 47,71 16,10 40,45 15,56 37,37 n.a. Value of Basal quantity used - (ZMK) 31920,3 92680,6 22202,3 238505,6 24409,9 87229,3 34564,7 95575,4 n.a. n.a. n.a. Value of Topdressing quantity used - (ZMK) Expenditure on Basal fertilizers per cultivated area (ZMK/Ha) 27770,4 80701,8 23052,7 241685,4 25208,1 89689,2 33167,1 86535,9 n.a. n.a. n.a. n.a. 12152,0 26389,7 7133,8 26384,2 10491,0 28902,8 15823,0 38979,2 n.a. n.a. n.a. n.a. Expenditure on Topdressing fertilizers per cultivated area (ZMK/Ha) 10284,9 19167,9 n.a. 8317,9 42566,3 10934,1 26732,4 15274,7 35724,6 n.a. n.a. n.a. Number of ploughs 0,374 0,865 0,29 0,77 0,30 0,77 0,27 0,65 n.a. n.a. n.a. n.a. Number of draught animals 0,649 1,741 0,54 1,45 0,57 1,55 n.a. n.a. n.a. n.a. n.a. n.a. Number of ploughs per household member 0,062 0,159 0,05 0,13 0,05 0,13 0,04 0,11 n.a. n.a. n.a. n.a. Number of draught animals per household members 0,099 0,260 0,09 0,25 0,09 0,27 n.a. n.a. n.a. n.a. n.a. n.a. Size of the land allocated to cotton 0,13 0,21 0,12 0,21 0,11 0,20 0,07 0,16 0,10 0,19 0,10 0,18 Total area under crops (ha) 1,97 1,77 1,86 1,74 1,87 1,96 2,10 2,06 1,73 1,65 1,83 1,74 Cultivated land per household member (ha) 0,38 0,33 0,37 0,33 0,35 0,32 0,39 0,45 0,34 0,37 0,36 0,31 Livestock raising 0,58 0,49 0,48 0,50 0,48 0,50 0,50 0,50 0,55 0,50 0,47 0,50 Usage of animal draught power for land preparation 0,27 0,45 0,25 0,43 0,24 0,43 0,28 0,45 0,35 0,48 0,35 0,48 Received agricultural loan 0,323 0,468 0,265 0,441 0,32 0,47 0,16 0,37 n.a. n.a. n.a. n.a. Rural transport infrastructure dummy (EPFRP) n.a. n.a. n.a. n.a. 0,84 0,37 0,83 0,37 0,83 0,37 0,83 0,37 Aggregate agricultural - Year effects - Length of Roads Network per total area of District (km / km2) Agricultural extension services 1997/1998 Full Sample Standard Deviation Mean Volume of cotton production per hectare produced (MT) Age Square of the household head Household demographics 1996/1997 Full Sample Standard Mean Deviation 7,47 4,32 7,47 4,32 7,47 4,32 7,47 4,32 7,47 4,32 7,47 4,32 Cotton-specific effect (OLS fitted values) 0,148 0,049 0,146 0,048 0,118 0,055 0,121 0,057 0,122 0,042 0,113 0,046 Information on marketing for agricultural products 0,46 0,50 0,39 0,49 0,33 0,47 0,30 0,46 n.a. n.a. n.a. n.a. Use any of the advice received on Crop husbandry 0,28 0,45 0,20 0,40 0,20 0,40 0,01 0,10 n.a. n.a. n.a. n.a. Use any of the advice received on Crop diversification 0,23 0,42 0,12 0,32 0,16 0,37 0,14 0,35 n.a. n.a. n.a. n.a. Information on agricultural input supply 0,41 0,49 0,35 0,48 0,32 0,47 0,23 0,42 n.a. n.a. n.a. n.a. Proportion of sample in Catchment Areas 0,85 0,36 0,85 0,36 0,84 0,37 0,84 0,37 0,83 0,37 0,83 0,37 Proportion of sample in Control Areas Distance to the nearest all-weather road 0,15 0,36 0,15 0,36 0,16 0,37 0,16 0,37 0,17 0,37 0,17 0,37 1,374 0,603 n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. Distance to the nearest input market 1,855 0,784 n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. Rainfall 831,5 122,9 716,0 81,4 788,2 148,1 667,1 93,8 980,1 203,6 723,7 89,4 Cotton Observations 421 378 388 279 467 492 Total number of Observations 1219 1197 1255 1427 1249 1403 Sourceμ Authors‘ estimations based on θHS. 108 Grain marketing was liberalized in 1992 by the MMD government barely one month after coming to power, the same year that financial liberalization occurred. 109 This outcome of interest is a continuous variable with a mean ranging from 6.54 in 1996/1997 to 6.83 in 1997/1998 and a standard deviation from 0.71 in 2001/2002 to 1.40 in 1999/2000. The treatment EPFRP variable is discrete and of on/off variety. Distribution of the Treatment and Comparison Samples The sample characteristics of the comparison group and the treatment group highlight the role of randomization in the sense that the distribution of the covariates for the treatment and control groups are not significantly different. The age of the head of household in 1996/97 was only 2 years higher in the catchment districts, whereas in 2001/2002 it was almost similar. The size of the household was likewise equivalent in both 1996/97 and 2001/2002, although a bit higher in the catchment areas in entire period, exclusive in 1998/1999. The same could be said about the number of males in the household with the number in the catchment areas again being slightly higher. A more synoptic way to view these differences is to use the estimated propensity score as a summary statistic (see section 4.5.2.1 below). 4.5.2. Evaluation of the EPFRP‟s impact on Cotton Productivity The standard problem in treatment evaluation involves the inference of a causal connection between the treatment and the outcome. In our single-treatment case in each cross-section we observe (yi, xi, Diν i = 1, …, ζ) the vector of observations on the scalar-valued outcome variable y, a vector of observable variables x, a binary indicator of a treatment variable D, and let N denote the number of randomly selected individuals who are eligible for treatment. Let NT denote the number of randomly selected individuals who are treated and let NNT = N – NT denote the number of nontreated individuals who serve as a potential control group. We would like to obtain a measure of the impact of the EPFRP intervention in D on y, holding x constant. The situation is akin to one of missing data, and it can be tackled by methods of causal inference carried out in terms of (policy-relevant) counterfactuals. We ask how the outcome of an average untreated individual agricultural household would change if such a person were to receive the treatment. That is, the magnitude Δy/ΔD is of interest. Fundamentally our interest lies in the outcomes that result from or are caused by the EPFRP interventions. Here the causation is in the sense of ceteris paribus (Cameron and Trivedi, 2005). 110 The average selection bias is the difference between programme participants and nonparticipants in the base state. Selection bias arises when the treatment variable is correlated with the error in the outcome equation (Baltagi, 2001). In our observational data the problem of selection of observables is solved in the subsequent sections using regression and matching methods. 4.5.2.1. Matching and Propensity Score Estimators Our data combine the treated units from a randomized evaluation of the EPFRP with nonexperimental comparison units drawn from the PHS data. The outcome of interest is logyield (logarithm of cotton yield); the treatment (EPFRP) is participation in the EPFRP treatment group. Control variables are: Age, age squared, sex (1 if male, 0 female), share of male in household, livestock ownership (1 if yes, 0 otherwise) and stratum (1 if small-scale, 2 if medium-scale). The treatment group contains 4070 observations, the control group 2755 observations, so the total number of observations is 6825. The propensity score - the conditional treatment probability - is estimated by the program on the independent variables. It is noted that the sort order of our data could affect the results when using nearest-neighbor matching on a propensity score estimated with categorical (non-continuous) variables. Or more in general when there are untreated with identical propensity scores. There are many options for fine tuning the matching estimators (Abadie et al., 2004). The output from running the estimation of the propensity score (pscore) using this linear specification is shown in table 4.10a. Table 4.10a: Estimation of the propensity score Table 4.10.b: Inferior bound, the number of treated, and the number of controls EPFRP Inferior of block of EPFRP (1) Age -0.0186 (0.0173) 0 1 Total pscore Age squared 0.0002 (0.0002) .2 3 5 8 Sex -0.1065 (0.0648) .4 207 180 387 hhsize 0.0393*** (0.0093) .5 452 503 955 Share of male 0.3913*** (0.1388) .55 1003 1452 2455 Stratum -0.0598 (0.0775) Livestock ownership 0.1873*** .6 872 1299 2171 (0.0532) Cotton land fraction -0.7536*** .65 166 397 563 (0.1313) Constant 0.5550 .7 13 33 46 (0.3595) Number of Obs. 6586 .8 0 1 1 R F Total 2716 3870 6586 Log likelihood(i) -4427.1405 2 Notes: Standard errors in parentheses * p<0.10, ** p<0.05, *** p<0.01. (i) Estimated after two iterations. Sourceμ Authors‘ estimations. Sourceμ Authors‘ estimations. 111 The region of common support is [0.35378466, 0.85246556]. Following the algorithm described in Becker and Ichino(2002) the first step is to estimate the propensity score is to identify the optimal number of blocks for which the average propensity scores of treated and controls differ are split in half. The algorithm continues until, in all blocks, the average propensity score of treated and controls does not differ. In our case, this happens for a final number of nine blocks, which ensures that the mean propensity score is not different for treated and controls in each blocks. In the second step we proceed to tests the balancing property of the propensity score for each covariate. The balancing property is satisfied in our case with the significance level of the tests of the balancing property p=0.001 and the rainfall variable removed from the original specification of the propensity score.109 Since, the balancing property holds, the final distribution of treated and controls across blocks is tabulated together with the inferior of each block in table 4.10b. After running pscore, we proceed to estimate average treatment effects (ATTs). In table 4.10c we present the results of a range of simple Matching methods. The first estimator that we consider in row 1 of table 4.10c is the „One-to-One propensity score matching‟. We find that the difference between the matched treated and the matched control is minus 0.223 while the T-statistics for H0 is minus 3.080 for ATT. In the 2nd row we present another method of selecting a set of comparison units, that is, the ‗Nearest-neighbour matching‟ without replacement for which the treated unit i is | pi – pj | = min | pi  pk | matched to that non-treated unit j such that: (5.1) k D  0 The nearest-neighbor method selects the m comparison units whose propensity scores are closest to the treated unit in question. We calculate and display in table 4.10c the effect by the difference between the matched treated and the matched controls, which is minus 0.201 and Tstatistics for H0 minus 2.75 in the case of ATT.110 Row 3 displays the results of the caliper matching method, which uses all of the comparison units within a predefined propensity score radius (or “caliper”). A benefit of caliper matching is 109 Note that this significance level applies to the test of each single variable of the vector X of pre-treatment characteristics, i.e. the balancing property is not rejected only in case it holds for every single X. This is a relatively conservative approach according to Becker & Ichino(2002). 110 The Abadie and Imbens (2002) procedure on match on the contrary allows individuals to be used as a match more than once, which generally lowers the bias but increases the variance. 112 that it uses only as many comparison units as are available within the calipers, allowing for the use of extra (fewer) units when good matches are (not) available (Dehejia and Wahba, 2002). Table 4.10c: Matching and Propensity Score Estimators Propensity score matching methods (i) Variable Sample Treated logyield Unmatched 6,650 ATT 6,701 ATU 6,916 ATE 1. One-to-One propensity score matching (ii) logyield Unmatched 6,650 ATT 6,724 ATU 6,909 ATE 2. K-nearest neighbors matching (iii) logyield Unmatched 6,650 ATT 6,729 ATU 6,872 ATE 3. Radius matching (iv) logyield Unmatched 6,650 ATT 6,760 ATU 6,852 ATE 4. Kernel (v) logyield Unmatched 6,6503 ATT 6,7638 ATU 6,8330 ATE 5.Local linear regression (vi) logyield Unmatched 6,650 ATT 6,724 ATU 6,909 ATE 6.'Spline-smoothing' (vii) logyield Unmatched 6,650 ATT 6,716 ATU 6,690 ATE 7. Mahalanobis matching (viii) Controls Difference 6,726 -0,075 6,924 -0,223 6,673 -0,243 -0,233 6,726 -0,075 6,926 -0,201 6,761 -0,148 -0,178 6,726 -0,075 6,919 -0,190 6,735 -0,137 -0,168 6,726 -0,075 6,762 -0,003 6,852 0,000 -0,001 6,7255 -0,0752 6,7786 -0,0148 6,8423 0,0093 -0,0045 6,726 -0,075 6,923 -0,198 6,700 -0,209 -0,203 6,726 -0,075 6,721 -0,005 6,700 0,009 0,001 S.E. 0,043 0,072 T-stat -1,740 -3,080 0,043 0,073 -1,740 -2,750 0,043 0,074 -1,740 -2,560 0,043 0,052 -0,012 -1,740 -0,050 0,0432 0,1456 -1,7400 -0,1000 0,0432 , -1,740 , 0,043 0,063 -1,740 -0,080 Notes: (i) A variety of propensity score matching methods to adjust for pre-treatment observable differences between a group of treated and a group of untreated. Treatment status is identified by depvar==1 for the treated and depvar==0 for the untreated observations. Sourceμ Authors‘ estimations using the PSMATCH2 Stata module. We achieve the best result by using „Kernel-based matching‟ as shown in row 5, that is the idea to associate to the outcome yi of treated unit i a matched outcome given by a kernel-weighted average of the outcome of all non-treated units, where the weight given to non-treated unit j is in proportion to the closeness between i and j: (5.2) K      Pi  Pj  yj h  j D  0  yˆ i   P  Pj   K  i  h  jD  0  113 By choosing the uniform kernel type and imposing common support on the treated,111 we find that the ATT difference between the treated and the control is almost zero (-0.003). In row 7 the difference is almost the same (-0.005) when carrying out Mahalanobis metric matching, by replacing pi – pj above with d(i, j) = (Pi – Pj)‘ S-1 (Pi – Pj), where  Pi is the (2x1) vector of scores of unit i  Pj is the (2x1) vector of scores of unit j  S is the pooled within-sample (2x2) covariance matrix of P based on the sub-samples of the treated and complete non-treated pool (Sianesi, 2001). The fact that there is substantial overlap in the distribution of the propensity score between the comparison and treatment groups, explains why most of the matching algorithms yields similar results in table 4.10c. Therefore finding a satisfactory match by matching without replacement is appropriate given our PHS datasets. In the output in table 4.10d we estimate respectively the ATE; ATT; and ATC for the sample. Since cotton productivity is recorded in natural logarithm, the output in row 1 in table 4.10d relying on only a single match implies that for the individual households in our sample, the SATE of benefiting from the EPFRP is a higher absolute increase for SATT of 0,192 compared to 0,057 for SATC. For all the specifications at hand we conclude that the sample ATTs are significantly different from zero at the 1% level, whereas the ATCs are insignificant, by using 3 matches.112 Since the standard error of the SATEs underestimates the standard error of the PATE, it is possible that the PATE might not be significantly different from zero at either the 5% nor the 1% level (Abadie et al., 2001, 2004). However, when considering launching another rural road rehabilitation programme in Eastern Province in which we would obtain another sample from the same population, the absolute increase in PATT of -0,208 is higher compared to PATC of -0,061 and that PATT is significantly different from zero at the 1% level. Moreover, since our productivity data are in terms of logarithms, our results would indicate a statistically significant but also 111 Treated units whose p is larger than the largest p in the non-treated pool are left unmatched. We chose 3 matches because it seemed to offer the benefit of not relying on too little information without incorporating observations that are not sufficiently similar. Like all smoothing parameters, the final inference can depend on the choice of the number of matches (Abadie et al., 2001, 2004). 112 114 economically important impact of the EPFRP on the individual rural household in the pooled PHS samples covering the period from 1996/1997 to 2001/2002. Finally, as discussed in Imbens (2003) and Heckman et al. (1998) the effects of the treatment on the sub-population of treated units (SATTs) are more important than the effect on the population as a whole (SATE) as shown by our results displayed in table 4.10d. Table 4.10d: Matching estimators for Average Treatment Effects Matching estimator: Average Treatment Effect Average Treatment Effect for the Treated Average Treatment Effect for the Controls 2 (i) Average Treatment Effect Average Treatment Effect for the Treated Average Treatment Effect for the Controls 3 (ii) Average Treatment Effect Average Treatment Effect for the Treated Average Treatment Effect for the Controls 4 (iii) Average Treatment Effect Average Treatment Effect for the Treated Average Treatment Effect for the Controls No. 1 Number of Number of matches matches, robust m(#) std. err. (h) 1 1 1 3 3 3 3 3 3 3 4 3 4 3 4 logyield SATE SATT SATC SATE SATT SATC SATE SATT SATC SATE SATT SATC Coef. -0,134 -0,192 -0,057 -0,148 -0,216 -0,057 -0,187 -0,268 -0,078 -0,187 -0,268 -0,078 Std.Err. 0,050 0,056 0,058 0,046 0,050 0,051 0,046 0,051 0,050 0,044 0,046 0,049 z -2,680 -3,410 -0,970 -3,210 -4,330 -1,120 -4,020 -5,280 -1,550 -4,270 -5,770 -1,570 P>z [95% Conf. 0,007 -0,232 0,001 -0,302 0,330 -0,172 0,001 -0,238 0,000 -0,314 0,262 -0,156 0,000 -0,277 0,000 -0,368 0,122 -0,176 0,000 -0,272 0,000 -0,359 0,116 -0,174 Interval] -0,036 -0,082 0,058 -0,058 -0,118 0,042 -0,096 -0,169 0,021 -0,101 -0,177 0,019 Notes: 4662 observations dropped due to treatment variable missing. Number of obs = 2163. Matching variables: Age Agesq Sex shareofmale loghhsize stratum basalprha Topdresprha livestock Areapc Clandfrac rain_EP. Bias-adj variables: Age Agesq Sex shareofmale loghhsize stratum basalprha Topdresprha livestock Areapc Clandfrac rain_EP. (i) Homoskedastic errors are estimated. (ii) The nnmatch estimate heteroskedasticity-consistent standard errors using # matches in the second matching stage (across observations of the same treatment level). (iii-iv) We estimate the ATE; ATT and ATC with bias-adjustment. The k*k diagonal matrix of the inverse sample standard errors of the k variables in varlist_nnmatch is used. (iii) Exclusively use the Bias Corrected Matching Estimator. (iv) Whereas the variance Estimation allows for Heteroskedasticity. The Bias Corrected Matching Estimator The simple matching estimator will be biased in finite samples when the matching is not exact. In practice one may therefore attempt to remove some of this bias term that remains after the matching. The bias-corrected matching estimator adjusts the difference within the matches for the differences in their covariate values. The adjustment is based on an estimate of the two regression functions (x) = E[Y ( )|X = x]. Following Rubin (1973) and Abadie and Imbens (2002) we approximate these regression functions by linear functions and estimate them using least squares on the matched observations (Abadie et al., 2001, 2004). Using the Bias Corrected Matching Estimator for the ATE: ˆMbcm  (5.3) 1 N ~ ~  (Y1i  Y0i ) N i 1 And the bias-adjusted matching estimators for ATT and ATC: 115 ˆMbcm,t  (5.4a) (5.4b) ˆMbcm, c  1 ~ ~ (Y1i  Y0i ) ,  N1 i:Di 1 1 ~ ~ (Y1i  Y0i )  N 0 i:Di  0 We estimate the SATE, SATT and SATC in rows 3. We find that this approach both increase the absolute size of the coefficients and decrease the standard errors, while not changing our previous conclusion that EPFRP treatment had an effect on its participants that still is significant at the 1% level.113 Variance Estimation Allowing for Heteroskedasticity In row 4 we show the results for the variance of the SATE: 1 N 2 sample ˆ  2  1  K Mi  ˆ D2 i ( X i ) V N i 1 (5.5a) Similarly the variance for the estimator for SATT is: (5.5b) and for SATC, (5.5c) 1 N sample, t ˆ V  2  Di  (1  Di K Mi ) 2  D2 i ( X i ) N1 i 1 1 N 2 Vˆ sample, c  2  Di K Mi  (1  Di   D2 i ( X i ) N 0 i 1 We estimate these variances by estimating the conditional outcome variance ˆ2 ( x) , which is assumed not to be constant (i.e. heteroskedastic) for both treatment groups ( ) and all values of the covariates (x). This is implemented using a second matching procedure, now matching treated units to treated units and control units to control units (Abadie et al., 2001, 2004). In other words, the SATE; SATT; and SATC is re-estimated in row 4, but compared to row 1-3 we estimate the standard error allowing for heteroskedasticity, while specifying 3 data matches in estimating the conditional variance functions. Our results show that when the standard error is estimated under these weaker conditions the estimated SATE and SATT are still significant at the 1% level. The in row 4 the EPFRP appears to have had exactly the same significant impact on the beneficiaries as in row 3, although standard errors are slightly smaller by taking account of heteroskedasticity. 113 The bias-adjustment does not affect the form of the estimator for the variance, although it may affect the numerical value. For the variance it does matter whether one is interested in the sample of population average treatment effect (or the average effect for the treated or controls) (Abedie et al., 2001, 2004). 116 4.5.2.2. Differences-in-Differences Estimators In our case we have data on the treated and the comparison (control) groups both before and after the experiment (i.e. implementation of the EPFRP). One way to improve on the one-group before and after design, which makes the strong assumption that the group remains comparable over time, is to include an additional untreated comparison group, that is, one not impacted by policy (i.e. EPFRP), and for which the data are available in both periods (Cameron and Trivedi, 2005). Since the work by Ashenfelter and Card(1985), the uses in differences-in-differences methods has become very widespread. The first-differences estimator for the fixed effects model reduces to a simple estimator called the differences-in-differences estimator. The latter estimator has the advantage that it can also be used when repeated cross-section data rather than panel data are available. However, it does rely on model assumptions that are often not made explicit (Blundell and Macurdy, 1999; Cameron and Trivedi, 2005). In the context of the analysis of our experimental data the simple comparison of the mean of the outcome in treatment and control groups (the „differences‟ estimator) is justified on the grounds that the randomization guarantees they should not have any systematic differences in any other pre-treatment variable.114 In the absence of treatment, the unobserved differences between treatment and control groups are the same over time. In this case one could use data on treatment and control group before the treatment to estimate the „normal‟ difference between treatment and control group and then compare this with the difference after the receipt of treatment. This removes biases in second period comparisons between the treatment and control group that could be the result from permanent differences between those groups, as well as biases from comparisons over time in the treatment group that could be the result of trends. The validity of the differences-in-differences estimator is based on the assumption that the underlying „trends‟ in the outcome variable is the same for both treatment and control group. This assumption is never testable and with only two observations one can never get any idea of whether it is plausible. But, from figures 4.2a-b depicting more than two observations we can get some idea of its plausibility. 114 In economic applications treatment and intervention usually mean the same thing (op.cit., p.860). The term outcome refers to changes in economic status or environment on economic outcomes of individuals (ibid.). 117 6.4 6.6 6.8 7 7.2 Fig. 4.2a: „Trends‟ in the Log of Cotton Yield for Treatment & Control group, 1996-2001 1996 1997 1998 year 1999 logyield_t 2000 2001 logyield_nt Sourceμ Author‘s estimations. 5.8 6 6.2 6.4 6.6 Fig. 4.2b: „Trends‟ in the log of Cotton Production for treatment & control group, 1996-2001 1996 1997 1998 year 1999 Avglcotprod_catch 2000 2001 Avglcotprod_control Sourceμ Author‘s estimations. An important feature of panel data is that it allows the estimation of parameters characterizing dynamics from individual level data. Several authors, including Moffitt(1993), argue that such parameters can also be identified from repeated cross-section (RCS) data such as the PHS data. While there are differences across estimators used by these authors, the approach employed is either explicitly or implicitly instrumental variables (IV). The estimators first aggregate the individual data into cohorts comprising individuals with some similar observed characteristic(s) (e.g. year of birth). Using this pseudo-panel the lagged dependent variable is then replaced by a predicted value from an auxiliary regression and the dynamic model is subsequently estimated via ordinary least squares (OLS), IV or GMM (Verbeek and Vella, 2004). A paper by Verbeek and Vella(2004) reviews the identification conditions underlying these estimators. They first conclude from their analysis that while it is possible to identify individual dynamics without having individual time series or panel data, this requires the existence of a set of 118 instruments that are observed for each person in the entire sample, that exhibit sufficient (timevarying) correlations with each explanatory variable in the model of interest, and that have no (time-varying) correlations with the equation‘s unobservables. Such instruments can be used to aggregate the data in a number of mutually exclusive groups (cohorts). While the availability of such instruments may be unrealistic in certain applications, their validity is, generally, not testable due to its identifying nature. Overall, the size of the standard errors is no indication for the amount of bias that is present. Secondly, Verbeek and Vella(2004) conclude that given moderate sample sizes, the sampling variation of the IVe stimator is mainly driven by the importance of the cohort effects in the exogenous variables. When the number of instruments (cohorts) is large relative to the number of individuals, a small sample bias is present in the IV estimator. When cohort sizes are 100 or more, these biases seem to be acceptable, provided sufficient cohort-specific variation is present in the exogenous variables. Thus, we use the repeated cross-section (RCS) data to construct pseudo or synthetic panel data that have some of the advantages of genuine panel data, most notably the ability to control for fixed effects. This data structure is useful for our policy analysis purposes. Since the distribution of variables from the pooling of independent cross-sections tends to change over time, the identical distribution assumption is not usually valid, but the independence assumption is. This approach gives rise to independent, not identically distributed (i.n.i.d) observations (Wooldridge, 2002).115 We have two time periods, 1996/1997-1997/1998 (i.e. the pre-EPFRP treatment period) and 1998/1999 – 2001/2002 (i.e. the post-EPFRP treatment period), which straddle the policy change. We include a time period dummy variable for the second (post-policy change) time period 1998/1999 – 2001/2002, which we denote D2 to account for aggregate changes (factors) that affect the dependent variable over time in the same way for the two groups, the neighboring three control districts (no.: 302, 306-307) and the five similar treatment districts (no.: 301, 303-305, and 308). Define i = 0 for the control group and i=1 for the treatment (catchment) group. Define t = 0 to be a pre-treatment period and t=1 to be the post-treatment period. 115 In a pooling of cross sections over time, unlike panel data set, there is no replicability over time. 119 The dummy variable (EPFRP) Treatment equals unity for those in the treatment group and is zero otherwise. The equation for analyzing the impact of the EPFRP treatment policy change known as the pooled cross-section time-series model or constant-coefficient model is:116 (5.6) yit = 0 + 0Dβ + 1 EθFRθ+ 1Dβ*EθFRθ + i Xit + uit, where y is the outcome variable (log yield) of interest. The presence of EPFRP by itself captures possible differences between the treatment and control groups before the policy change occurred. The coefficient of interest, δ1, multiplies the interaction term, D2*EPFRP, which is simply a dummy variable equal to unity for those observations in the treatment group in the second period. X comprise the additional covariates in equation (5.6), which account for the possibility that the random samples within a group have systematically different characteristics in the two time periods. Define μit to be the mean of the outcome in group i at time t. The difference estimator simply uses the difference in means between treatment and control group post-treatment as the estimate of the treatment effect i.e. it uses as estimate of (μ11 – μ01) (see table 4.11). However, this assumes that the treatment and control groups have no other differences apart from the treatment, a very strong assumption with non-experimental data. A weaker assumption is that any difference in the change in means between treatment and control groups is the result of the treatment i.e. to use an estimate of: (5.7) ˆ1 = ( 11 – 01) –( 10 – 00) = ( y 2  y1 )  ( y 2  y1 )   y   y tr tr nt nt tr nt as an estimate of the treatment effect – this is the differences-in-differences estimator (DID).117 The DID estimator is the OLS estimate of δ1, the coefficient on the interaction between EPFRP and D2. This is a dummy variable that takes the value one only for the treatment group in the posttreatment period. In practice we write the DID estimator ˆ1 as ( 116 11 – 01) –( 10 – 00) note that the In the statistics literature the model is called a population-averaged model, as there is no explicit model of yit conditional on individual effects. Instead, any individual effects have implicitly been averaged out. The random effects model is a special case where the error uit is equicorrelated over t for given i (Cameron & Triverdi, 2005:720). 117 Since one estimates the time difference for the treated and untreated groups and then takes the difference in the time differences (Cameron & Trivedi, 2005:769). 120 first term is the change in outcome for the treatment  y and the second term the change in tr outcome for the control group  y and then estimate the model (5.6). nt By comparing the time changes in the means for treatment and control groups, both groupspecific and time-specific effects are allowed for. Nevertheless, unbiasedness of the DID estimator still requires that the policy change not be systematically related to other factors that affect y (and are hidden in u). The estimated equation (column 4 in table 4.11) is (5.6b)  log( yield )  7.201 * * * - 0.0394 D2 * - 0.011 Treatment * * 0.0598D2 x Treatment     ... (0.2364) (0.0777) (0.0035) (0.1090) Therefore, ˆ1 = 0.0598 (t= 0.55), which implies that the average cotton yield in Zambia‘s Eastern Province increased by about 6 percent due to improved rural transport infrastructure development. The coefficient on D2 is both fairly small and statistically insignificant. The coefficient on the EPFRP Treatment is also negative at the same level, but statistically significant at the 1 percent level. Thus, in the absence of change in treatment districts, the treatment districts actually saw a fall in cotton yield over time, which corresponds to figures 4.2a-b above. (5.8) Δyi = yi1 – yi0, Equation (5.8) is simply the difference estimator applied to differenced data.118 The treatment effect δ can be consistently estimated by pooled ηδS regression of Δyit on ΔXit and a full set of time dummies (column 1 in table 4.11) (Cameron and Trivedi, 2005, Wooldridge, 2002). The individual-specific fixed effects αi is eliminated by first differencing of the fixed effects model for yit = ϕDit + + αi + it. 118 121 t Table 4.11 Basic Log-Productivity & Log-Production Regressions Productivity (Kg/HA) Difference-in-Differences Simple Difference Estimator Type Independent Variables CV Age of Head of HH CV Age Squared Production (Kg) Difference-in-Differences Simple Difference Estimator Model 1 Model 2 Model 1 Model 2 Model 1 Model 2 Model 1 (1) (2) (3) (4) (1) (2) (3) Model 2 (4) dlogyield dlogyield logyield logyield dlcotprod dlcotprod lcotprod lcotprod 0.0106 0.0096 0.0070 0.0054 0.0175* 0.0183* 0.0152 0.0153 (0.0085) (0.0086) (0.0085) (0.0086) (0.0100) (0.0101) (0.0101) (0.0101) -0.0002* -0.0001* -0.0001 -0.0001 -0.0002* -0.0002** -0.0002* -0.0002* (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) 0.2415*** 0.2406*** 0.2200*** (0.0001) DV Sex 0.1179** 0.1148** 0.1017* 0.1026* 0.2198*** (M=1; F=0) (0.0563) (0.0561) (0.0568) (0.0565) (0.0632) (0.0632) (0.0642) CV Household Size 0.0012 0.0021 0.0027 0.0020 0.0342*** 0.0344*** 0.0367*** (0.0076) (0.0076) (0.0077) (0.0077) (0.0080) (0.0080) (0.0081) (0.0081) CV Share of Males in HH -0.0669 -0.0746 -0.0846 -0.0989 -0.3063** -0.3045** -0.2950** -0.2950** (0.1176) (0.1177) (0.1193) (0.1190) (0.1328) (0.1327) (0.1355) DV Farm Type (Stratum) -0.0811 -0.0845 -0.1108* -0.1126* 0.4989*** 0.5003*** 0.4649*** (0.0643) 0.0367*** (0.1355) 0.4650*** (SSF=1; MSF=2) (0.0586) (0.0586) (0.0590) (0.0589) (0.0651) (0.0649) (0.0655) CV Livestock 0.0626 0.0617 0.0784* 0.0738* 0.2493*** 0.2485*** 0.2509*** (0.0417) (0.0416) (0.0421) (0.0419) (0.0479) (0.0479) (0.0479) CV Rainfall District Level 0.0006*** 0.0004** 0.0001 0.0000 0.0004** 0.0005** -0.0000 -0.0000 (0.0002) (0.0002) (0.0001) (0.0001) (0.0002) (0.0002) (0.0001) (0.0002) DV CV DV DV CV EPFRP Treatment (T=1; NT=0) -871.9633*** -0.1605** -96.4543 (237.0590) (0.0760) (304.8437) Cotton Landfraction -1.3831*** -1.3180*** -1.3408*** (0.1047) (0.0998) (0.1053) D2 D2*Treatment (0.1049) (0.0479) -0.1629* (0.0930) 1.0681*** 1.0667*** 0.1045 0.1050 (0.1330) (0.1326) (0.0927) (0.0929) 0.0000 0.0000 -0.0192 -0.0394 0.0000 0.0000 0.0000 -0.1690 (0.0000) (0.0000) (0.0774) (0.0777) (0.0000) (0.0000) (0.0000) (0.1296) 871.7184*** -0.0699 0.0000 0.0598 96.2213 -0.3147** 0.0003 (237.0540) (0.1182) (0.0000) (0.1090) (304.8399) (0.1393) (0.0043) -0.0077** EPFRP Treatment (share) Constant -1.3625*** (0.0654) 0.2510*** -0.0105*** (0.0037) (0.0035) 0.4334*** 0.3847*** 7.0582*** (0.0419) (0.0368) (0.2364) Observations 2364 2364 2364 R2 0,09 0,09 F 18,43 ll -3248,77 7.2007*** 0.0039 0.0003 (0.0044) (0.0043) -0.5039*** -0.5057*** 4.5598*** (0.0531) (0.0515) (0.2816) 2364 2073 2073 2073 0,08 0,09 0,13 0,13 0,12 0,12 19,06 16,87 16,75 25,21 25,72 23,05 21,28 -3251,69 -3269,71 -3264,97 -2965,14 -2964,77 -2984,92 -2984,91 (0.2423) 4.5561*** (0.2853) 2073 Notes: Robust Standard Errors in parentheses: * p<0.10, ** p<0.05, *** p<0.01. Sourceμ Authors‘ estimations. Browning, Deaton and Irish(1985) and Deaton(1985) in their empirical studies consider methods for analyzing RCS data. Deaton (1985) suggests tracking cohorts and estimating economic relationships based on cohort means rather than individual observations. Deaton (1985) argued that these pseudo panels do not suffer the attrition problem that plagues genuine panels, and may be available over longer time periods compared to genuine panels. For a pseudo-panel with N observations on C cohorts, the fixed effects estimator based on the within `cohort' transformation, is a natural candidate for estimating . However, Deaton (1985) argued that these sample-based 122 averages of the cohort means can only estimate the unobserved population cohort means with measurement error. Therefore, one has to correct the within estimator for measurement error using estimates of the errors in measurement variance covariance matrix obtained from the individual data. Details are given in Deaton (1985). There is an obvious trade-off in the construction of a pseudo panel. The larger the number of cohorts, the smaller is the number of individuals per cohort (Baltagi, 2001). Browning et al. (1985) computed group means of yi(r)t and Xi(r)t within cohort-age cells and regressed the y means on the x means and cohort dummies. This procedure is a special case of the one outlined in Moffitt(1993), with a full set of age, cohort, and cohort-age interaction dummies appearing. Their suggestion was to convert our N = 4493 observations of individual-household level data into C (46=65-20) age-group of cohorts observations at the cohort-level data. In this case, C is large and the pseudo panel is based on a large number of observations N. However, the fact that the average cohort size nc = N/C = 97.67 is not large implies that the sample cohort averages are not precise estimates of the population cohort means, or as expressed by Verbeek and Nijman(1992a) the observed cohort means can be regarded as error-ridden measurements of the true population cohort means.119 In our case, we have a large number C=46 of imprecise observations (Baltagi, 2001). If the number of observations per cohort is large, it is tempting to ignore the errors-invariables problem and to use standard software to handle the pseudo panel as if it were a genuine panel. This is what was done in empirical studies e.g. by Browning, Deaton and Irish(1985). Verbeek and Nijman (1992a) analyze to what extent this is a valid approach. Verbeek and Nijman (1992a) show that the effects of ignoring the fact that only a pseudo panel is available will be small if the cohort sizes are sufficiently large (100, 200 individuals) and if the true means within each cohort exhibit sufficient time variation. In contrast, a pseudo panel constructed with a smaller number of cohorts and therefore more individuals per cohort is trading a large pseudo panel with imprecise observations for a smaller pseudo panel with more precise observations. Verbeek and Nijman (1992b) find that nc = ∞ is a crucial condition for the consistency of the within estimator and that the bias of the within estimator 119 Blundell, Meghir and Neves (1993) use the annual U.K. Family Expenditure Survey covering the period 1970-1984 to study the intertemporal labour supply and consumption of married women. The total number of households considered was N=43,671. These were allocated to C=10 different cohorts depending on the year of birth. The average number of observations per cohort was nc = N/C = 364 (Baltagi, 2001). 123 may be substantial even for large nc. On the other hand, Deaton's estimator is consistent for , for finite nC, when either C or T tend to infinity (Baltagi, 2001). Although individual household production cannot be tracked through time, it is possible to do so for cohorts of individuals. However, in neither of these cohort specifications do we find that the treatment effect is significant (table A11). Moffitt(1993) concludes that there is a considerable amount of parsimony achieved in the specification of age and cohort effects. Also, individual characteristics are considerably more important than either age, cohort or year effects (Baltagi, 2001). 4.5.2.3. Further Specification Table 4.12 reports the productivity results corrected for entry and exit, where we consider the treatment variable as respectively a dummy variable (Model 1) and as a percentage share of the feeder road network (Model 2). Column (1) reproduces the Difference-in-Differences estimates from column (3) of Table 4.9, which does not include controls for . In columns (2) we present the results of linear OLS regression model with heteroskedasticity-robust standard errors for the monotone transformed dependent variable in natural logarithm on the same regressors. In both models the regressors are jointly statistically significant, because the overall F(10, 1346) statistics of respectively 33.13 and 32.95 have p-values of 0.0000 and 0.0000. At the same time, much of the variation is unexplained with Model 1‟s R2 = 0.1962 and Model β‟s R2 = 0.1945. By using a two-sided test at level 0.10, only the regressors: EPFRP; household size; share of males; farm type; rainfall and the estimated cotton land fraction in Model 1 are individually statistically significant, whereas for Model 2 only the EθFRθ‘s percentage share of feeder road network; and same covariates as in Model 1 are individually statistically significant. In Model 1 the significant coefficient of the key regressor EPFRP is only 0.2317, which means that the presence of the EPFRP is associated with a 23.17% increase in cotton productivity, whereas in Model 2 the EθFRθ‘s percentage share only is associated with a 0.9% increase. The sign of these EPFRP coefficients are correct, but there is a huge difference in the magnitude of the coefficient between the two models. Moreover, because of censoring, which is more severe for low values of the explanatory variable, the OLS regression line has a slope less than one.120 120 The censoring of low values of x keeps the OLS slope down. 124 A method to overcome the inconsistency of OLS is to follow Tobin, because the cross-section of PHSs reveal that a significant proportion ranging from 63 to 79 percent of the households with zero cotton production (i.e. censoring observations) and the rest with positive levels of cotton productivity. The sample is therefore a mixture of observations with zero and positive values. The censored regression model, or Tobit model, is relevant because the dependent „logarithm of cotton productivity‟ variable is observed only over some interval of its support.121 In fact this is a case of left-censored PHS data with the censoring point (L=0). Thus, in columns (3) we estimate a linear regression in the presence of censoring by using a Tobit model. In the face of the heteroskedasticity, the Tobit procedure yields estimates that are as biased up as OLS is biased down. Deaton(1997μ88) warns that ―there is no general guarantee that the attempt to deal with censoring by replacing OLS with the Tobit MLE will give estimates that reduce the bias.‖ Table 4.12: Tobit Model Comparisons of Log Likelihood Model 1 Model 2 Tobit -2959,10 -2962,87 Log Likelihood Two-limit(i) Two-part(ii) Type-2 Tobit (iii) Type-2 Tobit (iv) -2900,07 -4970,67 -4963,34 -4963,34 -2903,92 -4984,34 -4978,10 -4977,71 Notes: (i) We exclude the η.γ% of the sample (111 observations) where logyield exceeds 8.βγ and which doesn‘t correspond well with the in sample-fitted values. (iii) A bivariate Heckman sample-selection model without exclusion restrictions. (iv) A Heckman bivariate sample-selection model with exclusion restrictions. By comparison (table 4.10), the log likelihood for the two-limit tobit model fits the data considerably better than both the simple tobit model and the two-part model as well as the Heckman sample selection models. As a useful guide to failure of homoskedasticity or normality, comparing the Tobit estimates with Powell‟s estimator is done in Columns (4), where we calculates Powell's (1984) censored least absolute deviations estimator (CLAD) and bootstrap estimates of its sampling variance.122 100 121 Cotton productivity in levels is very heavily skewed and has considerable nonnormal kurtosis. The logarithmic transformation reduced both the skewness and nonnormal kurtosis significantly. 122 We choose the bootstrap estimate which assumes that the sample was selected in two-stages and which replicates the design by bootstrapping in two stages. An advantage of the two-stage bootstrap estimates is that if the sample was collected using a two-stage process, then the estimated standard errors will be robust to this design effect. Kish (1995) and Cochran (1997) show the importance of correcting mean values for design effects. Scott and Holt (1982) show that the magnitude of the bias for the estimated variance-covariance matrix for OLS estimates can be quite large when it is erroneously assumed that the data were collected using a simple random sample, if in fact a two-stage design had been used. 125 bootstrap replications are performed.123 Unlike the standard estimators of the censored regression model such as tobit or other maximum likelihood approaches, the CLAD estimator is robust to heteroscedasticity and is consistent and asymptotically normal for a wide class of error distributions. Due to insufficient observations and non achievement of convergence the final specification omits some of the covariates: Sex; stratum; livestock and rainfall. Consequently, the treatment in either model is not significant. In Table 4.11 reports our benchmark results. Columns (1) and (4) report estimates of equation (4.10), that is, a simple model of cotton yields per hectare that does not control for the unobservables (It, t, ϕt). There is evidence in both models in favor of decreasing returns to scale in cotton since there is a negative association between the size of land allocated to cotton and cotton yields. There is also negative association between farm size and cotton productivity and production. Ageν household size and rainfall don‘t seem to matter. The dynamics of cotton yields are closely linked to the market accessibility through the EPFRP treatment. In both the simple models of cotton yields, the estimated magnitudes are both significant and larger than the other covariates except cotton land share. Columns (2) and (5) report productivity results and columns (3) and (6) the production results both from equation (4.12), controlling for agricultural effects (It) and unobserved heterogeneity ( t). In model 1 the estimated magnitude of the impact of the EPFRP treatment is likewise significant but the coefficient is both larger and its sign is correct in column 2. In model 2 it is only the sign that is correct column 5. Using the logarithm of cotton production instead of productivity as the dependent variable doesn‘t seem to improve the results in both model specifications. 123 Rogers (1993) shows that these standard errors are not robust to violations of homoscedasticity or independence of the residuals and proposes a bootstrap alternative. 126 Table 4.13 Cotton Yields: Impacts of the EPFRP Baseline Regressions Model 1 Simple Model Controlling for I and η Controlling for I and η (1) (2) (3) (4) (5) (6) logyield lnY lnY_CM logyield lnY lnY_CM Age of Head of HH Age Squared Model 2 Simple Model 0.0070 0.0187 0.0201 0.0057 0.0193 0.0200 (0.0086) (0.0131) (0.0132) (0.0086) (0.0131) (0.0132) -0.0001 -0.0002 -0.0002* -0.0001 -0.0002* -0.0002* (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) Sex 0.1019* 0.0163 0.0479 0.1024* 0.0035 0.0489 (M=1; F=0) (0.0568) (0.0835) (0.0855) (0.0564) (0.0841) (0.0856) Hous ehold Size Share of Males in HH 0.0027 -0.0070 -0.0059 0.0021 -0.0059 -0.0060 (0.0077) (0.0104) (0.0109) (0.0077) (0.0105) (0.0109) -0.4241** -0.0847 -0.1635 -0.4234** -0.0975 -0.1487 (0.1193) (0.1842) (0.1847) (0.1192) (0.1849) (0.1848) Farm Type (Stratum) -0.1097* -0.1694* -0.1522* -0.1104* -0.1668* -0.1522* (SSF=1; MSF=2) (0.0590) (0.0898) (0.0848) (0.0590) (0.0899) (0.0848) Lives tock 0.0790* -0.0044 -0.0763 0.0760* 0.0087 -0.0773 (0.0420) (0.0637) (0.0618) (0.0419) (0.0642) (0.0620) Rainfall Dis trict Level 0.0001 -0.0003* -0.0006*** 0.0000 -0.0002 (0.0001) (0.0002) (0.0002) (0.0001) (0.0002) EPFRP Treatment (T=1; NT=0) -0.1759*** 0.2887*** -0.0185 (0.0417) (0.0639) (0.0585) Cotton Landfraction -1.3377*** (0.0002) -1.3571*** (0.1032) (0.1026) EPFRP Treatment (s hare) Cons tant -0.0006*** -0.0094*** 0.0085*** -0.0005 (0.0018) (0.0027) (0.0026) 7.0440*** -0.4795 -0.1622 7.1622*** -0.5260 -0.1607 (0.2359) (0.3556) (0.3645) (0.2381) (0.3573) (0.3645) Obs ervations 2364 2221 1909 2364 2221 R2 0.0818 0.0148 0.0173 0.0854 0.0096 Sourceμ Authors‘ estimations. 127 1909 0.0173 In table 4.14, we report the results with the entry and exit correction, thereby taking into account the compositional effects induced by entry and exit into cotton farming (Brambilla and Porto, 2007). Column (1) reproduces the DID estimates from column (3) of Table 4.9, which does not include controls for ϕ. Columns (2) use a linear OLS model; columns (3) use a Tobit model to estimate the selection equation, and columns (4) use a CLAD model. Model 1 and Model 2 are the same as before. We find that in both models, the Tobit model outperforms the other models with regards to the magnitude and sign of the EθFRθ coefficient (α). In all our specifications in columns (2) to (4) of Table 4.14, the estimates of b0 are similar to those from the DID model that does not correct for ϕ column (1). Table 4.14: Comparison of Cotton Productivity Estimation Models Model 1 Age of Head of HH Age Squared Model 2 (1) (2) (3) (4) (1) (2) (3) (4) DID OLS Tobit CLAD DID OLS Tobit CLAD 0.1921*** 0.0070 0.0230 0.0340 -0.1091*** 0.0054 0.0225 0.0322 (0.0085) (0.0179) (0.0295) (0.0247) (0.0086) (0.0178) (0.0297) -0.0001 -0.0003 -0.0003 0.0011*** -0.0001 -0.0003 -0.0003 (0.0003) (0.0001) (0.0002) (0.0003) 0.1026* -0.1140 -0.1945 (0.1973) (0.0719) -0.0020*** (0.0001) (0.0002) (0.0003) Sex 0.1017* -0.1005 -0.1475 (M=1; F=0) (0.0568) (0.1227) (0.1968) (0.0565) (0.1232) Household Size 0.0027 -0.0269* -0.0145 -0.0254* 0.0020 -0.0254* -0.0096 -0.0935* (0.0077) (0.0143) (0.0199) (0.0137) (0.0077) (0.0143) (0.0200) (0.0514) Share of Males in HH (0.0007) -0.0846 -0.6342** -0.5892 -1.1998*** -0.0989 -0.6285** -0.5734 -0.1354 (0.1193) (0.2645) (0.3855) (0.3475) (0.1190) (0.2651) (0.3880) (0.8756) Farm Type (Stratum) -0.1108* -0.3650*** -0.2274 -0.1126* -0.3646*** -0.2301 (SSF=1; MSF=2) (0.0590) (0.1183) (0.1640) (0.0589) (0.1183) (0.1651) Livestock Rainfall District Level 0.0784* -0.0821 0.0932 0.0738* -0.0719 0.1264 (0.0421) (0.0856) (0.1432) (0.0419) (0.0860) (0.1436) 0.0001 -0.0010*** -0.0012** -0.0008* 0.0000 -0.0010*** -0.0011** 0.0005 (0.0001) (0.0003) (0.0005) (0.0004) (0.0001) (0.0004) (0.0005) (0.0011) EPFRP Treatment (T=1; NT=0) -0.1605** 0.2317*** 0.6506*** 0.1105 (0.0760) (0.0895) (0.1350) (0.1241) Cotton Landfraction -1.3408*** -1.3625*** (0.1053) (0.1049) D2 D2*Treatment -0.0192 -0.0394 (0.0774) (0.0777) 0.0000 0.0598 (0.0000) (0.1090) Cotton Landfraction -3.8872*** -5.1098*** -7.5985*** (Residualsμ ϕ) (0.2281) (0.4106) EPFRP Treatment (share) Constant -3.8924*** -5.1328*** -14.9180*** (0.8439) (0.2289) (0.4136) -0.0105*** 0.0090** 0.0239*** (2.0747) 0.0094 (0.0035) (0.0045) (0.0068) (0.0133) 7.0582*** 1.6479*** 0.7817 4.4438*** 7.2007*** 1.6462*** 0.8296 -3.4324** (0.2364) (0.5302) (0.8603) (0.7830) (0.2423) (0.5378) (0.8819) (1.6898) Observations 2364 1357 1357 219 2364 1357 1357 R2 0.0818 0.1962 0.0855 0.1945 170 Notes: (1) Difference-in-Differences Estimator; (2) OLS: Log-linear models for which the parameters need to be interpreted as semielasticities. (3) Tobit regression. (4) Clad calculates Powell's (1984) censored least absolute deviations estimator (CLAD) and bootstrap estimates of its sampling variance. Sourceμ Authors‘ estimations. 128 Table 4.15: Cotton Productivity Non-Linearity of Unobserved Productivity Tobit (1) Age of Head of HH Age Squared Model 1 PLREG (2) SPF (3) Tobit (1) Model 2 PLREG (2) SPF (3) 0.0340 0.0395* 0.0189 0.0322 0.0378 0.0182 (0.0295) (0.0231) (0.0178) (0.0297) (0.0232) (0.0177) -0.0003 -0.0004* -0.0002 -0.0003 -0.0004* -0.0002 (0.0003) (0.0002) (0.0002) (0.0003) (0.0002) (0.0002) Sex -0.1475 -0.2063 -0.2528* -0.1945 -0.2229 -0.2654** (M=1; F=0) (0.1968) (0.1553) (0.1312) (0.1973) (0.1553) (0.1330) Hous ehold Size -0.0145 -0.0318* -0.0136 -0.0096 -0.0309* -0.0126 (0.0199) (0.0164) (0.0171) (0.0200) (0.0164) (0.0172) Share of Males in HH -0.5892 -0.6170** -0.3151 -0.5734 -0.6207** -0.3213 (0.3855) (0.3074) (0.3019) (0.3880) (0.3079) (0.3024) Farm Type (Stratum) -0.2274 -0.3817*** -0.3578** -0.2301 -0.3847*** (SSF=1; MSF=2) (0.1640) (0.1299) (0.1440) (0.1651) (0.1301) (0.1437) Lives tock 0.0932 -0.1223 -0.1767* 0.1264 -0.1072 -0.1701* (0.1112) (0.1005) (0.1111) (0.1005) Rainfall Dis trict Level -0.0012** (0.0005) (0.0004) (0.0004) EPFRP Treatment (T=1; NT=0) 0.6506*** 0.2609** 0.2412** (0.1350) (0.1092) Cotton Landfraction -5.1098*** (Res iduals μ ϕ) (0.4106) (0.1432) -0.0011*** -0.0012*** Obs ervations (0.0005) (0.2588) 0.0087 0.0100* (0.0068) (0.0055) (0.0052) 0.8296 (0.6092) (0.8819) 1357 1357 1356 0.0551 ll -1234.1097 -2480.5744 (0.2595) 0.0239*** 1.7468*** F (0.0004) -3.5440*** (0.4136) 0.7817 R2 (0.0004) (0.1048) (0.8603) 1357 -0.0011*** -0.0012*** -3.5341*** -5.1328*** EPFRP Treatment (s hare) Cons tant (0.1436) -0.0011** -0.3543** 1.7440*** (0.6242) 1356 1357 0.0517 -3.322e+05 -1239.4838 8.7259 -2482.9949 -3.324e+05 8.1615 Notes: (i) Simple Tobit model in levels; (ii) Partial Linear regression model with Yatchew's weighting matrix; (iii) Stochastic frontier model: Cobb-Douglas production function with half-normal distribution for inefficiency term. Sourceμ Authors‘ estimations. Columns 2 and 5 in Table 4.13 reports the Linear partial regression (plreg) estimates following the semiparametric regression model by the method of differencing: (5.9) yit f(z) + xit‘ + uit, = where f(z) is a smooth function with bounded first derivatives,124 the function f is known to lie in a particular parametric family, x are control variables that enter (5.9) linearily, u is the zero-mean innovation error, and is a vector of parameters. Standard errors of are adjusted according to Yatchew(1998)‘s method. Columns 3 and 6 show the fitted results of the stochastic production frontier models. It provides estimators for the parameters of a linear model with a disturbance that is assumed to be a mixture of two components, which have a strictly non-negative and symmetric distribution, respectively.125 Again the Tobit Model outperforms these two other non-linear models. 124 The smoothed values of f are estimated by the locally weighted regression using lowess. Frontier can fit models in which the non-negative distribution component (a measurement of inefficiency) is assumed to be from a half-normal, exponential, or truncated-normal distribution. 125 129 4.5.3. Robustness Checks 4.5.3.1. Tests of the Matching Assumption and Sensitivity of Estimates In this section we test the matching assumption and examine the sensitivity of our estimates to the specification of the propensity score. In table 4.14 we first calculate several measures of the balancing of the independent variables before and after matching. For each regressor it calculates (a) t-tests for equality of means in the treated and non-treated groups, both before and after matching. T-tests are based on a regression of the variable on a treatment indicator. Before matching this is an unweighted regression on the whole sample, after matching the regression is weighted using the matching weight variable and based on the on-support sample. (b) The standardised bias before and after matching, together with the achieved percentage reduction in abs(bias).126 The standardised bias is the difference of the sample means in the treated and non-treated (full or matched) sub-samples as a percentage of the square root of the average of the sample variances in the treated and non-treated groups (Leuven and Sianesi, 2003).127 Table 4.16: Covariate imbalance testing Mean Variable Sample Age Unmatched Matched Agesq Unmatched Matched Sex Unmatched Matched loghhsize Unmatched Matched stratum Unmatched Matched livestock Unmatched Matched Areapc Unmatched Matched Clandfrac Unmatched Matched Treated 41.381 40.607 1845 1777 1.2386 1.2169 1.6963 1.7106 1.1443 1.1572 .51474 .54003 .36411 .38772 .09763 .12518 Control 41.527 39.982 1859.9 1726.5 1.2319 1.1847 1.6342 1.7673 1.1397 1.2358 .4726 .66925 .34372 .48696 .12224 .31539 %bias -1.3 5.4 -1.5 5.0 1.6 7.6 11.5 -10.5 1.3 -22.5 8.4 -25.9 5.9 -28.8 -12.7 -97.8 %reduct bias t-test t p>t -0.51 0.607 -325.6 -4.29 0.000 -0.60 0.550 -239.5 -5.17 0.000 0.63 0.527 -385.9 -3.63 0.000 4.66 0.000 8.6 -4.12 0.000 0.52 0.602 -1637.3 -3.22 0.001 3.42 0.001 -206.6 -11.85 0.000 2.34 0.019 -386.8 -2.11 0.035 -5.10 0.000 -672.8 -1.21 0.228 Sourceμ Authors‘ estimations. P-values for majority of variables (7 out of 8) are significant at 0.05 level which means there is a significant difference between the two groups with respect to these characteristics. A direct comparison between two groups based on this study population would lead to biased and invalid inference if the outcome measure is related to any of these baseline characteristics. 126 Percentage bias reduction is calculated by (1- Di)/Dj *100% where Di and Dj are group difference in covariates means after matching and before matching, respectively. 127 The stata formula is taken from Rosenbaum and Rubin, 1985. 130 As we can see from table 4.16, the percentage of bias reduction for covariates varies from 1637% to 8.6% after matching. Difference between the matched pairs is evaluated using T-test for continuous variables and Chi-Square test for categorical ones. For none variables are there a significant difference between two matched groups, which shows that the two matched groups are well balanced. An estimate of the propensity score is not enough to estimate the ATT of interest. The reason is that the probability of observing two units with exactly the same value of the propensity score is in principle zero since p(X) is a continuous variable. Various methods e.g.: Nearest Neighbor Matching, Radius Matching, Kernel Matching etc. have been used in table 4.8a above to overcome this problem. It is clear from considerations in Becker and Ichino(2002) that these methods reach different points on the frontier of the tradeoff between quality and quantity of the matches and none of them is a priori superior to the others. However, their joint consideration in table 4.8a offers a way to assess the robustness of the estimates.128 Table 4.17a: The Baseline ATT estimation (with no simulated confounder) Outcome Type Logyield Continuous LYIELDDV Dummy ATT estimation with Nearest Neighbor Matching method Nearest Neighbor Matching method Number of (i) Treated Controls ATT Analytical Std.Err. t 3870 516 -0.114 0.054 -2.127 3870 516 -0.037 0.028 -1.355 Notes: Random draw version; (i) The numbers of treated and controls refer to actual nearest neighbour matches. Nannicini(2007b) and Ichino et al.,(2007) offers another way to implement the sensitivity analysis for matching estimators. This analysis builds on Rosenbaum and Rubin(1983a) and Rosenbaum(1987) and simulates a potential binary confounder to assess the robustness of the estimated treatment effects with respect to specific deviations from the conditional independence assumption (CIA). As a first simple step table 4.17a shows the estimation of the baseline average treatment effect on the treated (ATT) by using the ‗ζearest neighbor‘ propensity-score matching estimator instead of ‗radius‘, and ‗kernel‘ Becker and Ichino(2002). The propensity-score specification 128 BECKER, S. & ICHINO, A. 2002. Estimation of average treatment effects based on Propensity Scores. Stata Journal, 2, 358-377. also note that with all these methods the quality of the matches may be improved by imposing the common support restriction. 131 includes the same variables as in table 4.16 except ‗Areapc‘ replaced by ‗share of males in household.‘ The baseline ATT point neighbor matching estimate is equal to -0.114 (with an analytical standard error of 0.054). Nannicini(2007b) argues that the fact that we can compare the non-experimental estimates with this unbiased benchmark makes the sensitivity analysis useless. But following Nannicini(2007b) we assume that this is not the case, and we would like to assess the robustness of the above matching estimate in table 4.17a. As a second step, we successively simulates a couple of potential binary confounders (U) in the data, on the basis of four parameters: pij (with i,j=0,1). We define Y = LYIELDDV as a binary transformation of the outcome for the continuous outcomes (logyield) (with Y = 1 if logyield > meanlogyield, i.e. the 50 centile; or 0 if logyield < meanlogyield). T is defined as the binary EPFRP treatment, we find that each simulation parameter pij represents the probability that U=1 if T=i and Y=j. We specify the four parameters (with the options: p11(0.6) p10(0.5) p01(0.5), and p00(0.2)). We also let the distribution of U mimic the distribution of two relevant covariates (respectively Male and Rain). Finally, U is considered as any other covariate and is included in the set of matching variables used to estimate the propensity score and the ATT. The imputation of U and the ATT estimation are replicated 100 times, and a simulated ATT is retrieved as an average of the ATTs over the distribution of U in table 4.15b as well as the outcome and selection effects of U. This estimate is robust to the specific failure of the CIA implied by the parameters pij. A comparison of the simulated ATT and the baseline ATT tells us to what extent the latter is robust, with respect to the specific deviation from the CIA that we are assuming. The simulated ATT for the continuous outcome (ranging from -0.121 to -0.128) and the binary transformation (-0.071) is greater that the baseline estimates of respectively -0.114 and -0.037. To further emphasize the characteristics of the failure of the CIA implied by the simulated confounder (i.e., by the chosen pij): The estimated effect of U on the selection into treatment selection effect - and the estimated effect of U on the outcome of untreated subjects - outcome effect - are also reported as odds ratios in table 4.17b. The selection effect of the confounder is larger than the outcome effect when the binary variable (Male or Rain) is used to simulate the confounder. The opposite is the case when we directly specify the parameters pij. 132 Table 4.17b: ATT estimation with simulated confounder Potential Binary variable used to simulate the conOutcome founder Centile (U) (Y) 1 Binary Male 50 2 Binary Male 25 3 Binary Male 75 4 Continuous Male 50 5 Continuous Male 25 6 Continuous Male 75 7 Continuous 50 8 Continuous 25 9 Continuous 75 10 Continuous Rain 50 11 Continuous Rain 25 12 Continuous Rain 75 13 Continuous Young 50 14 Continuous Young 25 15 Continuous Young 75 ATT estimation with Nearest Neighbor Matching method Nearest Neighbor Matching method Nearest Neighbor Matching method Nearest Neighbor Matching method Nearest Neighbor Matching method Nearest Neighbor Matching method Nearest Neighbor Matching method Nearest Neighbor Matching method Nearest Neighbor Matching method Nearest Neighbor Matching method Nearest Neighbor Matching method Nearest Neighbor Matching method Nearest Neighbor Matching method Nearest Neighbor Matching method Nearest Neighbor Matching method ATT General multipleimputation Effect Std.Err. Outcome Selection ATT Withinimputation Effect Std.Err. Outcome Selection ATT The probabili ty of having U=1 if T=1 and Y=1 (p11) Betweenis equal imputation Effect Std.Err. Outcome Selection to: The probabili ty of having U=1 if T=1 and Y=0 (p10) is equal to: The probabili ty of having U=1 if T=0 and Y=1 (p01) is equal to: The probabili ty of having U=1 if T=0 and Y=0 (p00) is equal to: The probabili ty of having U=1 if T=1 (p1.) is equal to: The probabili ty of having U=1 if T=0 (p0.) is equal to: Confoun der U account for % of baseline estimate -0,071 0,035 3,536 0,606 -0,07 0,03 3,454 0,627 -0,074 0,018 3,051 0,578 0.98 0.99 0.99 0.99 0.99 0.99 -94% -0,07 0,034 2,728 0,664 -0,073 0,031 4,721 0,614 -0,068 0,018 2,876 0,619 0.98 0.99 0.99 0.99 0.99 0.99 -90% -0,07 0,034 4,861 0,677 -0,071 0,031 2,615 0,595 -0,073 0,017 5,128 0,553 0.98 0.99 0.99 0.99 0.99 0.99 -93% -0,123 0,063 0,226 1,228 -0,121 0,054 0,204 1,217 -0,122 0,039 0,209 1,195 0.96 0.99 0.95 0.99 0.97 0.96 -7% -0,121 0,063 0,248 1,195 -0,121 0,055 0,227 1,144 -0,119 0,034 0,217 1,208 0.96 0.99 0.97 0.96 -6% -0,125 0,064 0,187 1,262 -0,12 0,054 0,15 1,228 -0,122 0,034 0,169 1,242 0.96 0.99 0.95 0.99 0.97 0.96 -7% -0,198 0,071 4,101 1,657 -0,206 0,056 4,442 1,654 -0,207 0,037 4,49 1,659 0.60 0.50 0.50 0.20 0.58 0.45 -79% -0,187 0,069 4,471 1,598 -0,184 0,055 4,395 1,609 -0,186 0,037 4,521 1,585 0.60 0.50 0.50 0.20 0.59 0.47 -63% -0,201 0,069 4,397 1,778 -0,207 0,057 4,38 1,755 -0,195 0,042 4,372 1,761 0.60 0.50 0.50 0.20 0.58 0.43 -76% -0,121 0,069 0,499 1,398 -0,124 0,055 0,464 1,399 -0,122 0,04 0,485 1,401 0.49 0.40 0.39 0.56 0.48 0.40 -7% -0,127 0,074 0,487 1,414 -0,123 0,055 0,465 1,383 -0,12 0,042 0,481 1,404 0.49 0.40 0.39 0.56 0.48 0.40 -8% -0,128 0,071 0,725 1,404 -0,134 0,055 0,736 1,406 -0,134 0,046 0,742 1,385 0.49 0.40 0.39 0.56 0.48 0.40 -16% -0,128 0,071 0,962 1,069 -0,135 0,054 0,927 1,065 -0,129 0,04 0,93 1,068 0.23 0.26 0.22 0.25 0.24 0.23 -15% -0,136 0,069 1,131 1,07 -0,128 0,055 1,127 1,064 -0,138 0,042 1,148 1,066 0.24 0.24 0.23 0.21 0.24 0.23 -18% -0,131 0,068 0,883 1,074 -0,133 0,055 0,87 1,073 -0,134 0,038 0,832 1,071 0,23 0,26 0,22 0,25 0,24 0,23 -16% Sourceμ Authors‘ estimations based on (Nannicini, 2007b); (Ichino et al., 2007). 133 1.00 0.96 The simulated ATTs are in all cases bigger than the baseline ATT. However, the biggest ATT are when the outcome is continuous and we directly specify the parameters pij. In this case the potential confounder ―kills‖ by a large amount (63%-79%) the baseline estimate. In other terms, the sensitivity analysis is telling us that the existence of a confounder U behaving like: The Male dummy; Rain dummy; or Youth dummy might account for respectively: 6-7%; 7-16%; and 15-18% of the baseline estimate. These sensitivity conclusions depends on the fact that the U is simulated on the basis of a continuous outcome variable, since the account is much lower than if the outcome U is simulated on the basis of the binary transformation of Y that uses either the 25, the 50, or the 75 centile of the outcome (table 4.17b). The above simulations convey an image of robustness of the nearest neighbor matching estimate equal to -0,114. This image, however, might be produced by the particular characteristics of the covariates used to simulate U (Male; Rain and Youth), rather than by the fact that the baseline ATT is robust to possible deviations from the CIA (Nannicini, 2007b). 4.5.3.2. Specification tests and model diagnostic Given the fact that the EθFRθ wasn‘t implemented simultaneously in all five treatment districts, it may be difficult to justify assigning 1998 as the year where the entire project was implemented. To examine the robustness of the results, we re-estimate the model by first redefining the post-implementation phase to exclude the year 1998 and thus move it to the pre-implementation phase. Thus, we assign the first three years and the three years to two different phases of the EPFRP, where the first period is considered the pre-implementation phase, whereas the second period is considered the (post-) implementation phase. Thus in column 1 in table 4.20 we insert the estimates from the Tobit model in column 3 from table 5.6 above. In columns (2) and (4) we show the Tobit results for the same definition of the treatment variable but with a readjusted implementation phase starting in 1999. This reassignment of the start year raises the magnitude of the treatment variable‘s coefficient (α) to an even higher level in both models. When looking at the district and year effects in model 3 we only find that the 1999 year effect is significant at the 1% level, whereas year 2000 and 2001 are dropped due to collinearity. Finally, by looking at the EθFRθ‘s percentage share of the primary feeder roads (column 7) instead of as a share of the entire feeder road network (column 3) we get more or less the same results, although a bit inferior with regards to the magnitude of the coefficients. 134 Table 4.18: Sensitivity to the Definition of EPFRP and Reassignment of Implementation Year Tobit Model 1 (1) EPFRP (T=1; NT=0) (Primo: 1998) Cotton Landfraction (Residualsμ ϕ) Model 2 (2) (3) Model 3 (4) (5) Model 4 (6) (0.1350) -5.1098*** -5.2231*** -5.1328*** -5.1719*** -5.3043*** -5.3917*** -5.1071*** (0.4106) (0.3981) epfrp (T=1; NT=0) 1.6017*** (Primo: 1999) (0.1697) (0.4136) (0.3989) EPFRPpct (share) (i) 0.0239*** (Primo: 1998) (0.0068) epfrppct (share) 0.0751*** (Primo: 1999) (0.0085) (0.3984) D1998 0.0904 (1998 year effect) (0.1499) D1999 1.7143*** (1999 year effect) (0.1716) (0.4179) D301 0.1515 (District 301 effect) (0.8124) D302 0.1127 (District 302 effect) (0.8463) D303 0.5846 (District 303 effect) (0.7658) D304 0.6730 (District 304 effect) (0.7660) D305 -0.5912 (District 305 effect) (0.8016) D306 0.4568 (District 306 effect) (0.7995) D308 -0.1548 (District 308 effect) (0.7800) EPFRPPCT (share) (ii) Observations ll (0.4105) 0.0225*** (Primo: 1998) Constant (7) 0.6506*** (0.0047) 0.7817 -0.3277 0.8296 -0.1387 -0.7530 2.2258* 0.6497 (0.8603) (0.8287) (0.8819) (0.8326) (0.8378) (1.2171) (0.8679) 1357 1357 1357 1357 1357 1357 1357 -1234.1097 -1201.8225 -1239.4838 -1207.9393 -1191.7112 -1225.7072 -1234.2839 Notes: Standard errors in parentheses * p<0.1, ** p<0.05, *** p<0.01. 135 4.6. Conclusions and Policy Implications This chapter has investigated the dynamic impacts at the district level of the implementation of the EPFRP on farm cotton yield and production in Zambia‘s rural Eastern Province exclusively, contrary to Brambilla and Porto(2005, 2007) who cover all four cotton growing provinces at the provincial level. The key development objective of the EPFRP in our context was „to improve access to the productive areas of the Province to ensure the introduction of farm inputs as well as the timely evacuation of harvested agricultural product.‟ In addition to the short-run effects such as the creation of direct employment because of the use of laborbased technology, the EPFRP also had medium to long-run effects through reallocation of incumbent firms, entry and exit of firms, market creation and destruction (for inputs, outputs and credit), and contract enforcement mechanisms (i.e. the out-grower scheme). To explore these market dynamics of the EPFRP we identify two phases of the EPFRP. Starting with a baseline period from July 1996 to 1997/1998, which in our study is the pre-implementation period, where the project focused on building the capacity of the districts and the private contractors to rehabilitate and maintain rural feeder roads using labor-based techniques.129 The second (post-) implementation phase ran from 1998/1999 to 2001/2002. Only five districts (Chadiza, Chipata, Katete, Lundazi and Petauke) received capital assistance to carry out rehabilitation works. These five districts are considered our treatment districts. Moreover, at the mid-term evaluation carried out in July 1998 only 76 km of feeder road of the final 404 km had actually been rehabilitated to an excellent standard and had been maintained for one year, and another 196 km had received separate routine maintenance interventions.130 By following the Brambilla and Porto(2005, 2007) approach we have estimated the dynamic impacts of the implementation of the EPFRP by building a model of cotton 129 The project succeeded in training seven private, labor-based rehabilitation contractors; 15 private, laborbased maintenance contractors and 19 public officials—14 district supervisors. 130 Considering that not all essential inputs have been in place at any one time (national staff, the finance company, office accommodation, supporting projects). 136 yields and crop choices. We have compared average cotton yields across two phases, the pre-implementation phase and the post-implementation phase of the EPFRP, conditional on the aggregate trend in agricultural, observed covariates at the farm level, and unobserved farm effects like land quality and cropping ability. To correct for compositional effects associated with entry and exit into cotton farming and with cotton-specific unobserved heterogeneity, we have introduced a model of selection into cotton that provides proxies for unobserved cotton productivity. Adapting techniques from the industrial organization literature, these proxies are given by land cotton shares (i.e., the shares of total land allocated to cotton) purged of the effects of observed covariates. Our model thus provides an overall consistent estimator of the impacts of the EPFRP in the presence of confounding observed and unobserved effects and in the presence of compositional effects in cotton farming. Unlike Brambilla and Porto(2007) who besides Eastern Province also include Central, Southern and Lusaka Provinces in their sample, we don‘t find an increase in cotton productivity in the last two years 2000/2001 to 2001/2002, but rather a stagnation after a sharp fall from 1999/2000 to 2000/2001. This is because as they likewise acknowledge that there are significant differences across these four provinces. These provincial differences could be due to differences in the out-grower schemes offered by different firms. This piece of information unfortunately isn‘t captured by the Post Harvest Surveys. We also find some evidence indicating that cotton yields followed different patterns in the treatment districts compared with the control districts ( figure A2b). Finally, notwithstanding the sensitivity of our results to the choice of model specification, we do find that the EPFRP treatment had an effect that was significant. In other words, it does seem as though the EθFRθ‘s improvement and maintenance of the road network in Eastern Province may have contributed to the stimulation of the production of cotton (MT) and the yield (MT/HA). The concern is that the yield improvements weren‘t sustained beyond the short-term perhaps due to idiosyncratic factors in Eastern Province not captured by the models used. 137 In line with Brambilla and Porto(2007) we derive a number of lessons from our empirical analysis. First, in 2005, most cotton production in Zambia was carried out under the outgrower scheme. Farmers and firms have understood the importance of honoring contracts and the benefits of maintaining a good reputation. The out-grower programs have been perfected and there are now two systems utilized by different firms: the Farmer Group System and the Farmer Distributor System. Both systems seem to work well (Balat, 2005; Brambilla, 2005; Poulton et al., 2004; Tschirley et al., 2006). Second, our results generate some evidence on how labour-based technology rural transport infrastructure projects at the regional level in combination with other effects such as the persistent effects from the prior liberalization of the agricultural sector affects yields at the farm level via input and output prices, credit, input use, technical advice, information and technology, and efficiency. In other words, unlike Brambilla and Porto(2007) who exclusively associate the impact with the economic reform effect, we believe that the agricultural trade liberalization and the privatization of the parastatals wouldn‘t have had a significant impact on the crop choice and the crop productivity if the EθFRθ hadn‘t subsequently been implemented in Eastern θrovince as an indispensable complementary policy for rural development. Finally, Schulz and Bentall(1998) already noted in the mid-term evaluation of the EPFRP that the rate of implementation of the decentralization policy had a negative impact on the sustainability of the project results. Budget constraints forced the District Councils in the Eastern Province not to continue to carry out rehabilitation tasks after the project was completed and instead to concentrate on maintenance unfortunately not always on a routinely basis. Moreover, in the end only a total of 404km of feeder roads of the targeted 580km were completed within the project budget, which most likely diminished the EθFRθ‘s impact on the cotton yield and production. 138 Chapter 5: Regional Analysis of Eastern Province Feeder Road Project. District level estimation of the Poverty Alleviation Effects of Rural Roads Improvements in Zambia‟s Eastern Province 139 Felix qui potuit rerum cognoscere causas.131 Virgil. 5.1. Introduction Despite the relative abundance of data and analysis in Zambia there is still much that needs to be understood about policies that promote rural growth and poverty reduction. More specifically, trade facilitated and augmented due to rural transport infrastructure improvements introduces new opportunities and new hazards. Households are affected both as consumers and as producers or income earners (see chapter 4). As consumers, which is the focus of this chapter, households are affected when there are changes in the prices of goods consumed by the family (Winters, 2002b, Balat and Porto, 2005b). Hence, targeting aid in the form of transport infrastructure development to poor areas has been an important vehicle for development assistance (Chen et al., 2006). However, assessing the returns to the EPFRP as we have seen in the previous chapter has proven quite problematic. Not only is its contribution to local agricultural production (Chapter 4) and trade (Chapter 8) difficult to measure and to attribute, it is difficult assessing precisely who benefits from these community-level assets. Often, any indicators of impact on the poor are indirect (Devereux, 2002) as we shall explore in this chapter.132 In other words, chapter 5 endeavours to assess whether the EPFRP had any welfare impacts on the district levels both within the disbursement period (12th of June 1996 and 31st of December 2001) and beyond that period. The attempt to identify to what extent the EPFRP contributed or constrained propoor rural consumption growth in five out of eight districts constituting Eastern Province is complicated by confounding factors such as the εεD Government‘s reform-based policies, which are difficult to separate from other policies more directly aimed at and typically associated with generating growth and poverty-reduction. In many respects the 131 Happy he, who could understand the causes of things. Georgics: Book 2, Line 490. Rural road networks are needed not only for transporting passengers and commodities, but also for market integration, which reduces price seasonality and enhances food security. 132 140 limitation of chapter 5, is that it can only attempt to explain whether certain households within the catchment districts of these rehabilitated feeder roads coped better than the non-beneficiary households in the counterfactual districts. Following Chen et al.(2006) we must as we also did in chapter 4 deal with the selective geographic placement of the EPFRP, whereby it was targeted to areas with particular agricultural potential and the resulting selection bias is unlikely to be timeinvariant. There is also a concern about (time-varying) selection bias due to spill-over effects to non-participating areas (control districts). Some local spillover effects e.g. arising from the spending responses from the ZMK2,065 billion (US$480,233) that was paid in wages within the concerned districts of Eastern Province through the achievement of 870,000 workdays of the entire project are also expected to occur. It is rare to assess public works project impacts by repeated panel observations over a relatively long period. A few recent exceptions include e.g. Dercon and Hoddinott(2005), Chen et al.,(2006), Dercon et al.,(2007), Mu and van de Walle,(2007). In Zambia much of the data required is readily available from the Living Conditions Monitoring Surveys (LCMS) conducted by Zambia's CSO on a regular basis.133 Unfortunately, the LCMS dataset is only constituted by successive independent crosssections collected in 1996, 1998, 2003, 2004, and 2006. In this chapter one class of semi-parametric models known as partially linear models is first applied, which is also available for estimating conditional quantile.134 There has been relatively little research done on the ability of semi-parametric models to fit datasets that are encountered in applications. Horowitz and Lee(2002), Yatchew(2005), Zhoua et al.,(2008), and He et al.,(2005) report the results of some investigations on this issue. On the whole, however, the usefulness of semi-parametric 133 Caveat: Readers should be particularly cautious concerning the validity of our inferences based upon the analysis of the LCMS IV 2004 dataset, which is constrained by lack of unique identifiers (HID) in the original dataset. This has made it inevitable for us carry out a number of data manipulation in order to transform that dataset into a new dataset with all the usual negative consequences associated with these operations. 134 Other classes are: Nonparametric additive models, and nonparametric additive models with interactions. 141 representations of real LCMS datasets remains hitherto largely unexplored. This approach is complemented by cohort type approach, where cohorts are defined by date of birth, to facilitate the identification of the gainers and losers from the EPFRP. The chapter is organized as follows. Section 5.2 presents the background and the socio-economic setting as well as describes the LCMS datasets. Section 5.3 describes the various estimation methods used. Regression model estimation results are presented in Section 5.4. Section 5.5 discusses the estimation results, and section 5.6 concludes. 142 5.2. Background and setting In this section we will exclusively focus on a few additional characteristics of Zambia's Eastern Province not mentioned above as well as the cross-sectional datasets that we will use in our analysis. The CSO carried out the second Living Condition Monitoring Survey (LCMS II) in November-December, 1998, whereas the only comparable LCMS IV in terms of the survey designs was conducted between October 2004 and January 2005 covering the whole country on a sample basis.135 As with the PHSs used in chapter 4 these LCMSs neither contain any panel element and are simply repeated independent cross-sections or one-spot (single interview), which make welfare measures imprecise both due to sampling and non-sampling errors (e.g. under - or overestimation of household incomes and expenditures). Table 5.1: Sample Allocation (Standard Enumeration Areas -Primary Sampling Units) Surveys LCMS IV** 2004 LCMS III** 2002/2003 LCMS II* 1998 LCMS I* 1996 Province Zambia Eas tern Zambia Eas tern Zambia Eas tern Zambia Eas tern Total (% of total SEA) 1048 (6.28% of 16,683) n.a. 520 (3.12% of 16,683) 60 820 (6.31% of 12,999) n.a. 610 (4.69% of 12,999) 68 Rural n.a. n.a. 326 50 492 n.a. 348 54 Urban n.a. n.a. 194 10 328 n.a. 262 14 Source: Author based upon metadata of the LCMSs. Notes: * The sampling frame developed from the 1990 census of population and housing. ** The sampling frame developed from the 2000 census of population and housing. Households had been selected using a two-stage stratified cluster sample design. In the first stage, a sample of SEA was selected within each stratum (centrality) according to the number allocated to that stratum. Selection had been done systematically with probability proportional to the number of households within each SEA as registered in the 1990 / 2000 Population Census.136 In the second stage in each 135 In order to have equal precision in the estimates in all the districts and at the same time take into account variation in the sizes of the district, the survey adopted the Square Root sample allocation method, (Lesli Kish, 1987). This approach offers a better compromise between equal and proportional allocation methods in terms of reliability of both combined and separate estimates. The allocation of the sample points (PSUs) to rural and urban strata was almost proportional. 136 Sample allocation was done using the PPS method. This entailed allocating the total sample (1048 / 820) proportionately to each province according to its population share. 143 selected SEA, households were listed and each eligible household was given a unique sampling serial number.137 Given the stratified surveys the ex-ante probability of being surveyed is not constant across households (i.e. the sample is not purely probabilistic) and some re-weighting must take place before obtaining population's estimates. Each household has to be re-weighted with the inverse of its probability of being sampled.138 Economic Activities in Rural Eastern Province In 63% of the sampled rural households, the principal activity of the household head is farming in 1998. However, in 2004 the overall share had lowered, while increasing for the upper quintile (table 5.2). Table 5.2: Principal Economic Activity of Household Head, Rural Areas, Numbers of Household Heads by Quintile of Consumption All 3% 3,01% 63,09% n.a. n.a. 1,16% In Wage Employment Running a Busines/Self Employed Farming, Fishing, Forestry Piecework Unpaid family workers Not working but looking for work/means to do business Not working and not looking for work/means to do business but available or wishing to do so 7,49% Full time student 14,98% Ful time at home/home duties (homemaker) 7,49% Retired 0,00% Too old to work 1,47% Other 5,10% Total 1295 1998 Poorest 20% Richest 20% 0,70% 22,92% 1,63% 6,25% 67,13% 35,42% n.a. n.a. n.a. n.a. 0,93% 0,00% 0,58% 13,59% 7,43% 0,00% 1,74% 6,16% 861 0,00% 20,83% 10,42% 0,00% 0,00% 0,00% 48 All 3,07% 1,20% 54,80% 0,93% 0,27% 0,40% 0,27% 26,20% 2,00% 0,07% 1,47% 9,33% 1500 2004 Poorest 20% Richest 20% 1,40% 6,82% 0,47% 1,62% 58,97% 53,57% 0,93% 0,65% 0,00% 0,32% 0,00% 0,32% 0,23% 25,17% 1,17% 0,00% 1,17% 10,49% 429 0,65% 24,68% 0,97% 0,00% 0,97% 9,42% 308 Source: Author's calculations. Material Assets The large majority of sampled rural households again fall within the bottom quintile. Of the assets listed, only residential building, radios, bicycles and basic farm tools such as a plough and crop sprayer are owned by more than 10 percent of the sampled rural households. Motorized vehicles are practically non-existent, instead 10% of the poorest rural households (those in the bottom quintile) own a scotch cart in 1998. 137 Sample selected from roster of household members was obtained from a responsible adult household member. This may lead to unequal weighting in order to account for household size. 138 Hence while making statistical inferences about the population we should account for survey design effects (see, Deaton, 1997). 144 Table 5.3: Percentage of Households in Rural Areas Owning Particular Assets by Quintile 1.1. Plough 1.2. Crop sprayer 1.3. Fishing Boat 1.4. Canoe Brazier / Mbaula 1.5. Fishing net 1.6. Bicycle 1.7. Motor cycle 1.8. Motor vehicle 1.9. Tractor 1.10. Television 1.11. Video player 1.12. Radio Grinding/Hammermill (powered) 1.13. Electric (and non-electric) iron 1.14. Refrigerator/Deep freezer 1.15. Telephone (including cellular phone) 1.16. Sewing/knitting machine 1.17. (Electric/gas) Stove/cooker 1.18. Non-residential building 1.19. Residential house/building 1.20. Scotch cart 1.21. Donkeys Oxens Total number of households in Eastern Province sample 1998 All Poorest 20% Richest 20% 18,73% 19,13% 10,20% 12,39% 12,71% 8,16% 0,15% 0,12% 0,00% 0,69% 0,24% 0,00% n.a. n.a. n.a. 3,13% 2,42% 2,04% 50,84% 51,09% 59,18% 1,30% 1,45% 2,04% 1,83% 1,94% 2,04% 0,38% 0,24% 0,00% 1,61% 1,21% 0,00% 0,84% 0,36% 0,00% 43,88% 43,58% 48,98% n.a. n.a. n.a. 1,15% 0,97% 2,04% 0,61% 0,24% 2,04% 0,08% 0,00% 0,00% 7,11% 6,17% 2,04% 1,53% 1,09% 2,04% 2,60% 2,18% 0,00% 85,17% 82,45% 97,96% 9,33% 9,93% 6,12% 0,31% 0,36% 0,00% n.a. n.a. n.a. 1308 826 49 All 25,05% 17,47% 0,00% 1,44% 36,57% 1,50% 56,42% 0,13% 1,57% 0,44% 5,32% 3,07% 50,53% 2,19% 19,54% 1,88% 1,57% 4,88% 1,69% 2,44% 87,98% 13,34% 1,63% 19,66% 1597 2004 Poorest 20% Richest 20% 23,49% 23,74% 17,67% 15,73% 0,00% 0,00% 0,67% 2,97% 28,41% 37,69% 2,46% 0,59% 51,68% 49,85% 0,00% 0,00% 0,67% 1,48% 0,00% 0,00% 2,01% 7,72% 0,22% 4,15% 48,10% 49,55% 2,46% 2,67% 11,63% 25,22% 0,45% 3,26% 0,67% 0,30% 2,68% 8,01% 0,67% 2,97% 2,91% 1,48% 91,50% 85,76% 9,84% 12,46% 2,68% 2,67% 11,63% 22,55% 447 337 Source: Author's calculations. If we take a closer look at the district level (Table A5.a-b) we find that the catchment districts fare much better than the control districts at both the lowest quintile as well as the two highest quintiles in terms of improvements to the assets base with regards to a much steeper increased ownership of: Ploughs, crop sprayer, bicycle and a scotch cart. However, the rural households in the catchment districts experienced a fall in the ownership of motor cycles and motor vehicles in the same period from 1998 to 2004, with this kind of asset ownership being practically non-existent in the control areas. Access to Infrastructure, Service and Community Assets: Distance to Markets In 1998, on average the rural households in Eastern Province had to travel almost 40 km to reach an agricultural input market, which sell fertilizer and seeds. This distance had fallen significantly to around 13 km in 2004, although with no noteworthy differences between rural households belonging to separate consumption quintiles in 2004 (table 5.4). 145 Table 5.4: Mean distance to services and Community Assets, by Household, Rural Areas Quintile of Provincial Distriction, 1998 Quintile of Provincial Distriction, 2004 Poorest 20% 2 3 4 Richest 20% All Poorest 20% 2 3 4 Richest 20% 20,4 20,4 52,0 n.a. 8,0 27,5 9,5 10,3 9,2 9,8 9,2 8,8 32,1 32,2 52,0 n.a. 10,0 27,5 13,1 12,9 11,8 13,0 12,3 15,5 3,3 3,3 4,0 n.a. n.a. 1,5 3,8 3,4 2,4 2,9 6,0 4,3 n.a. n.a. n.a. n.a. n.a. n.a. 2,3 1,7 2,1 1,7 2,8 3,1 All 1.1. Food Market 1.2. Post Office/postal agency 1.3. Primary School Distance to Low Basic School(1-4) Distance to Middle Basic School(1-7) n.a. n.a. Distance to Upper Basic School(1-9) n.a. n.a. Distance to High School n.a. n.a. 32,5 1.4. Secondary School 1.5. Health Facility (Health 15,9 post/Centre/Clinic/Hospital) 6,0 1.6. Hammermill 1.7. Input market (for seeds, fertilizer, 40,0 agricultural implements) 34,7 1.8. Police station/post 1.9. Bank 1.10. Public transport (road, or rail, or 8,6 water transport) n.a. n.a. n.a. 32,6 n.a. n.a. n.a. 52,0 n.a. 16,0 6,0 24,0 n.a. 0,0 n.a. 40,3 34,8 52,0 n.a. 52,0 n.a. 8,6 47,0 n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. 29,0 3,0 4,3 28,8 16,4 2,8 4,2 27,7 15,5 2,9 4,1 26,9 16,8 2,5 4,6 33,1 18,7 2,8 4,1 30,6 15,9 3,7 4,4 27,5 16,1 6,0 0,0 13,0 0,5 6,1 4,1 5,9 4,3 6,1 3,4 6,2 3,6 6,4 4,0 6,1 4,9 6,0 29,0 29,0 12,6 10,9 21,7 11,8 10,5 21,4 12,1 10,8 20,5 13,5 10,9 23,4 12,5 10,4 22,1 13,6 12,0 21,9 3,0 29,0 5,7 5,7 5,4 5,4 5,1 6,8 n.a. Source: Author's calculations. 146 5.3. Theoretical Framework and Estimation Methods In this section we first present the analytical framework. Then we turn to a description of the range of models addressed in our chapter to estimate the conditional mean function of the logarithm of the total per adult equivalent (p.a.e.) expenditure of rural household in Eastern Province. 5.3.1. Analytical Framework An analytical framework has been constructed to identify at the national level the various channels through which price changes associated with the removal of border trade barriers (analogous to the creation of transport network connectivity) are ―passed through‖ the economic system to influence the welfare of richer and poorer households (Winters, 2002b, McCulloch et al., 2001, UNCTAD, 2004, Winters et al., 2004). We are focusing on households located in the rural areas of Eastern Province, which all were poorly served by transportation infrastructure as well as being plagued by high marketing costs prior to the launch of the EPFRP in 1996 (Chiwele et al., 1998). Within this analytical framework, transport network improvement through the EPFRP is seen as a price shock, which has:   Expenditure effects arising because of changes in the prices of the goods that are consumed; and Income and employment effects arising because of changes in the remuneration of factors of production. Assessing the impact of rural transport infrastructure improvement on poverty in Eastern Province is not an easy task as emphasized by Winters(2002a) “Tracing the links between trade [facilitated by rural transport infrastructure improvements] and poverty is going to be a detailed and frustrating task, for much of what one wishes to know is just unknown.” Since most of the links are very case specific we narrow the scope by carrying out an in-depth study of the EPFRP focusing on the longer-term impacts in the attempt to make these transmission mechanisms a little less opaque. The best way of thinking about poor self-employed rural households is in terms of the ―farm household,‖ which produces goods or services, sells its labor and consumes (Inderjit Singh, δyn Squire, and John Strauss 1986 referred to in Winters et al., 2004). An increase in the price of something of 147 which the household is a net seller (labour, goods, services) increases its real income, while a decrease reduces it. Winters et al., 2004 argue that the framework needs to ask how trade liberalization [or termination of geographic isolation through rural roads improvement in our case] affects all of the different sources of income, as well as considering consumption. Winters et al.(2004) further argue that if price changes are an important pathway through which liberalization affects the poor, then we must ask how trade liberalization and/or trade facilitation through rural roads improvement affects prices. More important than price changes according to Winters et al.,(2004) is whether markets exist at all: Trade reform and/or the opening of previously inaccessible remote rural areas can both create and destroy markets. Winters et al.,(2004) mention that a common worry is that opening up an economy, e.g. through rural roads rehabilitation, will expose it and its component households to increased risk. Certainly, it will expose them to new risks, but the net effect can be to reduce overall risk because world markets (which have many players) are often more stable than domestic ones. The application of this analytical framework evidently entails a number of shortcomings. In addition, our discussion is conducted entirely at the level of the household. The starting point for the definition of a household is the SNA93. In Zambia a household is described as follows: A household is a group of persons who normally live and eat together. These people may or may not be related by blood, but make common provision for food or other essentials for living and they have only one person whom they all regard as the head of the household. A household may also consist of one member (CSO, 2003).139 In the absence of an internationally applied definition of a household, the definition of a household to be adopted in this chapter is that used in Zambian national household surveys. This will be based on the single-dwelling concept. 5.3.2. Models and Estimators We consider three basic approaches to estimate of our welfare measure the logarithm of p.a.e. consumption. The first semi-parametric approach maintains the functional form assumptions but partially relaxes the distribution assumption. The second approach is parametric and is based on strong 139 Collecting data using households as the unit of analysis obscures intrahousehold inequalities in income and consumption. 148 assumptions about the conditional data distribution and functional forms. The final approach shows the variance of the logarithm of consumption as measured by tracking randomly selected representatives of semi-aggregated cohorts defined by date of birth through a time series of cross sections. Although cross-section survey data carried out every second year in Zambia is regarded as infererior to true panel data, some classes of models can be consistently estimated using RCS data, by constructing so-called pseudo panels (Hammer, 2007; Chapter 4). 5.3.2.1. Semiparametric models Nonparametric regression estimators are very flexible but their statistical precision decreases greatly if we as in our case include a vector x of a dimension exceeding two explanatory variables in the model. The latter caveat has been appropriately termed the curse of dimensionality. Consequently, researchers have tried to develop what is usually referred to as semiparametric models and estimators, which offer more flexibility than standard parametric regression but overcome the curse of dimensionality by employing some form of dimension reduction. Such methods usually combine features of parametric and nonparametric techniques to yield the semi-parametric regression model that could help obtain consistent estimates of the parameters of interest.140 Further advantages of semiparametric methods are the possible inclusion of categorical variables (which can often only be included in a parametric way), an easy (economic) interpretation of the results, and the possibility of a part specification of a model. One such example is the partial linear model, which takes the pragmatic point of fixing the error distribution but let the index be of non – or semiparametric structure (Härdle et al., 2004). According to Lokshin(2006) the econometric problem of estimating a partial linear model arises in a variety of settings. We apply the partial linear regression technique to estimation of the household consumption of poverty attributes. Parametric variables include household size; age and education of head of household; asset ownership; distance to input market area; and cotton share of total household income. The EPFRP location effect, which has no natural parametric specification, is incorporated nonparametrically. 140 That is, when estimating semiparametric models we usually have to use nonparametric techniques. 149 5.3.2.2. Partially linear models Thus, as we extend beyond two dimensions, one compromise is the partially linear regression model (PLRM) originally studied by Robinson(1988), which has the form: (3.1) (3.2) (3.3) yi E[ | x, z] E[y | x, z] Xi + f(zi) + 0 Xi + f(zi), = = = i where one part of the model is linear - the Xs - and a single variable has potentially nonlinear relationship with y. The p-dimensional random variable x  Rp and the random variable z  Rq do not have common variables, that is, they are non-overlapping sub-vectors. If they do, then the common variables would be regarded as part of z but not x and the coefficients that correspond to the common variables would not be identifiable.141 ε is i.i.d. mean-zero error term, such that Var[y | x, z] = 2 i function f is a smooth, single valued function with a bounded first derivative. Standard errors of . The are adjusted according to the Yatchew(1998a) method. If there is no cross terms of z among x's, then the model presumes additive separability of nonparametric f(zi) and the parametric Xi parts, which may be too restrictive in some applications. In the PLRM, the convergence rate n-1/2 depends only on the number of continuous regressors among z (Hidehiko, 2005).142 Thus, the curse of dimensionality is avoided in estimating . According to Lokshin(2006) following the methodology suggested by Yatchew(1998b), to estimate the partial linear model (3.1) we first rearrange (sort) the data in such a way that z <z <...<z 1 2 T where T is the number of observations in the sample. Then the first difference of (3.1) results in: (3.4) (yi(n) – yi(n-1)) = (f(zi(n) – f(zi(n-1)) + (xi(n)- xi(n-1)) + i(n)- i(n-1), n = β,…, T. That is the identification of β requires the exclusion restriction that none of the components of X are perfectly predictable by components of z. 142 The nonparametric method would be to use y = m(x, z) + model. The convergence rate would depend on the number of continuous regressors among (x, z). 141 150 When the sample size increases, f(zi(n) – f(zi(n-1)) → 0 because the derivative of f is bounded. Under standard assumptions, equations (3.4) could be estimated by the ordinary least squares (OLS). The vector of estimated parameters  Diff has the approximated sampling distribution:    1 1.5 2   diff  N   , , 2    T u  (3.5) Where 2 u = Var[x | z] is conditional variance of x given z. The error term in (3.4) has a MA(1) structure, thus reducing efficiency of the OLS estimator. The efficiency could be improved by using higher order differences (Yatchew 1998). The generalization of (3.4) for the mth-order differencing can according to Lokshin(2006) be expressed as: d m (3.6) j 1 j m  m  m yi j     d j xit    d j f ( zi j )   d j vi j j 1  j 1  j 1 Where d0,…, dm are differencing weights satisfying the conditions: dj  0 m (3.7) j 1 d m and j 1 2 j 1 The first condition in (3.7) ensures that the differencing removes the non-parametric component in (3.6) as the sample size increases. The second normalization condition implies that the residuals in 2 (3.6) have variance of u. With the optimal choice of weights equation (3.6) could be estimated by OLS. By selecting m sufficiently large, the estimator approaches asymptotic efficiency. Define ∆y to be the (T–m)x1 vector with elements [∆y]i =   m matrix with elements [∆x]i =   d j xi  j  , then:   j 1 (3.8)  Diff (3.9) 2 sdiff  (3.10)   ( x| z ) = (∆x‘∆x)-1∆x‘∆y → d m j 1 1 (x)' (x)   T ( x| z ) 151 yi j and ∆x to be the (T–m)xp 1 1 1 ) 2  N (  , (1  2m T ( x| z )   1 (y  x  diff )' (y  x  diff )   2 T  j This method allows performing inferences on as if there were no non-parametric component f in the model. Once  Diff is estimated, a variety of non-parametric techniques could be applied to  estimate f as if were known. Formally, subtracting the estimated parametric part from both sides of (3.1), we get: yi = xi  Diff = xi ( –  Diff ) + f(zi) +  (3.11)  i ≈ f(xi) + Because  Diff converges sufficiently quickly to true  i , the consistency, optimal rate of convergence, and construction of confidence intervals for f remain valid and could be estimated by the standard smoothing methods (Lokshin, 2006). Using estimates (3.8) it is possible to perform the differencing test for the parametric specification of f. Suppose g(z, π) is the known function of z and some unknown parameter π. We want to test the null hypothesis that y =g(z , π)+x i Parameters π and p i p against the alternative hypothesis that y =f(z )+x . i i i and mean square residual could be obtained by estimating the parametric regression of y on x and z. Then: (3.12) i V=  s2  s2 mT  res 2 diff  s diff     N (0,1) .   5.3.2.3. Model using a Times Series of Cross Sections Finally, we use the same two cross-country data sets (1998 and 2004) to follow cohorts of groups over time from one survey to another. No assumptions are made about the distribution of household consumption in the population; what is being estimated is simply the average consumption in the population in the survey year, not the parameter of a distribution (Deaton, 1997, 1985).Since it is impossible to track individual housholds over time, Deaton (1985) proposed to track cohorts through such RCS data. Successive LCMSs will generate successive random samples of individuals from cohorts based on age, sex or a combination of these variables. Summary statistics from these time series can then be used to infer behavioral relationships for the cohort as if it were a panel. Some disadvantages in the work with genuine panels do not arise in the work with pseudo panels. There is no 152 attrition problem since new samples of individuals are drawn in each LCMS, and a further advantage is often the availibility over a long time period (Hammer, 2007) from 1998 to 2004. In table 5.5 below the 19 rural constituencies in Eastern Province are used to make comparisons over time. These comparisons are often referred to as measurement of net changes in the population (Glewwe and Jacoby, 2000).143 These averages, which relate to households living in the same constituency, have many of the properties of panel data. This approach enables us to address the key question about the gainers and losers in the catchment and counterfactual districts from EPFRP by following the same „Constituency or Age‟ cohort over time. In this quest it is necessary to verify that the measured changes are statistically significant and thus unlikely to be caused by chance alone (Glewwe and Jacoby, 2000). Hence, in table 5.5 we compare the case of two independent crosssectional surveys carried out in the same Constituency. The change in average consumption, say, would be estimated by the difference in average consumptions in 1998 and 2004. The variance of the estimate would be the sum of the variances in the two periods because each cross-sectional sample is drawn independently (Deaton, 1997). Because, our Constituency and Age Cohort data are constructed from two fresh samples, there is no attrition. The way the constituency/age data are used will often be less susceptible to measurement error than in the case with panels. The quantity that is being tracked over time is the mean and the averaging will nearly always reduce the effects of measurement error and enhance the signal-to-noise ratio. In this sense, the constituency/age method can be regarded as an IV method (Deaton, 1997). Despite the fact that we record that two catchment constituencies experienced a negative annual growth rate per capita, of the 19 constituencies in Eastern Province the four control constituencies are only ranked respectively 12, 15-17. The rankings are almost equivalent when measured in p.a.e. terms. The catchment constituencies also largely outperformed the control constituencies with regards to the change in poverty gap, where the latter again only are ranked 14, 16, 17 and 19 far behind the catchment constituencies. 143 However, RCS data reveal nothing about the movements of individuals or households over time, often referred to as gross changes, because different households and individuals are interviewed in each survey. Measurement of gross changes over time can only be addressed using panel data. 153 Table 5.5: Ranking of Consumption Growth and Poverty Change at the Constituency level, 1998 & 2004 District Constituency 1998 Year 2004 1998 Consumption pc 2004 Consumption pc* 2004-1998 Growth in pc 2004-1998 Annual growth in pc Rank of Constituency wrt Growth 1998 Consumption pae 2004 Consumption pae* 2004-1998 Growth in pae 2004-1998 Annual growth in pae Rank of Constituency wrt Growth 1998 Below Food Poverty Line pae 2004 Below Food Poverty Line pae 1998-2004 Change in Poverty Gap Rank of Constituency wrt Poverty Change 1998 Below Total Poverty Line pae 2004 Below Total Poverty Line pae 1998-2004 Change in Poverty Gap Rank of Constituency wrt Poverty Change Location level 301 37 38 45 75 64 38 17905,2 13686,3 42230,6 37086,3 24325,4 23400,0 15,4% 18,1% 6 5 17988,8 13234,1 59969,9 32425,8 41981,1 19191,7 22,2% 16,1% 5 6 6176,1 10930,8 -35805,0 -8260,8 -41981,1 -19191,7 3 8 15573,6 20328,3 -26407,5 1136,6 -41981,1 -19191,7 3 8 302 39 40 45 30 11 12 25703,2 14311,4 26397,4 17856,1 694,1 3544,7 0,4% 3,8% 16 15 25259,7 13194,6 26087,4 17818,7 827,7 4624,0 0,5% 5,1% 16 14 -1094,8 10970,3 -1922,5 6346,3 -827,7 -4624,0 19 16 8302,7 20367,8 7475,0 15743,7 -827,7 -4624,0 19 16 303 41 149 66 12887,3 22802,3 9915,0 10,0% 9 12277,4 21771,4 9494,0 10,0% 10 11887,5 2393,5 -9494,0 13 21285,0 11791,0 -9494,0 13 42 94 22 44972,0 30435,6 -14536,3 -6,3% 19 44417,0 35798,4 -8618,6 -3,5% 18 -20252,1 -11633,5 8618,6 15 -10854,6 -2236,0 8618,6 15 43 15 58 5717,5 52567,3 46849,8 44,7% 1 5533,0 48661,6 43128,6 43,7% 1 18631,9 -24496,7 -43128,6 2 28029,3 -15099,2 -43128,6 2 304 44 45 46 15 59 46 53 36 56 11254,4 15539,4 39104,0 50329,6 21482,0 27012,8 39075,2 5942,6 -12091,2 28,4% 5,5% -6,0% 2 13 18 9791,8 14861,8 38709,7 57187,1 26486,1 26632,0 47395,3 11624,3 -12077,7 34,2% 10,1% -6,0% 2 9 19 14373,2 9303,1 -14544,8 -33022,2 -2321,2 -2467,1 -47395,4 -11624,3 12077,7 1 12 10 23770,6 18700,6 -5147,3 -23624,7 7076,3 6930,3 -47395,3 -11624,3 12077,7 1 12 10 47 44 47 8606,9 25784,3 17177,4 20,1% 4 8345,7 28764,7 20419,0 22,9% 4 15819,3 -4599,7 -20419,0 6 25216,7 4797,7 -20419,0 6 48 77 81 32889,3 55100,3 22211,0 9,0% 10 32841,2 53013,7 20172,5 8,3% 12 -8676,3 -28848,8 -20172,5 7 721,1 -19451,3 -20172,5 7 305 49 14 51 19234,3 35997,7 16763,4 11,0% 8 17424,7 30298,2 12873,5 9,7% 11 6740,2 -6133,3 -12873,5 9 16137,7 3264,2 -12873,5 9 50 78 77 34740,4 47037,8 12297,4 5,2% 14 33812,9 62419,4 28606,5 10,8% 8 -9647,9 -38254,5 -28606,5 5 -250,5 -28857,0 -28606,5 5 306 51 105 74 23470,2 33772,9 10302,7 6,3% 12 22808,2 32035,4 9227,1 5,8% 13 1356,7 -7870,4 -9227,1 14 10754,1 1527,0 -9227,1 14 307 52 121 85 26129,9 23092,0 -3037,9 -2,0% 17 25745,6 29284,8 3539,2 2,2% 15 -1580,7 -5119,8 -3539,2 17 7816,8 4277,6 -3539,2 17 53 75 81 9809,9 21270,5 11460,6 13,8% 7 9528,4 21577,7 12049,3 14,6% 7 14636,5 2587,2 -12049,3 11 24034,0 11984,7 -12049,3 11 308 54 105 54 22116,9 35461,7 13344,9 8,2% 11 21714,1 19700,9 -2013,3 -1,6% 17 2450,8 4464,1 2013,2 18 11848,3 13861,5 2013,3 18 55 54 33 12311,0 49565,4 37254,4 26,1% 3 11521,6 52942,2 41420,7 28,9% 3 12643,4 -28777,3 -41420,7 4 22040,8 -19379,9 -41420,7 4 Notes: * Constant values in 1998 prices. Sourceμ Authors‘ computations. Fig.5.2: Household Expenditure by cohort, 1998 & 2004 50000 45000 45000 15000 Age of household head Cohort: Age of Household in 1998 Notes: Five-year moving averages are shown in the figure. Sourceμ Authors‘ computations. 154 56 53 20 62 59 56 53 50 47 44 41 38 35 32 0 29 0 26 5000 23 5000 50 10000 47 10000 44 15000 pae0498 20000 41 20000 pae98 25000 38 pae0498 30000 35 pae98 25000 32 30000 35000 29 35000 40000 26 40000 23 Real Consumption in 1998 Prices (ZMK) 50000 20 Real Consumption in 1998 prices (ZMK) Fig.5.1: Household Expenditure by Age, 1998 & 2004. In table 5.6 we show that we can also use the survey data to follow cohorts of groups over time, where cohorts are defined by date of birth. We use the successive LCMSs to follow each cohort over the 6 year period from 1998 to 2004 by looking at the members of the cohort who are randomly selected into each survey. For example, we look at the mean p.a.e. consumption of 20-year-olds in the 1998 survey, of 26-year-olds in the 2004 survey, etc. These averages, because they relate to the same group of people, have many of the properties of panel data. Because there are many cohorts alive at one time, cohort data are more diverse and richer than the aggregate constituency data, but the semi-aggregated structure provides a link between the micro-level household level and the macro-economic data from national accounts. Table 5.6 shows, that there were 1287 members of the cohort in the 1998 survey and 907 in the 2004 survey (in which the sample size was increased for reasons explained above). Table 5.6 also illustrates the same process for ten cohorts, born in 1978, 1974 and 1973, 1969, and so backward at five-year intervals until the oldest, which was born in 1933, 1929, and the members which were from 65 to 69 years old in 1998. Tracking different cohorts through successive surveys allows us to disentangle the generational from life-cycle components in consumption profiles as shown in figure 5.1 and figure 5.2 (Deaton, 1997). Table 5.6: Number of Persons in Selected Cohorts by Survey Year, 1998 & 2004 Whole Group Catchment Group Counterfac tual Group Cohort: Age in 1998 15-19 20-24 25-29 30-34 35-39 40-44 45-49 50-54 55-59 60-64 65-69 70-74 75-79 80-84 1998 6 114 191 168 159 126 116 103 78 75 73 47 22 9 2004 90 159 142 132 102 96 59 64 46 54 28 17 4 4 1998 6 75 142 120 126 92 83 77 58 59 59 29 15 7 2004 66 136 114 104 83 80 48 45 34 40 26 14 4 4 1998 0 39 49 48 33 34 33 26 20 16 14 18 7 2 2004 24 23 28 28 19 16 11 19 12 14 4 1 0 0 >84 6 0 4 0 2 0 Total 1287 907 946 732 341 175 Notes: The Age groups matches the cohorts in 1998, so that the age in Year 2004 is exactly 6 years older than in 1998, corresponding to the difference in year between 1998 and 2004. The figures 5.1 and 5.2 above shows the average real household expenditure p.a.e. (in 1998 prices) of all households with heads of that age against various cross-sectional 155 age profile cohorts in Zambia‘s Eastern Province observed from 1998 through to 2004.144 For example, the data in figure 5.1 compares the consumption of a 20 year-old in 1998 with the consumption of 20 year-old in 2004, whereas in figure 5.2 show e.g. the cohort born in 1970, who were 28 years old in 1998 are compared to those aged 34 in 2004. The 1998 result is plotted in the blue pae98 curve in the figure 5.2. The average consumption of 34-year olds in the 2004 LCSM IV survey is calculated and is depicted in the red pae0498 curve on the same segment as the pae98. The rest of the two lines depict the relationship between age and consumption as captured by these two surveys.145 The graphs, especially the one depicting the 1998 survey, do not look much like the stylized life-cycle profiles, which sees individuals smoothing their consumption over their lifetime. Moreover, figure 5.1 shows a clear age effects on consumption. In 33 out of 45 ages (i.e. 73%), when the cohorts are observed as averages for a single year of age in each year, and not smoothed by combining adjacent years into moving averages as in figures 5.1-2 (e.g. 30 year-old in 1998 vs. 30 year-old in 2004), was the 2004 cohort above the 1998 cohort. This is an indication of the growth of real consumption raising the profiles through this distinct time period. This age effect was even more pronounced when the line for the younger 2004 cohort with a very few exceptions (13%) are always above the line for the older 1998 cohorts (e.g. 41 year-olds in 1998 vs. 47 year-olds in 2004). This could be a reflection of the macroeconomic growth in Zambia from 1999 to 2006 (chapter 2), which was making younger generations better-off. In other words, the average real household consumption in Eastern Province increases by ZMK11,248 per month, equivalent to a 42.4% increase from 1998 to 2004 (i.e. an annual increase of θ.1%) for the ‗same age cohort‘ of different households tracked over time (figure 5.2).146 144 That is the age profiles of consumption for 1200 and 900 rural households in the LCMS II and IV for 1998 and 2004.To plot consumption against age is perhaps the most obvious way to examine the life-cycle behaviour of consumption from survey data (Deaton, 1997:339). 145 In order to keep the age profiles of consumption relatively smooth, I have used the averages for each age to calculate the five-year moving averages shown in the two figures. Five-year smoothing also eliminates problems of ―age-heaping‖ when some people report their age rounded to the nearest five years (ibid). 146 For an enumeration of a number of advantages which cohort data have over most panels see Deaton(1997:120f). 156 There are a number of other difficulties with age profiles of consumption as those in figure 5.1-2 above. First, these profiles are simply the cross sections for the households in the surveys, and there is no reason to suppose that the profiles represent the typical or expected experience for any individual household or group of households. We are only looking at the experience of different ages of different groups of households, whose members were born at different dates and have had quite different lifetime experiences of education, wealth, earnings, and accessibility e.g. through the EPFRP. Without controlling for these other variables, many of which are likely to affect the level and shape of the age profiles, we cannot isolate the pure effect of age on consumption or of EPFRP on consumption. Panel Data from Successive Cross Sections Following Deaton(1997, 1985) we briefly consider the issues that arise when using cohort data as if they were repeated observations on individual units. We first consider the simplest univariate model with fixed effects, so that at the level of the individual household, we have a parametric the linear regression model with a single variable: (3.13) yit =  + ‘xit + i + t + uit, where the t are dummies and i is an individual-specific fixed effect.147 If we average (3.13) over the members of the age group who appear in the survey, and who will be different in 1998 and 2004, the “fixed effect” will not be fixed, because it is the average of the fixed effects of different households in each year. Because of this sampling effect, we cannot remove the age group fixed effects by differencing or using within-estimators. Deaton(1997) considers an alternative approach based on the unobservable population means for each cohort. Starting from the cohort version of (3.13), and denoting population means in cohorts by the subscripts c, so that, simply changing the subscript i to c, we have: For an exposition of the „Fixed Effects Model in the two-period case‟ see section 12.6 in Johnston&Dinardo(1996:395ff). 147 157 yct =  + ‘xct + c + t + uct, (3.14) And take first differences – to eliminate the fixed effects so that: Δyct = yc04 – yc98 = Δxct + Δt + Δuct, (3.15) where the first time is a constant in any given year. This procedure has eliminated the fixed effects, but we are left with the unobservable changes in the population cohort means in place of the sample cohort means, which is what we observe. If we replace Δy and Δx in (3.15) by observed changes in the sample means, we generate an error-invariables problems, and the estimates will be attenuated.148 If the number of observations per cohort is large, it is tempting, and done in practice, to ignore the measurement error problem. Verbeek and Nijman (1992a) analyse under which circumstances this can be a valid approach (see chapter 4). They show that the bias of the estimate will inded be small if there is sufficient variation within the cohorts. Verbeek and Nijman (1992) assume that the regressor variables xit are correlated with the cohort defining variable zi in the following way: (3.16) xit = t + tzi + it, The choice of larger cohorts will reduce the bias if either t , t or both vary with t. The cohorts can be choosen smaller if this variation is large relative to the variance of it. Verbeek and Nijman (1993) also point out that there is a trade-off between the number of observations in a cohort and the number of cohorts in a panel. Since a decrease in the number of obervations in a panel implies an increase in the variance of the FE-estimator, this results in a trade off between bias and variance. They show that consistency of the estimator proposed by Deaton requires that the number of time periods go to infinity (Hammer, 2007). Moffitt (1993) proposes an alternative approach to estimation from RCS. He starts from the observation that grouping can be regarded as IV procedure. This is e.g 148 There are at least two ways of dealing with the error-in-variables problem, see Deaton(1997:122f). 158 illustrated in Verbeek (1996). The IV interpretation allows a whole new range of estimators. But as Verbeek (1996) remark, while the FE-estimator is consistent as the number of individuals per cohort tends to infinity, this is not guaranteed for estimators in the class suggested by Moffitt. They are consistent if the instruments for xit vary with i conditional on the instruments for i. Having instruments for individual time series which do not exhibit individual variation might lead to inaccurate estimators (Hammer, 2007). Decompositions by age, cohort, and year In the case of lifetime consumption profile, if the growth in living standards acts so as to move up the consumption-age profiles proportionately, it makes sense to work in logarithms, and to write the logarithm of consumption as: (3.16) ln cct = + αa + c + t + uct, where the superscripts c and t (as usual) refer to cohort and time (year), and a refers to age, defined here as the age of the cohort c in year t (1998). A convenient way to label cohorts is according to Deaton(1997) to choose c as the age in year t=0 (i.e. 1998). By this, c is just a number like a and t. Given the way cohorts have been defined, with bigger values of c corresponding to older cohorts, we would expect c to be declining with c (i.e. the age of the cohort in year 0): (3.17) act = c + t, That is in year 2004 (t=6) the cohort who in 1998 (t=0) was 30 years-old (c=30) would have the age 36 (a30,6 = 36). Using LCMS data to extract information on growth in p.a.e. consumption In order to impose structure, we follow Dercon and Hoddinott(2005) who borrow from the conceptual frameworks used to understand growth at the national or crosscountry level such as that found in Mankiw et al.,(1992) (chapter 2). In the context of 159 cohort panel data on p.a.e. consumption, yct, of N cohorts c (c=1, ... , N) across periods t, a version of this empirical model can be written as in Islam(1995): (3.18) lnyct – lnyct-1 = α + lnyct-1 – kct-1 + ΔZct + Xc + uct. ΔZct are changes in time-varying characteristics of cohorts and communities that help to explain growth and Xc are fixed characteristics of the cohorts and the community. Examples of ΔZct could according to (Dercon and Hoddinott, 2005) be changing levels of different (exogenous) assets (i.e. not due to investment decisions, but exogenously changing endowments at the cohort and community). They also include exogenous shocks in the specification, for example, rainfall shocks. The presence of Xc would suggest that different types of cohorts may have a particular growth path, linked to fixed characteristics (such as distance to commercial bank in town, etc.). Following, Dercon and Hoddinott(2005) we add further “initial” conditions to the specification, i.e., variables related to assets whose presence may have growth effects (kct-1). Examples are levels of landholdings or infrastructure. Thus, we use equation (3.18) for our test to see whether the EPFRP infrastructure and accessibility matter for understanding growth in consumption outcomes in the period from 1998 to 2004. Because we want to focus on age cohort variables, it makes sense to run our regression using the most complete controls for household-level variables that do not change over time. This is accomplished by estimating a fixed-effect regression – essentially including a dummy variable for each household in the sample – that controls for all household characteristics that might affect the growth of consumption but do not change over time. Consequently, all our covariates are identified using changes over time. The attraction of such an approach is according to Dercon and Hoddinott(2005:16) that ―we avoid some standard issues, such as placement effects due to fixed factors and other sources of endogeneity affecting accessibility. However, while desirable in terms of ensuring that we can confidently identify the impact of variables that change over time, this approach comes with a cost: that we cannot identify factors that do not change over time.‖ 160 5.4. Estimation Results Our analysis use the data collected through respectively the 1998 LCMS (II) and the 2004 LCMS (IV) to evaluate the effects of the EPFRP on the well being of rural households in the catchment and control districts of Zambia‘s Eastern θrovince as well as the changes in living standards from 1998 to 2004, using LCMS 1998 as a useful benchmark of living standards, coming as it does at the end of the most intensive part of the economic reform period from 1991 to 1996. Table 5.7: Descriptive Statistics of the Data covering Rural Eastern Province LCMS II 1998 Full Sample (i) Variables Household Size Age of household head Age 0-14 Age 15-24 Age 25-65 Age 66 and over Education level of household head No education Grade 1-4 Grade 5-7 Grade 8-9 Grade 10-12 Grade 12 GCE (A) College/undergraduate/Bachelor's degree and above Maritual status of head Never married Married Separated Divorced Widower Not stated Employment Status? SELF EMPLOYED CENTRAL GOVT EMPLOYEE LOCAL GOVT EMPLOYEE PARASTATAL EMPLOYEE PRIVATE SECTOR EMPLOYEE NGO Employee Intern.Org. & Embassy Employee EMPLOYER/PARTNER HOUSEHOLD EMPLOYEE UNPAID FAMILY WORKER PIECE WORKER OTHER Food exp / total exp Total HH Income (ZMK / month) Total HH expenditure (ZMK / month) Food exp (ZMK / month) Total area under cultivation Sample size Mean 5,06 43,06 0* 120* 1037* 149* 5,85 408* 188* 255* 67* 37* 11* Std.Dev. 2,931 15,772 n.a. n.a. n.a. n.a. 2,843 n.a. n.a. n.a. n.a. n.a. n.a. 2* 3,98 240* 497* 22* 51* 95* 401* 4,65 403* 20* 2* 0* 16* n.a. 0* 5* n.a. 452* n.a. 1* 0,41 345402,5 67543,43 32338,62 1,97 n.a. 2,864 n.a. n.a. n.a. n.a. n.a. n.a. 3,44 n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. 0,26 8112332,0 110622,30 40705,97 3,637 LCMS IV 2004 Extremely Poor Households (iii) Poor Households (ii) Mean 5,16 43,43 0* 112* 970* 146* 5,78 398* 183* 237* 54* 27* 8* Std.Dev. 2,93 15,87 n.a. n.a. n.a. n.a. 4,54 n.a. n.a. n.a. n.a. n.a. n.a. 1* n.a. 4,04 2,89 223* n.a. 458* n.a. 20* n.a. 43* n.a. 92* n.a. 392* n.a. 4,74 3,45 377* n.a. 13* n.a. 1* n.a. 0* n.a. 13* n.a. n.a. n.a. 0* n.a. 5* n.a. n.a. n.a. 439* n.a. n.a. n.a. 1* n.a. 0,41 0,27 346626,0 8346002,0 52668,73 62883,78 28239,41 32459,35 1,83 2,41 1300 Full Sample (iv) Poor Households (v) Mean 5,19 43,64 0* 104* 927* 141* 5,68 386* 175* 228* 48* 23* 4* Std.Dev. 2,93 15,84 n.a. n.a. n.a. n.a. 4,58 n.a. n.a. n.a. n.a. n.a. n.a. Mean 5,43 42,42 0* 120* 1286* 177* 5,03 344* 389* 532* 171* 112* 3* Std.Dev. 3,01 15,23 n.a. n.a. n.a. n.a. 3,88 n.a. n.a. n.a. n.a. n.a. n.a. Mean 5,70 43,44 0* 20* 187* 52* 4,50 76* 1* 4,06 215* 433* 19* 40* 88* 377* 4,77 357* 8* 1* 0* 12* n.a. 0* 5* n.a. 419* n.a. 1* 0,41 351442,0 46711,38 25740,02 1,81 n.a. 2,89 n.a. n.a. n.a. n.a. n.a. n.a. 3,45 n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. 22* 2,43 36* 1643* 41* 88* 235* 1228 Extremely Poor Households (vi) 70* 24* 5* 0* Std.Dev. 2,96 15,45 n.a. n.a. n.a. n.a. 3,61 n.a. n.a. n.a. n.a. n.a. n.a. Mean 5,80 43,80 0* 9* 122* 37* 4,32 57* 57* 40* 12* 2* 0* n.a. 1,03 n.a. n.a. n.a. n.a. n.a. 1* 2,82 3* 225* 17* 32* 68* n.a. 1,24 n.a. n.a. n.a. n.a. n.a. 0* 3,02 1* 114* 15* 25* 49* 1,30 n.a. n.a. n.a. n.a. n.a. 3,85 869* 27* 3* 7* 34* 0* 0* 0* 4,10 n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. 4,24 155* 2* 0* 0* 0* 0* 0* 0* 4,33 n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. 4,10 99* 0* 0* 0* 0* 0* 0* 0* 4,29 n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. 0* n.a. 0* n.a. 0* n.a. 82* Std.Dev. 2,96 15,66 n.a. n.a. n.a. n.a. 3,54 n.a. n.a. n.a. n.a. n.a. 377* n.a. 81* n.a. 48* n.a. 0,27 0,79 0,20 0,78 0,19 0,78 0,20 8542870,0 52116,29 483340,70 567464,10 390953,70 468119,50 369773,40 458534,30 27650,76 381419,10 514630,20 305285,10 411886,20 289068,50 406648,00 2,40 1172 1583 1072 889 Note: * Number of heads of households. Source: Author's calculations. The set of transmission mechanisms suggested in the theoretical framework of Winters(2002a), namely prices, wages and employment, are associated with the consumption outcomes varying by personal characteristics such as age and education. The basic rural household characteristics are summarized in table 5.7. The two LCMSs 161 only show small differences in the characteristics of rural households amongst the whole sample versus the moderately and extremely poor rural households. The extremely poor households in rural Eastern Province are on average larger and have more children per working age adult than better off households. Although the average year of schooling has increased since 1998, the poorest heads of households still have less schooling than the heads of better off households. Moreover, they are also older. 5.4.1. Model Diagnostics We share the belief that better models and a better understanding of one's data result from focused data analysis as well as the inclusion and exclusion of respectively exogenous and endogenous regressors for the structural model guided by substantive theory.149 The descriptive statistics for the data of the covariates are presented in table 5.8. The sample size of the reduced model's dependent variable is now 1287 in 1998 and 999 in 2004, which is significantly larger than the full model.150 The sample size is distributed between the catchment and the control districts. That is, 948 and 339 rural households were sampled in 1998 within the catchment and control districts respectively. Likewise 817 and 182 rural households were sampled in 2004 within the same catchment and control districts (Table A7). Surprisingly, the mean of cotton sales as a share of household income increased faster between 1998 and 2004 in the control districts (294%) than in the catchment districts (157%). Even more unexpectedly is the finding that the average distance to input markets in Eastern Province decreased faster in the in the control districts (-76%) compared to the catchment districts (-42%). Yet the percentage change in the logarithm 149 Economic theory often provides some guidance in model specification but may not explicitly indicate how a specific variable should enter the model, identify the functional form, or spell out how the stochastic elements (ei) enter the model. Comparative static results that provide expected signs for derivatives do not indicate which functional specification to use for the model (Baum, 2006). 150 Our initial specification reported all theoretically possible variables found in the existing CSO datasets. However, only 9 out of 70 coefficients are statistically significant at the 10% level in the 1998 dataset, using a 2-tailed p-value testing the null hypothesis that the coefficient is null. This is probably because our sample size in 1998 dataset (n=198) is so reduced by missing data on the large number of explanatory variables (70) covering the rural areas in Eastern Province. 162 of p.a.e. expenditures increased at a higher rate in the catchment districts (+16%) than in the control districts (+10%) (Table A7). Table 5.8: Descriptive Statistics of covariates, 1998 and 2004 Type CV CV DV CV DV DV DV DV DV CV CV DV Variable Name Variable Log pae monthly household expenditure LNPAE98* Cotton Sales share of household income Cotincshare Stratum, excl. Large AHH Stratum124** Distance to Inputmarket Distiput EPFRP Treatment Infrastructure*** Plough Ownership Plough Bicycle Ownership Bicycle Scotchcart Ownership Scotchcart Motorvehicle Ownership Motorvehicle Age of Head of Household Age Age Squared Agesq Head of HH ever attended School s4q5 Obs Mean 1287 8,996 1274 1304 1311 1311 1311 1311 1311 1311 1306 1306 968 1998 Std.Dev. Min. 1,234 3,912 Max. Obs 14 Mean 999 10,293 2004 Std.Dev. Min. 1,160 Max. Percentage change 6,563 13,771 0,104 0,202 0,010 1 999 0,293 0,338 0 1,510 1,028 1 4 999 1,193 0,417 1 22,692 22,957 0,000 99 999 12,139 17,566 0 0 0 0 0 999 0,818 0,386 0 0,799 0 0 1 999 0,683 0,466 0 0,483 0 0 1 999 0,337 0,473 0 0,901 0 0 1 999 0,822 0,383 0 0,976 0,152 0 1 999 0,979 0,144 0 43,054 15,745 15,000 99 999 42,888 15,024 20 2101,322 1522,004 225 9801 999 2064,876 1464,556 400 0,421 0,494 0 1 997 0,217 0,412 0 1 4 99 1 1 1 1 1 90 8100 1 14% 183% n.a. -47% n.a. n.a. n.a. n.a. n.a. 0% -2% n.a. Notes: * Logarithm of the Per Adult Equivalent total household expenditure; ** Small - and medium scale farmers, and non-agricultural rural households. *** Dummy / indicator variable takes the values 0 or 1 to indicate the absence or presence of some categorical (EPFRP) effect that may be expected to shift the outcome. Source: Author's calculations. The dependent variable is still this natural logarithm of p.a.e. monthly expenditure in both 1998 and 2004. Our aim is to measure the differences between the estimate of the 1998 mean of Y=LNpae98 conditional on five covariates and the estimate of the 2004 mean of Y=LNpae04 conditional on the same covariates. These covariates described in table 5.8 are rural stratum excluding the small number of large scale farmers, the distance to the nearest input market, the head of household's ownership of a motor vehicle and the EPFRP Treatment dummy variable (denoted infrastructure) and finally the share of income derived from cotton sales to total rural household income (denoted cotincshare).151 In other words we explore the link between the rural households response to the price changes ensuing from improved rural road transport infrastructure investment by focusing on the single most important cash crop namely cotton. That is, the critical factor in our case is the share of household income generated by the sales of the cotton production (Brambilla and Porto, 2007, Balat and Porto, 2005a, Balat and Porto, 2005b, Govereh et al., 2000, Kabwe, 2009, Kabwe 151 Alternatively we could use a dummy variable of the cotton variable to show whether the household derived any income from cotton cultivation (=1) or not (=0). 163 and Tschirley, 2007, Tschirley and Kabwe, 2007, Zulu and Tschirley, 2002, Zambia Food Security Research Project, 2000). 5.4.2. Estimation results of Parametric and Semiparametric Models This section reports the results of estimating the mean of Y conditional on the five covariates of a linear parametric model (4.1) and of a semiparametric model (4.2). The parametric model is more parsimonious, and thereby more interpretable than models proposed through a selection estimation procedure. The models that we estimate are as follows for respectively the 1998 and 2004 datasets: (4.1) E(LNpae | Stratum124, Cotton sale / Household income, Distance Input market, Motor vehicle, Infrastructure) = 0 + 1*Stratum1β4 + 2*(ωotton Sale / HH Income) + 3*Distance Input market + 152 4*εotorvehicle ownership + 5*Infrastructure + eht, (4.2) E(LNpae | Stratum124, Cotton sale / Household income, Distance Input market, Motor vehicle, Infrastructure) = m(X) = 1*Stratum1β4 + β*ωotincshare + γ*Distiput + 4*s10q8 + G( η*Infrastructure) Where G is an unknown function and the s are unknown scalar parameters. The error term eht comprises the rural household-level error term of household h at time t (1998 or 2004). Model (4.2) is a semiparametric partially linear model (3.1), which is a further example of semi-parametric GLM that is able to handle (additional) nonparametric components as discussed in section 5.1.153 In order to estimate the coefficients of (4.1) we had to choose a correct OLS analysis for the LCMS II and LCMS IV survey design, in other words we do OLS regression with clusters.154 Rather than the standard linear predictor (4.1) the partially linear model allows (4.2) one predictor - infrastructure - to be nonlinear. 152 To avoid that our statistics becomes misleading from outliers associated with the different population of the seven large scale farmers in rural Eastern Province in 1998, which is different from the rest of the sample set. Hence these seven data points will all be censored in the remaining part of this paper. 153 A partially linear model, that is a semiparametric model of the type: Y = Xtâ + ö(Z) + U, where X and Z are multivariate explanatory variables. The apparent asymmetry between the effect of the variables X and Z in the partially linear model brings the analyst to include all dummy or categorical variables in the parametric component of the model. The GPLM can be viewed as a compromise between the GLM and a fully nonparametric model (He et al., 2005). 154 In Annex A6.a-b the tables provides a summary of the various parametric models shown that we tried. Some of the models only adjust the standard error by computing a cluster robust standard error for the coefficient. Other procedures do more complex modeling of the multilevel structure. And there are some procedures that do various combinations of the two. 164 In table 5.9 the OLS model is compared to its flexible generalization the standard Generalized Linear Model (GLM) with the unknown parameters, , fitted using NewtonRaphson (maximum likelihood) optimization.155 The dependent variable is assumed to be generated by the distribution function, f, from the Gaussian(normal) probability distribution family (Hardin and Hilbe, 2007). The results show that cotton‟s income share of total income went from negative and insignificant in 1998 to positive coefficients and statistical significance at the 0.05 level. Table 5.9: Comparison of Generalized linear models, 1998 and 2004 1998 Variable Names Covariates Cotton Sales share of household income Stratum, excl. Large AHH Distance to Inputmarket cotincshare Age of Head of Household Age OLS OLS 2004 GLM GLM OLS OLS GLM GLM -0.1237 -0.1650 -0.1237 -0.1650 -0.0673 -0.1241 -0.0673 -0.1241 (0.1888) (0.1699) (0.2360) (0.2674) (0.1073) (0.1068) (0.0776) (0.0852) stratum124 0.2432*** 0.2672*** 0.2432*** 0.2672*** -0.2783*** -0.3609*** -0.2783 -0.3609** (0.0383) (0.0336) (0.0315) (0.0248) (0.0964) (0.0885) (0.1827) (0.1415) Distiput -0.0033* -0.0044*** -0.0033 -0.0044*** -0.0018 -0.0027 -0.0018 -0.0027 (0.0017) (0.0015) (0.0030) (0.0016) (0.0021) (0.0021) (0.0018) (0.0021) Infrastructure 0.3228*** 0.3035*** 0.3228*** 0.3035*** EPFRP Treatment (0.0941) (0.0949) (0.1150) (0.1135) Motorvehicle Motorvehicle -1.5649*** -1.5959*** -1.5649*** -1.5959*** 0.9974*** 1.0724*** 0.9974*** 1.0724*** Ownership (0.2942) (0.2519) (0.4180) (0.1968) (0.2532) (0.2562) (0.1490) (0.1387) Plough 0.0266 0.0266 0.1083 0.1083 Plough Ownership (0.1193) (0.0902) (0.1011) (0.0772) Bicycle -0.3326*** -0.3326*** 0.2172*** 0.2172*** Bicycle Ownership (0.0799) (0.0663) (0.0803) (0.0792) Scotchcart -0.1089 -0.1089 -0.0431 -0.0431 Scotchcart Ownership (0.1672) (0.1133) (0.1209) (0.0825) Agesq Age Squared Head of HH ever attended School s4q5 _cons Constant N r2 F ll -0.0297** (0.0137) 0.0002* (0.0001) -0.0297* (0.0164) 0.0002 (0.0002) -0.0656*** (0.0145) 0.0007*** (0.0001) -0.0656** (0.0297) 0.0007** (0.0003) -0.4115*** -0.4115*** -0.1055 -0.1055 (0.0787) (0.0454) (0.0883) (0.1356) 11.4715*** 10.2921*** 11.4715*** 10.2921*** 10.6893*** 9.4946*** 10.6893*** 9.4946*** (0.4447) (0.2605) (0.5691) (0.2157) (0.4659) (0.3038) (0.4732) (0.2407) 920 1246 920 1246 997 999 997 999 0.1554 0.0888 0.0873 0.0521 16.7288 30.2329 8.5654 10.9101 -1420.2349 -1960.7012 -1420.2349 -1960.7012 -1517.3167 -1538.6864 -1517.3167 -1538.6864 Notes: Distribution of depvar (LNpae) is family(gaussian). The Generalized Linear Models (GLM) are fitted using Newton-Raphson (maximum likelihood) optimization with robust standard errors in parentheses. * p<0.10, ** p<0.05, *** p<0.01. Source: Author's calculations. 155 The GLM generalizes linear regression by allowing the linear model to be related to the response variable via a link function and by allowing the magnitude of the variance of each measurement to be a function of its predicted value. 165 Alternatively in order to calculate the linear partial regression we follow Michael Lokshin, who estimates the following semiparametric regression model by the method of differencing: (4.3) yi = Xi + f(zi) + i. The weighting matrix (W) accounts for the correlations among the set of l-moments when the errors are not i.i.d. The weighting matrix only plays a role in the presence of over-identifying restrictions.156 The significance test of the Treatment Infrastructure variable that enters the Yatchew specification non-linearly indicates that the infrastructure dummy variable is highly significant (P-value of 0.000) in 2004 (column 6 table 5.10). The same is the case if the form of the vector of differencing weights d ,…,d , are specified using Hall et. al.(1990) weights for differencing (column 8 table 1 m 5.8, cf. figures A8.1-2). Compared with the estimation of the fully parametric model (table 5.7) one finds that while the signs of the coefficients are almost the same in 1998 between the two specifications, the magnitudes of some coefficients are different. For example, the effect of „distance to the input market‟ on the logarithm of p.a.e. consumption of rural household change from -0.0044 in the fully parametric model to -0.0031 in the partial linear model estimation with Yatchew‟s weighing matrix in 1998. In the 2004 dataset the coefficient of the same covariate change from -0.0027 to -0.0007, whereas the „cotton sales share‟ change from wrong sign – 0.1241 to the correct positive coefficient 1.3910 in 2004 (table 5.10). 156 If the equation to be estimated is overidentified, l > k, we have more equations than we do unknowns. The optimal weighting matrix is that which produces the most efficient estimate (Baum, 2006). 166 Table 5.10: Partial Linear regression models, 1998 and 2004 1998 2004 Weighting Matrix Weighting Matrix Variable Names Covariates Yatchew Yatchew Hall Hall Yatchew Yatchew Hall Hall Cotton Sales share of cotincshare -11.5638 -15.1956 -0.3689 -1.7352 -0.3784 2.5665 -0.2792 1.3910* household income (17.8541) (16.7153) (4.4799) (4.9766) (1.8683) (1.7332) (0.7920) (0.7657) Stratum, excl. Large stratum124 0.1725 0.2327** 0.3116** 0.2865*** -0.2070 -0.3197*** -0.3481*** -0.4463*** AHH (0.1697) (0.1132) (0.1297) (0.1076) (0.1362) (0.1188) (0.1136) (0.1025) Distiput 0.0024 0.0031 0.0021 -0.0004 -0.0000 -0.0004 0.0007 -0.0007 Distance to (0.0037) (0.0025) (0.0029) (0.0025) (0.0031) (0.0028) (0.0026) (0.0025) Inputmarket Motorvehicle Motorvehicle -1.7509*** -1.4151*** -1.5850*** -1.7253*** 0.8686*** 0.8220*** 0.8870*** 0.9726*** Ownership (0.3543) (0.2524) (0.2969) (0.2466) (0.3278) (0.3045) (0.2634) (0.2565) Plough 0.2388 0.0676 0.0438 0.1061 Plough Ownership (0.1631) (0.1230) (0.1328) (0.1060) Bicycle -0.2198* -0.3005*** 0.2127** 0.1939** Bicycle Ownership (0.1186) (0.0839) (0.1071) (0.0881) Scotchcart -0.1415 -0.0269 -0.0141 -0.0390 Scotchcart Ownership (0.2090) (0.1728) (0.1540) (0.1288) Age -0.0124 -0.0247* -0.0720*** -0.0727*** Age of Head of (0.0174) (0.0139) (0.0190) (0.0151) Household Agesq 0.0001 0.0002 0.0007*** 0.0008*** Age Squared (0.0002) (0.0001) (0.0002) (0.0002) Head of HH ever s4q5 -0.3860*** -0.3557*** -0.0304 -0.0685 attended School (0.0937) (0.0796) (0.1051) (0.0922) N 919 1245 913 1239 996 998 990 992 r2 0.0923 0.0473 0.1024 0.0499 0.0526 0.0284 0.0745 0.0440 F 9.2443 15.4142 10.3005 16.2273 5.4765 7.2561 7.8838 11.3670 ll -1385.5857 -1625.6319 -1384.8174 -1863.7249 -1494.5794 -1428.8287 -1505.5017 -1486.7673 V(i) 2.266 25.016 4.445 13.901 1.768 8.067 1.197 8.168 P>|V| 0.012 0.000 0.000 0.000 0.039 0.000 0.116 0.000 Notes: (i) Significant test on Infrastructure. Standard errors in parentheses * p<0.10, ** p<0.05, *** p<0.01 Source: Author's estimations. 5.4.3. Estimation results of Panel Data from Successive Cross Sections In this section we use cohort panel data, which we likewise draw from the LCMS II and LCMS IV. There are 45 cohort-year pair observations. In this case the number of instruments (cohorts) is large relative to the number of individuals. Hence a small sample bias is present in the IV estimator. If on the other hand, the cohort sizes are 100 or more, these biases seem to be acceptable, provided sufficient cohort-specific variation is present in the exogenous variables (Verbeek and Vella, 2005). The data contain one cohort(age) identifier, two years of data on monthly p.a.e. expenditure, two years of cotton income shares, two years of landownership per household member, two years of rainfall data and of lagged rainfall data, two years of highest grades attained, and two years of distances to public or private services (input market and formal banks). The data are for 45 ages. By converting the wide form to long form we expand the dataset from 45 observations (45 individual ages) to 90 observations 167 (90 cohort age-year pairs). A year-identifier variable, year (t-1=1998 and t = 2004), has been created. By providing separate time-series plot for the 45 LNpae individual units in the sample, we see a clear upward going tendency between 1998 and 2004. If we begin our graphical analysis by looking at a scatter plot of the dependent variable (LNpae) on the following key regressors (distance to the input market; highest grade attained by head of household; rain; and cotton income share), using data from all panel observations we find a clear downward trend between log expenditure p.a.e. and distance to input market, whereas log expenditure steeply increases until around attainment of six grade after which the log expenditure gradually declines. Rain lagged seems to have a higher upward going effect on log expenditure than rainfall in the actual agricultural season. Finally, we see a clear incremental upward going trend in log expenditure against cotton income share. According to the panel summary statistics on within and between variation, we find that for the variables (rain; household size; distance to food market; distance to input market), there is more variation across individuals (between variation) than over time (within variation), so within estimation may lead to considerable efficiency loss. Pooled OLS regression We start our pseudo-panel analysis with the pooled OLS regression for log p.a.e. expenditure using data for all cohorts in both years. We include as regressors: Education (highest grade attained by head of household); experience (Age) and quadratic in experience (Age squared); rain; rain lagged; landownership per household member; cotton income share of total income; cotton income share squared; distance to input market; and distance to nearest formal bank.157 Experience is time-varying in a 157 The few commercial formal banks are all exclusively situated in the district centres. Therefore the distance to the formal banks could be considered as a proxy for the distance to the district centre. Moreover this is the only possible variable, which the limited dataset allows us to consider as an instrument variable. See the discussion in section 5 below. 168 deterministic way as the sample comprises people who work full-time in all years, so experience increases by one year as t increments by one.158 Regressing yct on xct yields consistent estimates of , if the composite uct are uncorrelated with the regressors in the pooled model or population-averaged model, which assume regressors are exogenous: (4.4) yct = α + x‘ct + uct = α + x‘ct + αc + ct, The error uct is likely to be correlated over time for a given cohort, so we use clusterrobust standard errors that cluster on the cohort. Table 5.11: Pooled OLS with and without cluster-robust standard errors Whole Sample Robust Lnpae Coef. Std. Err. Highgrade 0,078 0,051 Age ,0640019** 0,026 Agesquare -,00086*** 0,000 Rain -0,025 0,016 Rain Lagged ,0881*** 0,020 Landownership pc ,4568*** 0,098 Cotton Income share 2,975 2,481 Cotton Inc Share squared -2,880 3,126 Distance to Input market -0,006 0,008 Distance to Bank -,01898*** 0,007 Constant 4,138*** 1,142 0,895 R2 Adj R2 N 90 Catchment Sub-sample Robust Default Default Coef. Std. Err. ,0783* 0,043 ,064** 0,026 -,00086*** 0,000 -0,025 0,016 ,0881*** 0,021 ,4568*** 0,081 2,975 2,223 -2,880 2,614 -0,006 0,008 -,01898*** 0,006 4,138*** 1,270 0,895 0,882 90 Counterfactual Sub-sample Robust Default Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. 0,042 0,055 0,042 0,034 0,048 0,040 0,048 0,035 -0,001 0,020 -0,001 0,022 -0,030 0,033 -0,030 0,042 0,000 0,000 0,000 0,000 0,000 0,000 0,000 0,000 -0,0378*** 0,011 -0,0378*** 0,012 -0,002 0,008 -0,002 0,011 0,028 0,018 0,0282* 0,016 0,055** 0,023 0,055** 0,023 0,193*** 0,061 0,193*** 0,062 0,195 0,154 0,195 0,135 0,284 1,228 0,284 1,792 -0,691 1,968 -0,691 1,547 -0,003 1,912 -0,003 2,028 0,838 2,199 0,838 1,746 -0,001 0,006 -0,001 0,007 -0,004 0,009 -0,004 0,008 -0,005 0,004 -0,005 0,005 -0,005 0,006 -0,005 0,006 10,43*** 0,895 10,434*** 1,050 5,847*** 1,911 5,847*** 1,782 0,672 0,672 0,447 0,447 0,631 0,364 90 90 78 78 Sourceμ Authors‘ estimations. The standard errors are small except for the two cotton share regressors. Moreover, the cluster-robust standard errors are smaller than the default standard errors for the following regressors: Age; Age square; Rain; Rain lagged (excl. Catchment sample); and distance to input market (excl. Control sample), and they are larger for the remaining regressors. Given the very high R2 it is almost certain that log p.a.e. expenditure is overpredicted in both 1998 and 2004. This is probably associated with how the cohort dataset was constructed. The failure to control for this error correlation leads to underestimation of standard errors. On the other hand, the difference between default and 158 Education is a time-invariant regressor, taking the same value each year for a given individual. However, since the education variable is the mean of all the head of households in the same cohort, it could be justified that education is varying, if we by education also mean life-long learning, which includes technical vocational education and training. 169 cluster-robust standard errors for pooled OLS is quite small given the small number of time periods (T=2). Table 5.12: Between Estimator with Default Standard Errors Whole Sample Lnpae Highgrade Age Agesquare Rain Rain Lagged Landownership pc Cotton Income share Cotton Inc Share squared Distance to Input market Distance to Bank Constant R2: within R2: between R2: overall N Number of groups Obs per group sd(u_i + avg(e_i.)) F(10, 34) Prob > F Catchment Sub-sample Counterfactual Sub-sample Between Estimator Default Between Estimator Default Coef. Std. Err. -0,032 0,045 0,015 0,022 0,000 0,000 0,024 0,015 0,014 0,023 -0,054 0,117 0,223 2,747 -0,712 3,094 -0,012 0,008 -,0124* 0,006 8,161*** 1,452 0,431 0,492 0,269 90 45 2 0,189 3,290 0,005 Coef. Std. Err. -0,017 0,048 0,018 0,023 0,000 0,000 -0,020 0,014 0,002 0,026 0,095 0,095 -1,191 2,538 0,297 2,942 -0,011 0,012 -0,012 0,007 11,814*** 1,551 0,131 0,544 0,225 90 45 2 0,223 4,060 0,001 Sourceμ Authors‘ estimations. Between Estimator Default Coef. 0,070 0,020 0,000 -0,012 0,063** 0,406* -0,217 0,088 -0,009 -0,007 4,722 Std. Err. 0,044 0,044 0,001 0,012 0,024 0,183 2,212 2,461 0,012 0,009 1,917 0,426 0,592 0,426 78 44 2 0,456 4,780 0,000 We next calculate the first-order autocorrelation coefficient for LNpae to be 0.1230 (whole sample). 45 observations are used to compute the autocorrelation at lag 1, -0.151, which provides a rough estimate of the intraclass correlation coefficient of the residuals that is far from indicating perfect anti-correlation (-1). The between estimator uses only between or cross-section variation in the data and is the OLS estimator from the regression of y i on x i . The Between Estimator estimates and standard errors are closer to those obtained from Pooled OLS (table 5.11) than those that could have been obtained from within estimation.159 Table 5.12 shows that the distance to the formal bank (district centre) correctly is negatively associated with the dependent variable, but only significant for the whole sample. The signs of the coefficients for the Lagged rain and Landownership regressors 159 Some groups have fewer than 3 observations therefore it was not possible to estimate correlations for those groups. 45 groups omitted from estimation. 170 correct for both the catchment and control areas, and they are significant at respectively the 0.05 and 0.1 level, but only for the control sample. Table 5.13 presents the results of our comparison of three cross-sectional timeseries regression models. And in table 5.14, several features emerge when we compare some of the panel estimators and associated standard errors, variance components estimates, and R2. The estimated coefficients vary considerably across estimators, especially for the time-varying regressors. This reflects quite different results according to whether within variation or between variation is used. Cluster-robust standard errors for the FE and RE models exceeds the default standard errors except for distance to input market and distance to bank in the former case and rain; rain lagged and distance to bank in the latter case. The various R2 measures and variance-components estimates also vary considerably across models. First-Difference Estimator Consistent estimation of in the fixed-effects (FE) model requires eliminating the αi. One way to do so is mean-difference, yielding the within estimator.160 An alternative way is to first-difference, leading to the first-difference estimator. This alternative has the advantage of relying on weaker exogeneity assumptions (Cameron and Trivedi, 2009). The first-difference (FD) estimator is obtained by performing OLS on the firstdifferenced variables: (4.5) 160 (yit – yi,t-1) = (xit – xi, t-1)‘ + ( it – i,t-1) The within estimator is traditionally favoured as it is the more efficient estimator if the 171 it are i.i.d. Table 5.13: RE Estimator: Comparison of cross-sectional time-series regression models. Catchment Sub-sample Lnpae Highgrade Age Agesquare Rain Rain Lagged Landownership pc Cotton Income share Cotton Inc Share squared Distance to Input market Distance to Bank Constant R2: within R2: between R2: overall N Number of groups Obs per group theta Wald chi2(11) / chi2(10) Prob > chi2 sigma_u sigma_e rho Log likelihood LR chi2(10) Prob > chi2 Likelihood-ratio test of sigma_u=0: Prob > chibar2 Counterfactual Sub-sample Maximum-likelihood random-effects GLS randomestimator effects estimator Robust Default GEE Populationaveraged estimator(i) Default Coef. Std. Err. 0,045 0,055 -0,005 0,020 0,000 0,000 -0.039*** 0,012 0,029 0,018 0.188*** 0,058 0,340 1,256 0,027 1,951 -0,001 0,005 -0,004 0,005 10,487*** 0,901 0,776 0,354 0,672 90 45 2 0,103 93553,2 0 0,099 0,283 0,108 Coef. Std. Err. 0,039 0,031 0,002 0,020 0,000 0,000 -0.037*** 0,011 0.028* 0,015 0.198*** 0,058 0,213 1,688 -0,026 1,910 -0,001 0,007 -0,006 0,005 10.39*** 0,975 Coef. 0,042 -0,001 0,000 -0.038*** 0.028** 0.193*** 0,284 -0,003 -0,001 -0,005 10.435*** Std. Err. 0,031 0,021 0,000 0,011 0,015 0,058 1,679 1,900 0,007 0,005 0,983 90 45 2 90 45 2 GLS randomeffects estimator Robust Coef. Std. Err. 0,048 0,040 -0,030 0,033 0,000 0,000 -0,002 0,008 0.055** 0,023 0,195 0,154 -0,691 1,968 0,838 2,199 -0,004 0,009 -0,005 0,006 5,847*** 1,911 0,476 0,544 0,447 78 44 1,8 170,11 0 44,25 0 0 0,747 0 0 0,125 0,340 0,025 0 . -30,668 100,35 0 0 1 Maximum-likelihood random-effects estimator Default Coef. 0,048 -0,030 0,000 -0,002 0.055** 0,195 -0,691 0,838 -0,004 -0,005 5,847*** Std. Err. 0,032 0,039 0,000 0,010 0,022 0,125 1,434 1,618 0,008 0,006 1,652 78 44 1,8 GEE Populationaveraged estimator(i) Default Coef. 0,049 -0,019 0,000 -0,002 0.055** 0.213* -0,486 0,495 -0,004 -0,004 5,592*** Std. Err. 0,032 0,037 0,000 0,010 0,022 0,125 1,432 1,603 0,008 0,006 1,627 78 44 1,8 57,75 0 0 0,175 0,611 0,049 0 -72,265 46,2 0 0 1 Notes: (i) Correlation exchangeable. Sourceμ Authors‘ estimations. Table 5.14a Panel Estimator, Comparison, Catchment Variable highgrade OLS_rob 0.0419 0.0546 Age -0.001 0.0202 Agesq -0.0001 0.0002 rain -0.0378 0.0113 rainlagged 0.0282 0.0183 Landowners~c 0.1931 0.061 cotincshare 0.2839 1.2282 cotincshar~q -0.0029 1.9123 distiput -0.0013 0.0055 distbank -0.0047 0.0043 _cons 10.4348 0.8947 N 90 r2 0.6721 r2_o r2_b r2_w sigma_u sigma_e rho Betweeneffects (BE) model -0.0174 0.0484 0.0176 0.0229 -0.0004 0.0003 -0.0201 0.0141 0.0017 0.0259 0.0945 0.0951 -1.1913 2.5376 0.2974 2.942 -0.011 0.0123 -0.0121 0.0075 11.8145 1.5507 90 0.544 0.2245 0.544 0.1307 Fixedeffects (FE) model FE_rob (i) 0.0502 0.0502 0.0378 0.0424 0.0944 0.0944 0.0551 0.0508 0.0012 0.0012 0.0004 0.0003 -0.0009 -0.0009 0.0181 0.0129 -0.0206 -0.0206 0.0201 0.0159 -0.1426 -0.1426 0.0889 0.0696 -1.7587 -1.7587 1.929 2.0335 2.4718 2.4718 2.147 2.1065 -0.0039 -0.0039 0.0066 0.0069 -0.0039 -0.0039 0.006 0.0068 4.4949 4.4949 2.122 1.9112 90 90 0.8844 0.8844 0.0172 0.0172 0.3606 0.3606 0.8844 0.8844 2.774 2.774 0.2835 0.2835 0.9897 0.9897 Table 5.14b Panel Estimator Comparison, Control GLS randomeffects (RE) model RE_rob (i) 0.0448 0.0448 0.0336 0.0549 -0.0047 -0.0047 0.0227 0.0226 -0.0001 -0.0001 0.0003 0.0003 -0.0386 -0.0386 0.0124 0.0123 0.0285 0.0285 0.0162 0.0187 0.1877 0.1877 0.0621 0.0634 0.3402 0.3402 1.7816 1.4581 0.0267 0.0267 2.0157 2.0074 -0.0014 -0.0014 0.0069 0.006 -0.0036 -0.0036 0.0054 0.0046 10.4869 10.4869 1.0586 0.9965 90 90 0.6715 0.3541 0.7763 0.0988 0.2835 0.1084 0.6715 0.3541 0.7763 0.0988 0.2835 0.1084 Variable highgrade OLS_rob 0.0478 0.0402 Age -0.0304 0.0328 Agesq 0.0002 0.0004 rain -0.0024 0.0085 rainlagged 0.055 0.023 Landowners~c 0.1954 0.1541 cotincshare -0.6905 1.9683 cotincshar~q 0.838 2.1987 distiput -0.0036 0.0094 distbank -0.0045 0.0065 _cons 5.8466 1.9113 N 78 r2 0.4469 r2_o r2_b r2_w sigma_u sigma_e rho Betweeneffects (BE) model 0.0701 0.0438 0.0196 0.044 -0.0003 0.0005 -0.0121 0.0117 0.0635 0.0239 0.4058 0.1832 -0.2174 2.2119 0.0881 2.4608 -0.0085 0.0125 -0.0068 0.009 4.722 1.9174 78 0.5917 0.4256 0.5917 0.4257 Notes: (i) vce(robust) uses the robust or sandwich estimator of variance. Sourceμ Authors‘ estimations. 172 Fixedeffects (FE) model FE_rob (i) 0.0198 0.0198 0.0631 0.0565 -0.1497 -0.1497 0.2338 0.2421 0.0021 0.0021 0.0013 0.0014 0.0037 0.0037 0.0255 0.0231 0.0347 0.0347 0.0474 0.0457 -0.0024 -0.0024 0.2674 0.2438 -1.4411 -1.4411 2.5926 2.7286 2.3725 2.3725 3.056 3.0741 0.0014 0.0014 0.0149 0.0122 -0.0055 -0.0055 0.0163 0.0159 8.5311 8.5311 8.7127 8.3592 78 78 0.5468 0.5468 0.0814 0.0814 0.0009 0.0009 0.5468 0.5468 0.8738 0.8738 0.7473 0.7473 0.5776 0.5776 GLS randomeffects (RE) model RE_rob (i) 0.0478 0.0478 0.035 0.0445 -0.0304 -0.0304 0.0421 0.0432 0.0002 0.0002 0.0005 0.0005 -0.0024 -0.0024 0.0106 0.0089 0.055 0.055 0.0234 0.0214 0.1954 0.1954 0.1354 0.1556 -0.6905 -0.6905 1.547 1.8223 0.838 0.838 1.7458 2.0804 -0.0036 -0.0036 0.0084 0.0098 -0.0045 -0.0045 0.006 0.0065 5.8466 5.8466 1.7821 1.6822 78 78 0.4469 0.5441 0.4763 0 0.7473 0 0.4469 0.5441 0.4763 0 0.7473 0 First-differencing has eliminated αi in (4.4), so OLS estimation of this model leads to consistent estimates of in the FE model. The coefficients of time-invariant regressors are not identified, because then xit – xi,t-1 = 0, as was the case for the within estimator (ibid.). As expected, the coefficient for education is not identified because Age here is time-invariant in the whole sample only. The signs of the coefficients for the other regressors do not change compared with the other estimators (table 5.14). And again it is the regressors: Lagged rain; landownership and distance to bank (table 5.13) that are statistically significant in the whole sample, whereas it is the regressors: Age; age squared and rain in the catchment sub-sample. The FD estimator like the within estimator, provides consistent estimators when the individual effects are fixed. For our panel with T=2, the FD and within estimators are equivalent. Thus, table 5.15 gives the results of a simplified version of model (3.18) above that explain the growth in p.a.e. consumption between 1998 (―t-1‖) and β004 (―t‖). Table 5.15: First-Differences Estimator with Cluster-Robust Standard Errors Whole Robust Std.Err. Coef. D.Lnpae Coef. highgrade D1. 0,066 0,082 Age D1. (dropped) Agesq D1. (dropped) rain D1. -0,049 0,032 rain_lag D1. 0,117** 0,047 Landowners~c D1. 0,477*** 0,128 cotincshare D1. 3,174 3,408 cotincshar~q D1. -4,204 4,151 distiput D1. -0,005 0,014 distbank D1. -0,0205** 0,009 R2 0,940 N 45 ,132* Catchment Robust Std.Err. Coef. Robust Std.Err. Coef. 0,070 0,050 0,045 0,094* Counter Factual Robust Robust Std.Err. Coef. Std.Err. Robust Std.Err. Coef. 0,049 0,020 0,062 0,054 -0,150 0,266 0,001*** 0,000 0,002 0,002 -0,001 0,014 -0,050** 0,021 0,004 0,051 0,000 0,063 0,025 0,010 0,023 ,071** 0,029 -0,021 0,017 0,032 0,022 0,035 0,050 0,040 0,054 ,544*** 0,118 -0,143* 0,074 0,121** 0,053 -0,002 0,268 -0,054 0,239 3,888 3,636 -1,759 2,148 0,315 2,105 -1,441 2,995 -1,918 3,088 -5,377 4,813 2,472 2,225 0,646 2,741 2,373 3,374 2,500 3,278 -0,004 0,007 -0,005 0,007 0,001 0,013 0,001 0,013 -,024** 0,010 0,935 45 -0,004 0,007 0,884 45 Sourceμ Authors‘ estimations. 173 0,005 0,007 0,803 45 -0,006 0,017 0,547 34 -0,008 0,013 0,498 34 5.5. Discussion of Estimation Results 5.5.1. Specification tests of the functional form A key assumption maintained in section 5.2 is that the functional form was correctly specified for the estimated relationship of the list of included regressors. In this section we will check the validity of this assumption. Formal specification tests can distinguish between systematic lack of fit and random sampling errors. If the zeroconditional-mean assumption: (5.1) E[e | x1, x2, …, x5] = 0 is violated, the coefficient estimates are inconsistent. The three main problems that cause the zero-conditional-mean assumption to fail in a regression model are: Improper specification of the model; endogeneity of one or more regressors; or measurement error of one or more regressors (Baum, 2006). This section reports the results of formal specification tests of models (4.1)–(4.2), whereas section 5.5.2 addresses endogeneity and measurement errors. First consider testing the specification of the parametric model (4.1). The consistency of the linear regression estimator requires that the sample regression function corresponds to the underlying regression function or true model for the response variable y = LNpae98 (or LNpae04). The cost of omitting relevant variables is high. A variable mistakenly excluded from our model is unlikely to be uncorrelated in the population or in the sample with the regressors (Table A8.1-2). Misspecification of the functional form Our model (5.1) that includes the appropriate five regressors may be misspecified because of the model may not reflect the algebraic form of the relationship between the response variable and those regressors. In the words of Baum(2006), this problem may be easier to deal with than the omission of relevant variables. In a misspecification of the functional form, we have all the appropriate variables at hand and only have to choose the appropriate form in which they enter the regression function. 174 Ramsey's RESET It is seen (table A9) that both the parsimonious Ramsey's (1969) omitted-variable regression specification error test (RESET),161 which augments the regression using the  second, third, and fourth powers of the fitted values y series (of LNpae) as well as the Ramsey RESET test using the powers of the individual independent regressors themselves reject RESET's null hypothesis of no omitted variables for the model, albeit at the 10% significant level in the first test.162 Since the hypothesis is rejected, then these powers cannot be excluded from the regression without compromising the level of explication of the dependent variable. That indicates that the original regression was not specified correctly. We re-specify equation (4.1) to include: The square of age (age2), whether household own bicycle (s10q6/assetownd07) and whether head of household ever attended school (s4q5).163 We see that the respecified and extended model's values no longer reject the RESET at the 1% significance level and so we may conclude that the regression was specified correctly in 1998 only (table A9). The relationship between squared age (age2), school attendance (s4q5) and bicycle ownership (s10q6) appears to be nonlinear (although with wrong pattern of signs on their coefficients). Specification plots Next, we evaluate the specifications of the model and the extended model by use of two types of plots. First we graph the residuals on the y-axis versus the predicted (i.e. fitted) values on the x-axis.164 Then we plot the residuals against a specific regressor. The residuals ―bounce randomly‖ around the 0 line (i.e. linear is reasonable). ζo one residual ―stands out‖ from the basic random pattern of residuals (i.e. no outliers).165 The residuals 161 The Wald test is a statistical test, typically used to test whether an effect exists or not. In other words, it tests whether an independent variable has a statistically significant relationship with a dependent variable. 162 The idea that a polynomial constructed from these estimated values can be seen as a ―reduced form‖ for many different combinations of powers and cross-products involving the independent variables. 163 Unfortunately, there are only 24 observations in rural Eastern Province in 1998 if we use the variable "highest grade ever attended by head of household," which would have given us an opportunity to measure the association between the level of education and level of poverty. 164 This helps to identify non-linearity, outliers, and non-constant variance. 165 We have deliberately from the outset chosen to disregards the commercial large-scale farmers stratum. 175 roughly form a ―horizontal band‖ around 0 line (i.e. constant variance). This in turn indicates that there doesn't seem to be a problem with either models. With regards to the residuals for the continuous variables – age2, distance to input market, and cotton income share – the assumption of homoskedasticity doesn't seem to be challenged by the residuals on the y axis -versus- the values of a predictor on the x axis plots, which through the random scatter indicated that the models are probably good. The only exception might be the case of „household ownership of motor vehicles‟ in the  original model, where the spread in the estimator (i.e. the observed residuals) ei = yi - y are larger when the household expressed no ownership, leading to the suspicion that the errors are heteroscedastic in both 1998 and in 2004. Outlier statistics and measures of leverage To evaluate the adequacy of the specification of the fitted models, we will consider evidence relating to the models' robustness to influential data.166 We calculate a measure of each data point's leverage, calculated from the diagonal elements of the "hat matrix", hj = xj(X'X)-1x'j, where xj is the jth row of the regressor matrix. From table 5.8 displaying summary statistics of the dependent variable LNpae98 (LNpae04), we see that it ranges from minimum 4,605 (6,629) to maximum 12,662 (13.226). The five largest values of the leverage measure are listed in tables A10.1-2. The five largest squared residuals are listed in tables A11.1-2. These leverage values versus the (normalized) squared residuals are displayed in figures A6.1-2, which show that there in 1998 are around a dozen of points with very high leverage or very large squared residuals.167 Several of the largest values of leverage or the squared residuals correspond to the extreme values of the pae98 recorded in the dataset. 166 An outlier in a regression relationship is a data point with an unusual value and a high degree of leverage on the estimates. An outlier may be an observation associated with a large residual (in absolute terms), a datapoint that the model fits poorly. On the other hand, an unusual data point that is far from the center of mass of the xj distribution may also be an outlier, although the residual associated with that data point will often be small because the least-squares process attaches a squared penalty to the residual in forming the least-squares criterion (Baum, 2006). 167 A large value of leverage does not imply a large squared residual (Baum, 2006). 176 DFITS statistics Every data point affects the estimates of the slope, the intercept, the predicted values, and the error variance of the regression line to some degree; an influential observation does so more than other points. Hence, we will next compute the DFITS statistics of Welsch and Kuh(1977) to provide a summary of the leverage values and magnitudes of residuals.168 The DFITS measure is a scaled difference between the insample and out-of-sample predicted values for the jth observation. DFITS evaluates the result of fitting the regression model including and excluding that observation. Tables A.12.1-2 displays the 55 large values of DFITS for the 1998 dataset and 27 large values of DFITS for the 2004 dataset for which cutoff = 1. That is, about 6% of the 1998 observations and 4% of the 2004 observations are flagged by the DFITS cutoff criterion. Many of those observations associated with large positive DFITS have the topcoded values for pae98, however the magnitude of the positive and negative DFITS are more or less the same in both the 1998 and 2004 datasets. The identification of top-coded values that represent an arbitrary maximum recorded pae98 suggests that we consider a different estimation technique for this model that can properly account for the censored nature of the pae98. DFBETA statistics Finally, we compute DFBETAs for the discrete Treatment dummy variable "infrastructure" (ℓ) regressor in our regression model, which measures the distance that this regression coefficient would shift when the jth observation is included or excluded from the regression, scaled by the estimated standard error of the coefficient.169 168 Since RStudent is the first component of DFFits, an outlier is more likely than another observation to be identified as strongly influencing the predicted value. Even though outliers are more distant from the regression line than others points are, they still tend to tilt the regression line towards them. The second component of DFFits is the ratio, which increases as the observation‘s leverage increases. This means that observations with independent variable values far from the mean tend to tilt the regression line more strongly than do the central observations. 169 One rule of thumb suggests that a DFBETA value greater than unity in absolute value might be reason for concern since this observation might shift the estimated coefficient by more than one standard error (Baum, 2006). 177 Compared to the DFITS measure, in tables A13.1-2 we see a similar pattern for the DFBETA for "infrastructure" with 7.6% of the 1998 sample of 919 observations and 7.1% of the 2004 sample of 647 observations exhibiting large values of this measure. As with DFITS, the large positive values are more or less the same size in magnitude as their negative counterparts in both 1998 and 2004. Around 45% of the positive values are associated with the top-coded pae98 values above ZMK50,000 in 1998. These presumably better off rural households have values well in excess of its minimum or mean. Thus, there is evidence of many data points with a high degree of leverage. Whether these pae98 / pae04 data have been improperly measured, we can only speculate about this. But these observations in particular have been identified by the DFITS and DFBETA measures. Removing the bottom-coded and top-coded observations from the sample would remove rural households from the sample non-randomly, affecting the wealthiest and poorest rural households.170 Mathematically, measurement error (commonly termed errors-in-variables) has the same effect on an OLS regression model as endogeneity of one or more regressors. This measurement error is of concern, because the economic behaviour we want to model that of the rural households in Eastern Province - presumably is driven by the actual measures, not our mis-measured approximations of those factors. So if we fail to capture the actual measure, we may misinterpret the behavioural response (Baum, 2006). 5.5.2. Instrumental-variable estimators The zero-conditional mean assumption must hold for us to use linear OLS regression. There are three common instances where this assumption may be violated: Endogeneity (simultaneous determination of response variable and regressors), omittedvariable bias, and errors in variables (measurement error in the regressors) since households are unlikely to be able to recall household expenditure. The solution to each is the same econometric tool: the instrumental-variables (IV) estimator (Baum, 2006). 170 A version of the tobit model, two-limit tobit, can handle censoring of both lower and upper limits (Baum, 2006). 178 To derive consistent estimators of (4.1), we must find an IV that satisfies two properties: The instrument z must be uncorrelated with e (that is, the orthogonality assumption) but must be highly correlated with xj. A variable that meets those two conditions is an IV or instrument for xj that deals with the correlation of xj and the error term (Johnston and DiNardo, 1997a, Baum, 2006, Wooldridge, 2002). In order to capture the transitory component in household expenditure, we use rainfall as an instrument.171 As most of the rural households in our sample rely on crop yields as the main source of income, rainfall can explain a non-trivial share of the intertemporal variation in total household expenditure (LNpae). We argue that a rainfallinduced variation in household income will be less tainted by measurement error. Rainfall fluctuations arguably captures a transitory and exogenous component in household's income and expenditure, uncorrelated to life-cycle decisions, knowledge or other variables that may enter the households preferences over choice of cash crops. Thus, our identification strategy rests on the assumption that rainfall affects consumption outcomes only via the total income variable and not via some omitted variable. A potential scenario is that households are able to completely smooth output in the event of a weather shock by adjusting their consumption of leisure. This type of income smoothing suggests that a realized weather shock can affect the consumption of leisure without affecting the observed level of rural production. As noted by Rosenzweig and Wolpin(2000), the credence of using weather variation as an instrument for rural income rests on how the market structure is defined, and on how expenditure and income is observed. Rainfall has a decisive impact on expenditure in our sample, implying that rural households in Eastern Province are unable to borrow and save across transitory income shocks. The rural households are completely liquidity constrained, because they 171 We do not have more than one potential candidate instrument. A valid instrument variable z requires E[u|z]= 0 (exogeneity, i.e. z is validly excluded from the outcome equation of interest. In other words, z has no direct effect on y but only indirectly through its impact on x) and ωov(z,x) ≠ 0 (relevance of the instrumentμ F>0). If the instrument is weak, βSδS is no longer reliable since the estimator will not only be badly biased but the estimator will also have a nonnormal sampling distribution making statistical inference meaningless. 179 are unable to save or borrow across aggregated income shocks to achieve at least perfect intertemporal consumption smoothing. Empirical implementation172 Our extended baseline regression equation is (4.1), where we model the logarithm of total household expenditure p.a.e. (LNpae) as a function of a couple of continuous variables: Distiput and age2 (distance to input market and age of head of household squared); and a set of indicator variables: Stratum124, s10q8, s10q6, and s4q5 (rural household stratum, motor vehicle ownership, bicycle ownership, and school attendance), and infrastructure, and indicator for residency in one of the five districts affected by the EPFRP. The endogenous variable is cotincshare, cotton sales as a share of total household income. Here we do not consider LNpae and cotincshare are simultaneously determined, but rather that cotincshare cannot be assumed independent of the error term: The same correlation that arises in the context of an endogenous regressor in a structural equation (Baum, 2006). The cotincshare is instrumented with two factors excluded from the equation (4.1): The average yearly rainfall in each district in Eastern Province for the agricultural season of the year of the survey (1998/1999 and 2004/2005) and the average yearly rainfall lagged with one year (1997/1998 and 2003/2004).173 We fit the IV model using first-stage regression to evaluate the degree of correlation between these two factors and the endogenous regressor. From tables A14.ab the first-stage regression results suggest that one of the two excluded instruments, namely the one year lagged rainfall variable is highly correlated with the endogenous variable (cotincshare) in the 1998/99 agricultural season and that this is the case for both the lagged (rain04) and rainfall during the 2004/05 agricultural season (rain05). The exception is the rainfall variable rain99. 172 Since we do not have two candidate instruments, we won't use the alternative approach, 2SLS, which combines multiple instruments into one optimal instrument, which can be used in the simple IV estimator. The order condition is often stated as requiring that there be at least as many instruments as endogenous variables. The order condition is necessary, but not sufficient, for the rank condition to hold (Baum, 2006). 173 Instruments that satisfy the rank condition but are not sufficiently correlated with the endogenous variables for the large-sample approximations to be useful are known as weak instruments (Baum, 2006). 180 Tables A15.a-b show that the endogenous regressor cotincshare has a distinguishable negative IV coefficient in 1998 but a positive coefficient in 2004 as expected in the latter case. Thus, conditioning on the other factors included in the equation, cotincshare does seem to play a role in determining the LNpae98 and especially in the case of LNpae04, although the sign of the coefficient estimate doesn't agree with the predictions of theory and empirical findings in 1998. Furthermore, in 1998 only the distance to input market coefficient estimate seem to agree with the predictions of theory. One cannot directly test for the exogeneity assumption but there are indirect ways of testing it. Table A16.1 shows the Anderson-Rubin-Sargan-Basmann test results for the validity of overidentifying structural restrictions, which signals a strong rejection of the null hypothesis that the instruments are uncorrelated with the error term (i.e. the disturbance process) and thereby suggests that we should not be satisfied with this specification of the equation according to the 1998 dataset. In the 2004 dataset the instrument is weak given the F-statistics in our specification is less than 2. This leads to the weak instruments problem,174 that is, instruments that are only weakly correlated with the included endogenous variables. Unfortunately, weak instruments pose considerable challenges to inference using GMM and IV methods (Stock et al., 2002). From table A17.1 we see that the endogenous regressor cotincshare still does play a role in the equation. The Hansen J statistics is the GMM equivalent to the Sargan test.175 The independence of the instruments and the disturbance process is called into question by the strong rejection of the J test null hypothesis. In tables A18.1-2 we test whether the subset of the excluded rainfall instruments is appropriately exogenous. The equation estimated without suspect instruments, free of one additional orthogonality condition on rain99, doesn't have a Hansen J statistics, whereas It is not useful to think of weak instruments as a ―small sample‖ problem. ψound et al.,(199η) provided an empirical example of weak instruments despite having 329,000 observations (Stock et al., 2002). 175 Hansen's J is the most common diagnostic used in the GMM estimation to evaluate the suitablity of the model. a rejection of the null hypothesis implies that the instruments do not satisfy the required orthogonality conditions (Baum, 2006). 174 181 the C statistic for the instrument, rain99, tested is highly significant. Hence rain99 does not appear to be valid in this context. To evaluate whether we have found a more appropriate specification, we reestimate the equation with the remaining instrument rain98. From the results shown in tables A19.1-2 we see that cotincshare no longer appears a significant regressor and the equation's J statistics is zero again. The following regressors: Stratum124, s10q8, s10q6, and s4q5 all seem to be playing a role in this from of the estimated equation. Detection of heteroskedasticity (unequal variance of the errors) can be achieved by many different tests under the assumption of a linear statistical model of the form (4.1). In table 5.16 below we compute several of the tests for heteroskedasticity appropriate in the IV context from the last regression reported in table A19.1. All of the tests using the 1998 dataset signal a problem of heteroskedasticity in the estimated equation's disturbance process. Table 5.16: Testing for heteroskedasticity in the IV context in 1998 and 2004 IV heteroskedasticity test(s) using levels of IVs only Ho: Disturbance is homoskedastic Pagan-Hall general test statistic 19,4 Pagan-Hall test w/assumed normality 22,574 White/Koenker nR2 test statistic 21,554 Breusch-Pagan/Godfrey/Cook-Weisberg 24,855 1998 Chi-sq(9) P-value Chi-sq(9) P-value Chi-sq(9) P-value Chi-sq(9) P-value = = = = 0,0220 0,0072 0,0104 0,0031 5.802 4.786 12.062 12.177 2004 Chi-sq(9) P-value Chi-sq(9) P-value Chi-sq(9) P-value Chi-sq(9) P-value = = = = 0.7596 0.8525 0.2099 0.2035 IV heteroskedasticity test(s) using fitted value (X-hat*beta-hat) & its square Ho: Disturbance is homoskedastic 1998 Pagan-Hall general test statistic 7,711 Chi-sq(2) P-value = Pagan-Hall test w/assumed normality 8,961 Chi-sq(2) P-value = White/Koenker nR2 test statistic 8,381 Chi-sq(2) P-value = Breusch-Pagan/Godfrey/Cook-Weisberg 9,665 Chi-sq(2) P-value = 0,0212 0,0113 0,0151 0,008 2.767 2.267 4.851 4.897 2004 Chi-sq(2) P-value Chi-sq(2) P-value Chi-sq(2) P-value Chi-sq(2) P-value = = = = 0.2507 0.3219 0.0884 0.0864 Notes: Pagan-Hall statistics is robust to the presence of heteroskedasticity elsewhere in a system of simultaneous equations and to non-normally distributed disturbances. White's general test (White, 1980), or its generalization by Koenker (1981), also relaxes the assumption of normality underlying the BreuschPagan test (see Deaton, 1997:79).176 The Breusch-Pagan (1979) and White tests for heteroskedasticity can be applied in 2SLS models, but Pagan and Hall(1983) point out that they will be valid only if heteroskedasticity is present in that equation and nowhere less in the system. The other structural equations in the system corresponding to the endogenous regressors must also be homoskedastic even though they are not being explicitly estimated. Source: Author's calculations. 176 The Breusch-Pagan test for heteroskedasticity uses a test-equation: the squared residuals divided by the residual variance are explained by all exogenous variables. The test statistic is computed as half the difference between the Total Sum of Squares and the Sum of Squared Residuals, which has a Chi-square distribution. Warning: this test should only be used if the endogenous variable is NOT used as lagged exogenous variable and if the number of observations is VERY LARGE. All OLS assumptions should be satisfied, including normality of the error term. 182 Weighted Least Squares Weighted least squares provides one method for dealing with heteroscedasticity. For our data, we obtain the results shown in table 5.17. In the 1998 data the sign of the coefficient of cotton income share change from negative in the OLS estimation to positive in the WLS estimation, plus with a smaller magnitude. The other noteworthy change is the R2 falls from 19.22% in the OLS model to 15.9% in the WLS model in 1998. In the 2004 data the coefficient sign change for three regressors: Distance to input market; Infrastructure dummy; and School attendance. Moreover, stratum becomes significant and motor vehicle significant at the 0.01 level from the 0.1 level. Finally, the R2 contrary to the 1998 data goes slightly up in the 2004 dataset. Table 5.17: Weighted Least-Squares Estimator vs OLS estimator, 1998 & 2004 1998 WLS LNpae98 Cotton Income Share Stratum, excl. Large Distance Inputmkt Motorvehicle Ownership Infrastructure Age Age Squared Bike Ownership School Attendance Constant N R2 Coef. 0,083 0,226*** -0,002 -1,885*** -0,093 -0,041** 0,000 -0,427*** -0,353*** 11,974*** 873 0,159 2004 OLS Robust Std. Err. 0,219 0,048 0,002 0,314 0,107 0,020 0,000 0,100 0,109 0,548 Coef. Std.Err. -0,104 0,188 0,247*** 0,038 -0,003* 0,002 -1,591*** 0,286 -0,006 0,085 -0,015 0,014 0,000 0,000 -0,368*** 0,076 -0,420*** 0,078 11,220*** 0,433 919 0,192 WLS Robust Std. Err. 0,158 0,118 0,002 0,257 0,151 0,021 0,000 0,110 0,123 0,577 Coef. 0,158 -0,258** 0,001 0,726*** 0,064 -0,029 0,000 -0,049 0,058 11,380*** 647 0,019 OLS Coef. 0,132 -0,058 -0,001 0,54* -0,037 -0,024 0,000 -0,005 -0,012 11,363 Std.Err. 0,139 0,118 0,002 0,314 0,126 0,019 0,000 0,100 0,109 0,570 647 0,017 Source: Author's estimations. Testing the relevance of instruments We will now test whether the instrument variable are highly correlated with the included endogenous variable - cotincshare - by examining the fit of the first-stage regressions.177 The relevant test statistics here relate to the explanatory power of the excluded instruments in these regressions. The statistics proposed by Bound, Jaeger, and Baker(1995) can diagnose instrument relevance only in the presence of one endogenous regressor (Baum, 2006).178 177 The first-stage regressions are reduced-form regressions of the endogenous regressors, x1, on the full set of instruments, z. 178 When multiple endogenous regressors are used, other statistics are required. 183 The tables A20.1-2 illustrate the weak-instrument problem with a variation on the LNpae equation using rain98 and rain99 as instruments.179 In order to ensure that we have found a strong instrumental variable we provide Shea‟s(1997) partial R2 statistic and its associated F-statistic. In the first-stage regression results, Shea's partial R2 statistic is very small for this equation indicating that the instrument is insufficiently relevant to explain the endogenous regressor in both 1998 and 2004, and the CraggDonald statistics rejects its null hypothesis of under identification. The Anderson canonical correlation statistic rejects its null hypothesis, suggesting that the instrument may be adequate to identify the equation.180 Finally, the redundant (rain98) option indicates that rain98 does provide useful information to identify the equation. For one endogenous regressor - cotincshare -, a F statistics less than 10 is not the case for the 1998 dataset,181 contrary to the 2004 dataset, which is cause for concern (Staiger and Stock, 1997:557) Durbin-Wu-Hausman tests for endogeneity in IV estimation There are three equivalent ways of obtaining the Durbin component of the DurbinWu-Hausmann (DWH) statistics. The different commands implement distinct versions of the tests, which although asymptotically equivalent can lead to different inference from finite samples (Baum, 2006). The first method fit the less efficient but consistent model using IV. Then fit the fully efficient model. The comparison in table 5.18 is restricted to the point estimate and estimated standard error of the endogenous regressor, cotincshare; the Hausmann test statistics accept the exogeneity of the rain98 variable. The command also warns of difficulties computing a positive-definite covariance matrix. The small chi2 value indicates that estimation of the equation with regress yields consistent results. 179 There is the familiar difficulty of finding convincing instruments, and it is usually easier to justify the role of instruments such as assets and lagged income as predictors of e.g. permanent income than it is to define their absence from the direct determination of consumption in the equation of interest (Deaton, 1997:352). Although average rainfall is predictably difference from place to place, the deviation of each year‘s rainfall from its local mean is serially uncorrelated and thus unpredictable (op.cit., p.γηγ). 180 Canonical Correlation Measure of the strength of the overall relationships between canonical variates (also referred to as linear composites, linear compounds, and linear combinations) for the independent and dependent variables. In effect, it represents the bivariate correlation between the two canonical variates. 181 One recommendation when faced with a weak-instrument problem is to be parsimonius in the choice of instruments. 184 Table 5.18: Durbin-Wu-Hausmann Tests for Endogeneity in IV estimation 1998 stratum124 distiput s10q8 infrastruc~e s1q3b age2 s10q6 s4q5 _cons (b) iv .2031568 -.0039727 -1.690421 -.0330978 -.0121234 -.0000538 -.4330964 -.4429078 11.5889 ---- Coefficients ---(B) (b-B) sqrt(diag(V_b-V_B)) . Difference S.E. .2510582 -.0479013 .0215904 -.0032202 -.0007525 .0003487 -1.583606 -.1068143 .0481315 -.0025029 -.0305949 .0138882 -.0151452 .0030217 .0013891 -6.21e-06 -.0000476 .0000215 -.3649632 -.0681332 .0316846 -.4174164 -.0254913 .0114684 11.19786 .3910402 .1785454 b = consistent under Ho and Ha; obtained from ivreg2 B = inconsistent under Ha, efficient under Ho; obtained from regress Test: Ho: difference in coefficients not systematic chi2(2) = = (b-B)'[(V_b-V_B)^(-1)](b-B) = 5.02 Prob>chi2 = 0.0812 (V_b-V_B is not positive definite) 2004 stratum124 distance11 assetownd09 infrastruc~e s1q3b age2 assetownd07 s4q5 _cons (b) iv -.4984822 .0019052 -.8796261 -.0354586 -.0585523 .0007242 .3288292 .2676141 11.82469 ---- Coefficients ---(B) (b-B) sqrt(diag(V_b-V_B)) . Difference S.E. -.0484214 -.4500609 .0763337 -.0012524 .0031576 .0005356 .57119 -1.450816 .2460694 -.0369459 .0014874 .0002523 -.023063 -.0354893 .0060193 .0001812 .0005429 .0000921 -.0119875 .3408167 .0578051 -.018386 .2860001 .0485078 11.3528 .471887 .0800356 b = consistent under Ho and Ha; obtained from ivreg2 B = inconsistent under Ha, efficient under Ho; obtained from regress Test: Ho: difference in coefficients not systematic chi2(1) = (b-B)'[(V_b-V_B)^(-1)](b-B) = 34.76 Prob>chi2 = 0.0000 (V_b-V_B is not positive definite) Source: Author's calculations. Tables A20.3-4 illustrates the second method, which fits the fully efficient model and specifies the regresssors to be tested. The second method's C test statistic agrees qualitatively with that from Haussman by accepting the exogeneity/orthogonality of the rain98 variable. Finally table A20.5 illustrates the method. The test statistic is identical to that provided by the C Statistic in table A20.1. All forms of the test agree that the estimation of this equation with linear regression yields consistent results. The regressor cotincshare must be considered exogenous in the fitted model. Quantile Regression We now turn to the "identically distributed" assumption, and consider the consequences of heteroskedasticity. Just as lack of independence appears to be the rule rather than the exception, so does heteroskedasticity seem to be almost always present in survey data (Deaton, 1997).182 We follow Deaton(1997) who suggests that the computation of quantile regressions is useful, both in its own right, because quantile regression estimates will often have better properties than OLS, as a way of assessing the 182 Even when individual behaviour generates homoskedastic regression functions within strate or villages, but there is heterogeneity between villages, there will be heteroskedasticity in the overall regression function. In the presence of heteroskedasticity, OLS is inefficient and the usual formulas for standard errors are incorrect (Deaton, 1997). 185 heteroskedasticity in the conditional distribution of the logarithmic transformation of p.a.e., and as a stepping stone to the nonparametric methods. The tables 5.19.1-4 and tables A21.1-2 shows the quantile regression outputs corresponding to the 20th, 40th, 60th and 80th percentile in the distribution of LNpae98 and the distribution of δζpae04. The slopes that are the response coefficients of the covariates of these four regression functions differ. These differences and the different spread between the regression functions show the increase in the conditional variance of the regression among better-off rural households.183 183 These regressions do not tell us anything about the causal processes that generate the differences, but they present the data in an interesting way that can be suggestive of ideas for a deeper investigation (Deaton, 1997). 186 Table 5.19.1 Quantile 1: 20%, 1998 LNpae98 observations LNpae98 observations LNpae98 observations stratum124 (1) cotincshare (2) distiput (3) .2716251** (.1373107) 186 -.0782083 (.5336642) 186 .001294 (.0065642) 186 .168185** (.0684282) 186 -.0343721 (.2667184) 186 -.0014761 (.0029171) 186 .1700681** (.069249) 186 .1138567 (.8488862) 186 -.001169 (.0033632) 186 s10q8 infrastructure s1q3b age2 s10q6 s4q5 _cons (4) (5) (6) (7) (8) (9) (10) Panel A: Quantile Regression -.4652535 .1687908 .0015369 -.0001599 -.6087141** -.1944722 7.488913 (.3969689) (.2379402) (.034054) (.000338) (.2355047) .2209168 1.050488 186 186 186 186 186 186 186 Panel B: OLS Regression -.1687206 .0944224 .0054081 -.0001351 -.450483*** -.1376368 7.509946 (.7416053) (.1163264) (.0186177) (.0001829) (.113021) (.1075116) (.9047165) 186 186 186 186 186 186 186 Panel C: IV Regression -.0933139 .0933699 .0056862 -.0001356 -.4348589*** -.1381655 7.388749 (.8479329) (.1165689) (.0186953) (.000183) (.1414588) (.1076443) (1.119847) 186 186 186 186 186 186 186 Standard errors in brackets, * significant at 10%, ** significant at 5%, *** significant at 1%. Table 5.19.2 Quantile 1: 20%, 2004 stratum124 cotincshare (1) (2) LNpae98 observations LNpae98 observations LNpae98 observations .1877515 (.2221348) 130 -.4743539 (.3468837) 130 .0086425 (.1093686) 130 -.072652 (.1359383) 130 .1117769 (.246006) 130 -.7570402 (1.431557) 130 distiput (3) s10q8 (4) infrastructure s1q3b age2 (5) (6) (7) Panel A: Quantile Regression -.0004347 .3145396 .1307892 -.034932 .0004011 (.0052569) (.4758589) (.3070168) (.059044) (.0006046) 130 130 130 130 130 Panel B: OLS Regression .0029035 .0541163 .0571551 -.0430928** .0004307** (.0022261) (.2024322) (.1214499) (.0204027) (.0002074) 130 130 130 130 130 Panel C: IV Regression .0014271 .1494557 -.0130489 -.0370546 .0003344 (.0039288) (.2982788) (.1979758) (.0257289) (.0003037) 130 130 130 130 130 s10q6 (8) s4q5 (9) _cons (10) -.4790948** -.1251921 9.632436 (.2295752) (.2738291) (1.569067) 130 130 130 -.1509163 -.1656449 10.52856 (.0936526) (.1023013) (.5196166) 130 130 130 -.1774808 -.1686605 10.52329 (.1169498) (.1127629) (.5719717) 130 130 130 Standard errors in brackets, * significant at 10%, ** significant at 5%, *** significant at 1%. Source: Author's calculations. Table 5.19.3 Quantile 4: 80%, 1998 LNpae98 observations LNpae98 observations LNpae98 observations stratum124 (1) cotincshare (2) .0656825 (.0703631) 188 -.1010467 (.4745047) 188 .0366734 (.0312563) 188 -.1632588 (.2288021) 188 .0434536 .0334405 188 .1065753 .5142713 188 distiput (3) s10q8 infrastructure s1q3b age2 s10q6 s4q5 _cons (4) (5) (6) (7) (8) (9) (10) Panel A: Quantile Regression -.004214 -1.378079*** .5049315** -.0442934 .0004334 .112295 -.1021425 13.07247 (.0041808) (.3869489) (.2162816) (.0402143) (.0004523) (.1939267) (.2243541) (.9556869) 188 188 188 188 188 188 188 188 Panel B: OLS Regression -.0034037** -.7335591*** .3297382*** -.0097287 .0000602 .0220449 -.0932895 11.47478 (.0017186) (.1617527) (.090622) (.0167093) (.0001824) (.0809641) (.0956322) (.4021212) 188 188 188 188 188 188 188 188 Panel C: IV Regression -.0033837** -.7149364*** .3453743*** -.0111327 .0000793 .0248216 -.1114647 11.43999 .0017256 .1654593 .0948021 .0169444 .000186 .0814176 .1008837 .4080262 188 188 188 188 188 188 188 188 Standard errors in brackets, * significant at 10%, ** significant at 5%, *** significant at 1%. Table 5.19.4 Quantile 4: 80%, 2004 stratum124 cotincshare (1) (2) LNpae98 observations LNpae98 observations LNpae98 observations -.0969562 (.4026205) 142 .3841689 (.4181356) 142 .0168404 (.1321313) 142 .1963199 (.1407112) 142 .1104283 (.2428376) 142 -.6054509 (1.660179) 142 distiput (3) s10q8 (4) infrastructure s1q3b age2 s10q6 s4q5 _cons (5) (6) (7) (8) (9) (10) Panel A: Quantile Regression .0006647 .0823007 -.0983912 .0149554 -.0001947 .1317238 .1019841 12.79436 (.0080136) (.63335) (.349107) (.0444247) (.0004248) (.3000069) (.2936413) (1.443234) 142 142 142 142 142 142 142 142 Panel B: OLS Regression .0002025 -.0637579 -.0974462 .0000327 -.0000102 .114323 .0201334 12.76118 (.0024202) (.3917926) (.1155226) (.0169465) (.0001717) (.0975078) (.1065241) (.5901007) 142 142 142 142 142 142 142 142 Panel C: IV Regression .0002117 .2312543 -.020249 .0109664 -.0001538 .1024231 .0208296 12.3533 (.0027015) (.7490474) (.2048184) (.0294243) (.0003527) (.1115709) (.1189138) (1.068066) 142 142 142 142 142 142 142 142 Standard errors in brackets, * significant at 10%, ** significant at 5%, *** significant at 1%. Source: Author's calculations. 187 5.5.3. Testing Linear Panel-Data Models The essential distinction in microeconometric analysis of panel data is that between FE and RE models. If effects are fixed, then the pooled OLS and RE estimators are inconsistent, and instead the within (or FE) estimator needs to be used. The within estimator is otherwise less desirable, because using only within variation leads to less efficient estimation and inability to estimate coefficients of time-invariant regressors (Cameron and Trivedi, 2009). Hausman test for fixed effects Under the null hypothesis that individual effects are random, these estimators should be similar because both are consistent. Under the alternative, these estimators diverge. This juxtaposition is a natural setting for a Hausman test, comparing FE and RE estimators (Cameron & Trivedi, 2009).184 In tables 5.20a-b we compare the estimable coefficients of time-varying regressors. Sigmamore specifies that both covariance matrices are based on the (same) estimated disturbance variance from the efficient estimator. We obtain that for all the coefficients of the regressors, that the test of RE against FE yields t-values that are higher in the catchment case than in the control areas for both the default version of the Hausmann as well as for sigmamore and sigmaless. Moreover, the t-values for Age; Age square; rain; rainlagged; landownership; and cotton income share indicate a highly significant difference for the catchment areas only, except for age square, which is significant at the 10% level in the control areas. Finally, the overall statistics for the catchment areas only has p<=0.001 for all three hausman tests, which leads to a strong rejection of the null hypothesis that the RE provides consistent estimates. 184 Or can be applied to a key subset of these (often one key regressor). 188 Table 5.20a: Hausman Test for Fixed Effects, Catchment areas highgrade Age Agesq rain rainlagged Landowners~c cotincshare cotincshar~q distiput distbank chi2(8) = (b-B)'[(V_b-V_B)^(-1)](b-B) Prob>chi2 Coefficients (b) (B) FE RE 0,050 0,045 0,094 -0,005 0,001 0,000 -0,001 -0,039 -0,021 0,029 -0,143 0,188 -1,759 0,340 2,472 0,027 -0,004 -0,001 -0,004 -0,004 (b-B) Difference 0,005 0,099 0,001 0,038 -0,049 -0,330 -2,099 2,445 -0,003 0,000 Hausman Hausman Sigmamore Sigmaless Hausmann FE RE sqrt(diag(V_b-V_B) sqrt(diag(V_b-V_B) sqrt(diag(V_b-V_B) S.E. S.E. S.E. 0,017 0,032 0,026 0,050** 0,063* 0,052** 0,0003*** 0,0005** 0,0004*** 0,013*** 0,0184** 0,015** 0,012*** 0,019** 0,0152*** 0,064*** 0,089*** 0,073*** 0,74*** 1,541 1,262* 0,7393*** 1,676* 1,373* . 0,004 0,003 0,003 0,005 0,004 27,210 26,820 39,990 0,001 0,002 0,000 Notes: b = consistent under Ho and Ha; obtained from xtreg. B = inconsistent under Ha, efficient under Ho; obtained from xtreg. Test: Ho: difference in coefficients not systematic. sigmamore and sigmaless specify that the two covariance matrices used in the test be based on a common estimate of disturbance variance (sigma2). sigmamore specifies that the covariance matrices be based on the estimated disturbance variance from the efficient estimator. This option provides a proper estimate of the contrast variance for so-called tests of exogeneity and overidentification in instrumental variables regression. sigmaless specifies that the covariance matrices be based on the estimated disturbance variance from the consistent estimator (Stata Manual).185 Sourceμ Author‘s estimations. Table 5.20b: Hausman Test for Fixed Effects, Control Areas highgrade Age Agesq rain rainlagged Landowners~c cotincshare cotincshar~q distiput distbank chi2(8) = (b-B)'[(V_b-V_B)^(-1)](b-B) Prob>chi2 Sourceμ Author‘s estimations. Coefficients (b) (B) FE RE 0,020 0,048 -0,150 -0,030 0,002 0,000 0,004 -0,002 0,035 0,055 -0,002 0,195 -1,441 -0,691 2,373 0,838 0,001 -0,004 -0,006 -0,005 Hausman Hausman Sigmamore Sigmaless Hausmann FE RE (b-B) sqrt(diag(V_b-V_B) sqrt(diag(V_b-V_B) sqrt(diag(V_b-V_B) Difference S.E. S.E. S.E. -0,028 0,052 0,043 0,049 -0,119 0,230 0,202 0,229 0,002 0,001* 0,001* 0,001* 0,006 0,023 0,020 0,023 -0,020 0,041 0,035 0,039 -0,198 0,231 0,193 0,219 -0,751 2,080 1,685 1,910 1,535 2,508 2,055 2,329 0,005 0,012 0,010 0,011 -0,001 0,015 0,013 0,015 5,070 7,310 5,690 0,828 0,605 0,770 185 Sigmamore or sigmaless are recommended when comparing fixed-effects and random-effects linear regression because they are much less likely to produce a nonpositive-definite differenced covariance matrix (although the tests are asymptotically equivalent whether or not one of the options is specified) (Stata on-line Manuel, 2009). 189 5.6. Conclusions In this chapter evidence was presented through the analysis of existing Zambian rural household survey (LCMS) data. I do find that there seem to a linkage between consumption growth and rural feeder road improvements in rural areas in Zambia‘s Eastern Province. However, both the results from cross-section and pseudo-panel data analysis have proven sensitive to the model specification. The testing of the different models used in this chapter from the simple plain vanilla OLS to the more sophisticated models doesn‘t exactly lead to the conclusion that one is a far superior model over the others.186 One important reason for these far from perfect results is certainly the measurement of expenditure and the other key regressors, especially with regards to the imperfect 2004 dataset the quality of which has been affected by the unavoidable choice to resort to using the collapsing technique. This approach seriously affected the variation, and hence explains the low R2s in 2004. This evidently, makes it difficult to base policy prescriptions on these somewhat unclear results, which are sensitive to the model specification . The results in this chapter confirms what I found in the previous chapter 4, namely that agriculture is the overwhelming dominant activity in the rural areas in Zambia‘s agriculture-based Eastern Province and therefore the main source of pro-poor economic growth. Despite the fact that farming as the principal activity among the sampled rural households had fallen from 63 % in 1998 to 55% in 2004 (table 5.3), which could be considered as a testimony to a slow but steady diversification trend of income sources towards the labour market and the rural nonfarm economy and migration out of rural areas, the cotton production still generated the largest share of these households‟ income in both 1998 and 2004. 186 Model 1: The plain vanilla OLS that does not account for clustering; [...]; Model 16: Mixed-effects REML regression. 190 This happened notwithstanding the fact that the world cotton market in the late 1990s was severely marked by the collapse in the world prices as illustrated by price indices of cotton products in figure A9 and likewise observed by (Hertel & Winters, 2006; Balat & Porto, 2005b; see chapter 4). Hence, it is somewhat surprising that cotton‘s share in income among the rural households in Eastern province rose by 183% for the total sample, and 157% and 294% for respectively the catchment and control districts over the same period. Moreover, on average, the sampled rural households farm just less than one hectare for food crops and 1.28 hectare for non-food crops hadn‘t change noteworthy between 1998 and 2004. On the other hand, the mean distance to services and community assets for the rural households had diminished significantly in the same period, mainly due to the improvement in rural transport infrastructure not only associated with the EPFRP catchment districts but probably also due to similar projects being implemented in the province (chapter 4).187 One weakness of my approach is that although I stratify the rural households into small-, medium and rural-nonfarm the district level approach doesn‘t allow me to properly capture the heterogeneity within the same scale of households e.g. subsistence small scale farmers versus commercial smallholder farming. The same applies to the eight provincial districts, e.g. according to their agricultural potential and access to markets. I try to address this shortcoming through the use of an infrastructure dummy variable in the case of the whole sample, with the catchment districts considered as proxies for agricultural potential and access to markets but with the drawback of not capturing the diverse local conditions within each district according to distinct agroecologies, which produce a wide range of farming systems and crops. 187 The EPFRP approach was replicated as the roads component of the Smallholder Enterprise and Marketing Project (SHEMP) and the implementation by the Zambian Social Investment Fund (ZAMSIF) of the Community Transport Infrastructure (CTI) a sub-project of the Road Sector Investment Programme. 191 Following Horowitz and Lee(2002), I also believe that it might be useful to develop additional semi-parametric models that achieve both good estimation precision and a high degree of flexibility. Although the usefulness of semiparametric models in econometrics applied on rural development and statistics is not fully understood, virtually any new application of these models could provide useful additional information for the ex-post policy evaluation of poor area public infrastructure projects.188 The covariate set in the pseudo-panel model is different from the covariate set in the semi-parametric and parametric models. The reason for the different choice of covariates comes from the fact that I in the two latter models used a more mechanical way of finding the best performing model specification by working my way through more than 16 different models from the simple plain vanilla OLS model to much more sophisticated models. On the other hand, the choice of the covariates in the specification of the pseudo panel model was partly inspired by my review of the neo-classical and endogenous macro economic growth literature (see chapter 2). Finally, the cohort data approach seem to have many advantages over both the parametric and the semi-parametric approach based on independent cross-sectional household surveys, which could be explored further in future research. 188 The GLM framework is the standard nonlinear model framework in many areas of applied statistics. Cameron&Trivedi(2009:321) mention that it is little used in econometrics. 192 Chapter 6: Poverty over Time in Zambia‟s Eastern Province 193 6.1. Introduction In the development-driven approach to trade literature an analytical framework has been constructed to identify at the national level the various channels through which price changes associated with the removal of border trade barriers (analogous to the creation of transport network connectivity) are “passed through” the economic system to influence the welfare of richer and poorer households (Winters, 2000b, McCulloch et al., 2001, UNCTAD, 2004, Winters, 2002a). Within this analytical framework, trade policy reform (or alternatively transport network improvements) is seen as a price shock, which has (i) expenditure effects, which arise because of changes in the prices of the goods that are consumed; and (ii) income and employment effects, which arise because of changes in the remuneration of factors of production (UNCTAD, 2004). The choice of the geographical unit to which data apply primarily depend on what data are available from Zambia‘s ωSη in terms of both time and spatial coverage. The risk is that the data may be at too broad a geographical scale for local effects to be identified using an econometric approach (Chapter 5 and Chapter 7). Fortunately, serious efforts have been made in Zambia to collect data on a wide range of topics. The institutionalized collection of poverty data for the monitoring of the social dimensions of adjustment programmes started in 1991 through the first Priority Survey (PS I) (Nsemukila, 2001).189 From table 6.1 it appears that Zambia possess a run of national household expenditure surveys (1991, 1993, 1996, 1998, 2002, 2004 and 2006).190 These surveys are conducted in an integrated manner and as the core of the National Household Survey Capability Programme (NHSCP), which CSO has implemented since 1983 (CSO, 2003). In fact, Zambia is one of the few SSA countries to have implemented a series of 189 Participatory poverty monitoring exercises, based on revisits to Participatory Poverty Assessment (PPA) sites, have been carried out in Zambia by an NGO formed by the team which undertook the participatory component of the initial assessment. Booth, et al., 1998. Participation and Combined Methods in African Poverty Assessment: Renewing the Agenda. Report commissioned by the UK Department for International Development for the Working Group on Poverty and Social Policy, Special Program of Assistance for Africa. For more information see MILIMO, J. T. 1995. An Analysis of Qualitative Information on Agriculture from beneficiary assessments, participatory poverty assessment and other studies which used qualitative research methods, Milimo, 1997. Participatory Poverty Monitoring III. Lusaka: A Report by the Participatory Assessment Group, Milimo, et al., 2002. The poor of Zambia speak. Who would ever listen to the poor? , Saasa, 1990. 190 Paris21 has no activity with Zambia for the moment, however Zambia has participated in the General Data Dissemination System (GDDS) since November 2001. Most of the data disseminated follow the GDDS recommendations for periodicity and timeliness (IMF, 2005). 194 national household surveys over the 1990s all of which are using the sampling frame from the 1990 Census of Population and Housing. Table 6.1: Relevant Zambian Household Surveys since 1991 Year Survey Data Collection Sampling Method 1991 Income/Expenditure/Household Survey (IES-HBS) I June/July n.a. Sample Size Visits 2.930 4 1991 SDA Priority Survey I Oct./Nov. Multi-stage Stratified Random Sample 9.886 1 1 1992 Demographic and Health Survey (DHS) I* Jan/May Multi-stage Stratified Random Sample 6.209 1993 SDA Priority Survey II March/June Multi-stage Stratified Random Sample 10.121 1 1993/1994 Income/Expenditure/Household Survey (IES-HBS) II July/Oct. Multi-stage Stratified Random Sample 4.500 1 1996 Living Conditions Monitoring Survey (LCMS) I Sept/Oct. Multi-stage Stratified Random Sample 11.752 1 1996/1997 Demographic and Health Survey (DHS) III* July/Jan. Multi-stage Stratified Random Sample 7.286 1 1 1998 Living Conditions Monitoring Survey (LCMS) II Nov./Dec. Multi-stage Stratified Random Sample 16710 2001 Demographic and Health Survey (DHS) IV* Nov/May n.a. 7.126 2001/2002 Living Conditions Monitoring Survey (LCMS) III Rolling Sample Multi-stage Stratified Random Sample 19.600 6 2002 Demographic and Health Survey (DHS) V* Aug/Oct. Multi-stage Stratified Random Sample 4.245 1 2004/2005 Living Conditions Monitoring Survey (LCMS) IV Oct./Jan. Multi-stage Stratified Random Sample 20.000 1 2005 Labour Force Survey II n.a. n.a. n.a. 2006 Living Conditions Monitoring Survey V (LCMS V) n.a. Multi-stage Stratified Random Sample n.a. 2007 Demographic and Health Survey (DHS) VI* April/Nov Multi-stage Stratified Random Sample 7164 1 Notes: Two weeks recall period for food, one month recall for other expenditures. Recall periods for education expenditures vary between surveys. * DHSs could be used for evaluating poverty interventions. Source: Authors‘ based on IHSζ ωentral Survey ωatalog (accessed on-line 2010). Hence, in chapter 6 we will follow an approach, which seeks to use existing secondary data derived from the LCMSs, on the composition of expenditure sources of households at different levels within the overall expenditure distribution to obtain a much more socially disaggregated view of the impact of rural transport infrastructure improvements, which took place in Zambia‘s Eastern θrovince from 1998 to β001 combined with the effects from the prior trade liberalisation from 1991 to 1995. With the aim of calculating the descriptive socio-economic statistics of the impacts of access to local transport infrastructure, the CSO LCMSs are useful sources, because they enable us to monitor the following rural development indicators over time: Health care and education; labour force; household income (sources) and assets; household expenditures; home consumption; agriculture; and access to infrastructure (e.g. markets) all derived from the independent cross-section 1998 LCMS II and the 2004 LCMS IV which are not compatible with the 1996 and 2003 LCMSs due to the variations in the data collection periods and the survey designs. Both the 1998 and 2004 LCMSs are cross-sectional or one-spot (single interview) surveys. This may make welfare measures imprecise both due to sampling and non-sampling errors. The rest of the chapter is organized as follows. Section 6.2 presents a brief description of the data sources and types of data used for the poverty and inequality analysis. Section 6.3 195 analyzes the evolution of the household expenditures between 1998 and 2004. Section 6.4 analyzes the evolution of poverty rates and inequality using various measures of poverty and inequality. Section 6.5 concludes. 6.2. Household Survey Data Before applying the analysis tools described below, we first assess the available data sources. Each data source in table 6.1 tends to have particular strengths (Coudouel et al., 2002b). However, our attention is exclusively devoted to the multi-topic LCMSs displayed in table 6.2. 6.2.1. Living Condition Monitoring Survey II 1998 The LCMS II survey was intended to highlight and monitor living conditions of the Zambian society. It included a set of repeated priority indicators on poverty and living conditions. The LCMS II had a normative point of departure aimed at illustrating living conditions that require appropriate policy decisions. The CSO carried out the second LCMS II in November-December, 1998. The sample had to be large enough to be broken down at provincial level, district level, by rural/urban, by centrality and by administrative level. These levels constitute the domains of study and data disaggregation for the survey analysis. Table 6.2: Sample Allocation (Standard Enumeration Areas - Primary Sampling Units) Surveys Province Total (% of total SEA) LCMS IV** Zambia 1048 (6.28% of 16,683) 2004 Eastern Rural Urban LCMS III** Zambia 520 (3.12% of 16,683) 326 194 2002/2003 Eastern 60 50 10 LCMS II* Zambia 820 (6.31% of 12,999) 492 328 1998 Eastern LCMS I* Zambia 610 (4.69% of 12,999) 348 262 1996 Eastern 68 54 14 Source: Author based upon metadata of the LCMSs. Notes: * The sampling frame developed from the 1990 census of population and housing. ** The sampling frame developed from the 2000 census of population and housing. Thus, the LCMS 2004 was carried out nation-wide on a sample basis in all of the 72 districts of Zambia, and covered 16,710 households representing a sampling fraction of about 1 household per every 113 households. The 2004 survey covered 8,487 households in rural areas and 8,223 households in urban areas. The 1998 LCMS covered nearly 13,000 196 households. The eligible household population consisted of all civilian households (table 6.2).191 Households had been selected in two stages. In the first stage, a sample of Standard Enumeration Area (SEA) was selected within each stratum (centrality) according to the number allocated to that stratum. Selection had been done systematically with probability proportional to the number of households within each SEA as registered in the 1990 Population Census.192 The second stage in each selected SEA, households were listed and each household was given a unique sampling serial number.193 A circular systematic sample of households was then selected. The circular systematic sampling method assumes that households are arranged in a circle (G. Kalton, 1983) and the following relationship applies: Let N = nk, Where, N = Total number of households listed in an SEA n = total sample size required from the SEA k = the sampling interval in a given SEA calculated as k=N/n. For the rural strata in LCMS, k = N/15, as 15 households were selected from each rural SEA. In the rural areas, 7 households have been selected from the stratum of small scale farmers, 5 from medium scale, 3 from non-agricultural and the large scale households were selected on a 100 percent basis, if any were found. Therefore, the number of selected households from a rural SEA was more than 15 where there were large scale farmers. In Micro-project areas the number of households to select was double, 14 in the small scale category, 10 in the medium, 6 in the non-agriculture, and all large scale farmers. The N in the rural SEAs differed from stratum to stratum within an SEA depending on how many households were identified as large scale in the listing. At this stage, a random-start number is obtained using a table of random numbers. This number is between 1 and N. The household whose random number lies between 1 and the 191 Excluded from the sample were, institutional population in hospitals, boarding schools, prisons, hotels, refugee camps and orphanages and diplomats accredited to Zambia in embassies and high commissions. Private households living around these institutions were not excluded. 192 Sample allocation was done using the PPS method. In 1998 this entailed allocating the total sample (820) proportionately to each province according to its population share. First-stage sampling units facilitate clustering of the sample to control costs and facilitate development of complete frames of housing unit in sampled areas. 193 Vacant residential housing units, non-contact households, refusals and partially responding households are not assigned sampling serial numbers. 197 random start is the first to be selected. Then k, the sampling interval is added to the sampling serial number of each selected household in the respective strata until the required n is achieved. Thus SEAs forms the primary sampling units (PSU), whereas the unit of analysis is the household (i.e. the secondary sampling unit (SSU)). In 2004 the LCMS SEA total sample allocation was 1084 whereas in 1998 the LCMS was based on a total sample of 820 SEAs of which 492 were rural SEAs (table 6.2). Unlike the integrated Household Budget (LCMS III), which have a carry-alone household budget module, employing a rolling sample meant to capture changes in living conditions due to seasonal effects,194 the LCMS I, LCMS II, and LCMS IV surveys were designed to provide reliable district estimates (table 6.1). 6.2.2. Living Conditions Monitoring Survey IV 2004 The Living Conditions Monitoring Survey IV was conducted between October 2004 and January 2005 covering the whole country on a sample basis. The major objective was to provide poverty estimates, and provide a platform for comparing with previous poverty estimates derived from cross-sectional survey data. Using similar survey design to that earlier conducted in 1998, the poverty estimates from the 2004 survey are comparable to the survey of 1998 (CSO, 2005), that is used as a baseline in our chapter to:    Monitor living conditions of households over time in terms of access to various facilities and infrastructure and basic needs; Monitor the impact of the EPFRP on the well being of the population in Eastern Province; Monitor poverty and its distribution in Eastern Province. In line with LCMS II 1998, the LCMS IV 2004 collected data on the living conditions of households and persons in the areas e.g. of education, health, economic activities and employment, income sources, income levels, food production, household consumption expenditure, access to various socio-economic facilities and infrastructure, etc. (CSO, 2005). The LCMS IV likewise employed a two-stage stratified cluster sample design whereby during the first stage, 1048 SEAs were selected with Probability Proportional to 194 The diary method was used to collect household consumption expenditure data. 31 days of total diary entries for all the households were collected. 198 Estimated Size (PPES). The size measure was taken from the frame developed from the 2000 census of population and housing. During the second stage, approximately 20,000 households were systematically selected from an enumeration area listing (CSO, 2005).195 6.3. Average Monthly Household Expenditure in 1998 and 2004 We use consumption rather than income as the welfare measure and monetary dimension of well-being (Coudouel et al., 2002b), and use the estimate of total monthly per adult equivalent (p.a.e.) household expenditure calculated by CSO and provided on CD-ROM issued by the African Household Survey Databank, World Bank Africa Region. Household expenditure plays a vital function in the economy in several ways. Firstly, it is most closely associated with household poverty, well-being and living standards. In general, households are assigned a particular poverty status (poor or not poor) on the basis of their expenditures on goods and services which include, among other things, basic human needs such as food, shelter, clothing, etc., while household well-being and living standards are usually judged by the amount of goods and services that the household is able to access in a given time period. Secondly, household consumption expenditure constitutes a sizeable proportion of private consumption expenditure which significantly affects aggregate demand, output, income and employment in an economy. Thirdly, household expenditure serves as a useful proxy for household income, which in many cases, households tend to under-report (CSO, 2005, Coudouel et al., 2002b). We start by looking at some standard summary statistics shown in table 6.3 below. The column one and seven in the table shows the weighted average over the households in the 1998 LCMS and 2004 LCMS respectively of total monthly expenditure in real terms. The units are Zambian Kwacha (ZMK) with 2195 ZMK in 1998 and 4848 ZMK in 2004 worth 1 US dollar (table 6.5), and the consumer price indexes are those of Lusaka used to convert the nominal 2004 LCMS measures into constant 1998 prices. 195 In order to have equal precision in the estimates in all the districts and at the same time take into account variation in the sizes of the district, the survey adopted the Square Root sample allocation method, (Lesli Kish, 1987). This approach offers a better compromise between equal and proportional allocation methods in terms of reliability of both combined and separate estimates. The allocation of the sample points (PSUs) to rural and urban strata was almost proportional. 199 From table 6.2 above, we see that the sampling fraction in the first stage is small enough (i.e., n/N<10%) to be ignored in the LCMS, which means that the variation between the PSU means or totals will automatically incorporate any finite population correction (fpc) that applies to the sub-sampling within the PSUs. Therefore, since the first stage sampling fraction is small, we can essentially ignore sub-sampling within the any later stages that is within clusters (i.e. the PSUs).196 In Eastern Province the household's average monthly expenditure increased 139% from 1998 to 2004. Analysis by residence (rural/urban) shows that although urban households in Eastern Province spent a much higher average monthly amount on food and non-food than their rural counterparts did in both 1998 and 2004 the expenditure growth was driven mainly by the increase in the rural areas (table 6.3). This indicates that expenditure and income inequalities were high between rural and urban areas both in 1998 and 2004 in line with the national-wide expenditure pattern (CSO, 2005, CSO, 1998a). Disregarding the large-scale agricultural households due to the small sample size, analysis by rural strata shows that although the medium-scale households incurred higher total expenditure in both 1998 and 2004 than the small-scale households the gap significantly narrowed. At the district level in rural Eastern Province, the average expenditure on non-food dominates over food in 1998 in all districts except in Lundazi (305) and in Petauke (308). In 1998 the households in the control district Mambwe(306) had the highest average per capita expenditure ZMK18,375, slightly followed by the households in the pre-treatment Lundazi district (ZMK17,395). In 2004 after the treatment, the treatment district Chipata had experienced the highest growth rate in total household expenditure per capita from 1998 to 2004, whereas two of the three control districts, namely Mambwe and Nyimba had experienced the lowest growth rate (column 15, table 6.3). The households in the treatment districts average outperformance of the households in the control districts in terms of per capita expenditure growth is likewise captured in tables 6.4a and 6.4b which measures the mean p.a.e. expenditure percentage change. However, the picture is different when focusing on mean-per capita income growth, where we find that the control districts are slightly catching up with the treatment districts (tables 6.4c-d). 196 For clustered designs it is usual to describe the first stage sampling unit as the PSU. When using STATA that compute complex standard errors from multi-stage clustered samples it is only necessary to have a PSU variable (SEA) in the dataset. Any clustering after the first stage does not have to be identified - the variance between PSUs automatically incorporates later stages of clustering. This method of calculating standard errors is sometimes known as the 'ultimate cluster method.'. 200 Table 6.3: Average Monthly Household Expenditure by Rural/Urban, Rural Stratum and District (ZMK), 1998 & 2004 1998 2004 Monthly Average Expenditure Household Monthly Average Expenditure Households Pct.ch. Residence No pct. Total Total Food Non Food Per Capita size No pct. Total Food Non Food Per Capita size 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 All Eastern Province 107981,4 51919,1 56062,3 19740,7 5,0 1739 100% 257560 190196,6 67363,4 47085,9 5,5 2108 (1995) 100% 139% Rural (1) 67513,3 32327,4 35185,9 12252,9 5,0 1304 75% 226126,7 181812,4 44314,3 41039,3 5,5 1583 (1479) 75% 235% Urban (2) 229292,8 110649,2 118643,6 42858,5 5,0 435 25% 347656,5 214227,9 133428,6 64982,5 5,4 525 25% 52% Rural Strata (scale of agricultural activities) Small Scale (1) 55101,1 28436,1 26665,0 9599,5 5,0 982 (975) 75% 224820,5 182077,1 42743,5 39167,3 5,7 1253 79% 308% Medium Scale (2) 134952,1 49372,1 85580,0 28652,3 7,9 140 11% 241288,4 189882,4 51406,0 51229,0 4,7 193 12% 79% Large Scale (3) 288870,2 93985,7 194884,5 55766,4 7,5 7 1% 218417,1 214254,9 4162,2 42165,5 5,2 6 0% -24% Fish Farming (4) n.a. n.a. n.a. n.a. n.a. n.a. n.a. 320429,4 105255,1 215174,4 65661,8 4,9 3 0% Non-Agriculture (5) 74358,1 38061,1 36297,0 14299,6 3,3 175 (169) 13% 214524,1 167435 47089,1 41254,6 5,2 128 8% 189% District (Rural) Chadiza (301) 57287,9 25980,7 31307,3 10932,8 5,1 119 (117) 9% 233233,7 193260,3 39973,4 44510,2 5,2 150 9% 307% Chama (302)* 49135,7 23008,9 26126,8 9150,0 4,8 90 7% 198546,8 163464,8 35082,0 36973,3 5,4 135 9% 304% Chipata (303) 61557,2 25689,2 35868,0 10042,0 5,5 288 (286) 22% 268068,3 216601,5 51466,8 43730,6 6,1 271 17% 335% Katete (304) 60676,5 27267,3 33409,2 11448,4 4,5 150 (148) 12% 176284,8 131137,1 45147,7 33261,3 5,3 211 13% 191% Lundazi (305) 97760,4 52338,2 45422,2 17395,1 4,8 169 (168) 13% 360493,9 325326,9 35167,0 64144,8 5,6 275 17% 269% Mambwe (306)* 89670,2 43689,5 45980,7 18375,0 4,7 133 (131) 10% 143115,2 103829 39286,2 29326,9 4,9 120 8% 60% Nyimba (307)* 68960,6 31544,2 37416,4 11951,6 4,8 120 (119) 9% 140002,3 90781,78 49220,5 24263,8 5,8 165 10% 103% Petauke (308) 56361,6 30053,8 26307,8 10859,7 5,4 235 (232) 18% 194230 142397 51832,9 37423,9 5,2 256 16% 245% Notes: The 1998 figures in brackets represent number where the 13 households without any total average expenditure the last month were censored away. In 2004 there were no observations in Eastern Province with zero total household expenditure. The 2004 nominal expenditure are deflated using a foodbasket04 = 216.3. A good rule of thumb is to use the finite population correction factor (fpc) to define both the standard error of the mean and the standard error of the proportion in order to measure how much extra precision we achieve when the sample size n become close to the population size N, when the sample is 510% or more of the population, so that n/N<0.05. However, since the LCMSs only sample a small fraction of the population at the national and provincial level of around 1%, which gives a fpc close to 1 with almost no effect, we have not used fpc to generalize to a larger population than the one we sampled from. * The Control Districts. Source: Author's calculations based upon LCMS II and LCMS IV. 201 Table 6.4a: Mean Per Adult Equivalent for different groups in Eastern Province Mean Pooled 2004 1998 Urban 54 291,7 63 995,6 34 761,2 Percentage change 45,7 Rural 31 717,2 56 862,9 11 086,0 80,5 Control 27 612,8 42 889,7 14 192,9 66,9 Treatment 36 303,2 61 115,1 13 032,1 78,7 Lowest quintile 2 336,7 8 564,9 1 253,1 85,4 2 7 352,3 19 976,2 3 909,9 80,4 3 16 082,9 34 619,3 7 000,9 79,8 4 32 839,1 59 245,8 13 411,0 77,4 116 289,2 169 712,0 40 667,1 76,0 34 901,3 58 259,1 13 224,4 77,3 Infrastructure Treatment Highest quintile Total Notes: (i) Percentage Changes shown between years 2004 and 1998. (ii) The nominal 2004 p.a.e. figure is deflated into a real 1998 p.a.e. measure by dividing the nominal 2004 values by a CPI based on the foodbasket equal to 216.3 and multiplying it with 100. Source: Authors' calculations using Adept version 4.1. Table 6.4b: Mean per adult equivalent for different groups in Eastern Province Mean Pooled 2004 1998 Urban 39 101,3 41 257,6 34 761,2 Percentage change 15,7 Rural 22 616,3 36 659,2 11 086,0 69,8 Control 20 486,4 27 650,8 14 192,9 48,7 Treatment 25 799,3 39 400,6 13 032,1 66,9 2 198,5 5 521,8 1 253,1 77,3 2 6 418,6 12 878,5 3 909,9 69,6 3 12 778,2 22 318,9 7 000,9 68,6 4 24 547,5 38 195,5 13 411,0 64,9 Highes t quintile 78 782,0 109 412,5 40 667,1 62,8 Total 24 942,0 37 559,3 13 224,4 64,8 Infras tructure Treatment Lowes t quintile Notes : (i) Percentage Changes s hown between years 2004 and 1998. (ii) The nominal 2004 p.a.e. figure is deflated into a real 1998 p.a.e. meas ure by dividing the nominal 2004 values by a CPI equal to 335.51 and multiplying it with 100. Source: Authors ' calculations us ing Adept vers ion 4.1. Table 6.4c: Mean per-capita incomes in real terms in Zambia‟s Eastern Province Mean Pooled 2004 1998 Urban 56 346,1 75 639,8 42 104,3 Percentage change 44,3 Rural 37 158,7 43 899,3 32 480,9 26,0 Control 30 860,4 39 162,1 26 891,2 31,3 Treatment 40 287,8 47 912,5 34 632,6 27,7 Lowes t quintile 3 353,2 4 814,8 2 547,9 47,1 2 8 387,7 9 989,2 7 458,5 25,3 3 13 823,4 16 514,9 12 360,6 25,2 Infras tructure Treatment 4 Highes t quintile Total 22 919,4 31 002,1 19 334,3 37,6 146 277,9 172 941,6 125 435,5 27,5 38 930,2 46 922,3 33 350,1 28,9 Notes : (i) Percentage Changes s hown between years 2004 and 1998. (ii) The nominal 2004 p.a.e. figure is deflated into a real 1998 p.a.e. meas ure by dividing the nominal 2004 values by a CPI bas ed on the foodbas ket equal to 216.3 and multiplying it with 100. Source: Authors ' calculations us ing Adept vers ion 4.1. Table 6.4d: Mean per-capita incomes in real terms in Zambia‟s Eastern Province Mean Pooled 2004 1998 Urban 42 104,3 48 764,6 42 104,3 Percentage change -13,7 Rural 32 480,9 28 301,6 32 480,9 14,8 Control 26 891,2 25 247,6 26 891,2 6,5 Treatment 34 632,6 30 888,9 34 632,6 12,1 Lowes t quintile 2 547,9 3 104,1 2 547,9 -17,9 2 7 458,5 6 440,0 7 458,5 15,8 3 12 360,6 10 647,1 12 360,6 16,1 Infras tructure Treatment 4 Highes t quintile Total 19 334,3 19 986,9 19 334,3 -3,3 125 435,5 111 494,6 125 435,5 12,5 33 350,1 30 250,6 33 350,1 10,2 Notes : (i) Percentage Changes s hown between years 2004 and 1998. (ii) The nominal 2004 p.a.e. figure is deflated into a real 1998 p.a.e. meas ure by dividing the nominal 2004 values by a CPI equal to 335.51 and multiplying it with 100. Source: Authors ' calculations us ing Adept vers ion 4.1. 202 6.4. Poverty Trends in Rural Eastern Province The LCMSs have adopted the material well-being perception of poverty in which the poor are defined as those members of society who are unable to afford minimum basic human needs, comprising of food and non-food items. Despite their limitations they will provide benchmark data for our poverty and inequality analysis in Zambia's Eastern Province. The principal steps involved in calculating a poverty measure are the following: i. First, one has to choose the relevant dimension and indicator of well-being (e.g. measure household consumption), ii. adjust for variation in cost of living, iii. one has to select a poverty line, that is, a threshold below which a given household or individual will be classified as poor (Ravallion, 2004), and iv. finally, one has to select a poverty measure to be used for reporting for the population as a whole or for a population subgroup only (Coudouel et al., 2002a, World Bank, 2005d). Each of the four steps is carried out below. 6.4.1. Poverty Lines in Zambia Deaton(1997) refers to writers, which have expressed grave doubts about the idea that there is some discontinuity in the distribution of welfare, with poverty on one side and lack of it on the other, and certainly there is no empirical indicator, e.g. calories, where there is any perceptible break in the distribution or in behaviour that would provide an empirical basis for the construction of a poverty line. Thus, without an empirical basis such as a discontinuity in some measure, the construction of poverty lines involves arbitrariness. Notwithstanding the fact that the calorie-based procedure of setting a poverty line is subject to a number of serious difficulties (Deaton, 1997), the poverty lines in Zambia have been based on the Food-Energy Intake (FEI) approaches. The methods attempt to establish a monetary value at which basic needs are met. The FEI method sets the minimum food requirement by finding the consumption expenditure level at which 203 food energy intake is just sufficient to meet pre-determined average food energy requirements for normal bodily functions (CSO, 2005, Karlsson and Berkeley, 2005). Thus, using the household survey data, the minimal standard of living is proxied by the level of consumption expenditure that will enable the household or individual to attain their basic needs. This usually means, being able to purchase a basket of goods containing the minimum quantity of calories and non-food commodities. The predetermined food energy requirement used in LCMS analysis is based on the minimum calorie intake of 2,094 calories per day per person. This amount is derived from a study carried out by the National Food and Nutrition Commission, which constructed a basic food basket necessary to maintain the nutritional requirements of an average Zambian family (The food basket comes from (ILO, 1981)). All households not able to achieve this critical level of consumption expenditure or income are described as poor (Nsemukila, 2001, CSO, 2005).197 Table 6.5: Zambian Poverty Lines Official Exchange Rate (USD/ZMK) Total Poverty Line (ZMK per month) Food Poverty Line (ZMK per month) Upper Poverty Line (ZMK per month) Lower (Food) Poverty Line (ZMK per month) Total Poverty Line (USD per month) Food Poverty Line (USD per month) Upper Poverty Line (USD per month) Lower (Food) Poverty Line (USD per month) Total Poverty Line per day (ZMK) Food Poverty Line per day (ZMK) Upper Poverty Line per day (ZMK) Lower (Food) Poverty Line per day (ZMK) Total Poverty Line per day (USD) Food Poverty Line per day (USD) Upper Poverty Line per day (USD) Lower (Food) Poverty Line per day (USD) Consumer Price Index Foodbasket Index 1996 1275,0 n.a. n.a. 28979,4 20181,0 n.a. n.a. 22,7 15,8 n.a. n.a. 966,0 672,7 n.a. n.a. 0,8 0,5 n.a. n.a. 1998 2195,1 33562,4 24164,9 47187,0 32861,0 15,3 11,0 21,5 15,0 1118,7 805,5 1572,9 1095,4 0,5 0,4 0,7 0,5 100,0 100,0 2003 4737,2 n.a. n.a. 92185,0 64530,0 n.a. n.a. 19,5 13,6 n.a. n.a. 3072,8 2151,0 n.a. n.a. 0,7 0,5 n.a. n.a. 2004 4848,9 72595,5 52268,7 111747,0 78223,0 15,0 10,8 23,1 16,1 2419,8 1742,3 3724,9 2607,4 0,5 0,4 0,8 0,5 335,5 216,3 Note: Used the CPI for the four years to convert the nominal survey measures to constant 1998 prices. In November 2004 Total = 17.5% and Non Metropolitan Group (covering households in rural areas) 17.8%. Sources: CSO(2005:113) and Author's calculation. When a household does not have enough income to cover the cost of the ‗minimum food basket‘ it is considered to be core poor or severely food insecure. If its income level is such that it can cover the cost of the minimum food basket with up to 70 percent of its income, it is considered to be moderately poor or moderately food insecure. Households that require less than 70 percent of their income to cover the cost of the minimum food basket are considered to be non-poor or food secure. 197 204 The analysis of poverty has revealed that there is a 'hardest-hit' category of people consisting of those who cannot afford to meet the basic minimum food requirements even if they allocated all their total spending on food. This group is frequently referred to as the extremely poor or the ultra poor in the literature of poverty. The Extreme (lower) Poverty Line is normally set at the total expenditure equivalent to the Food Poverty Line. For example in LCMS IV, these are households whose total monthly expenditures are less than K78,223 per adult equivalent (p.a.e.). This is converted from the 1998 poverty line of Kγβ,8θ1 by using ωSη‘s Consumer Price Index: ωθI98/ωθI04 = π98,04 (Table 6.5). A special character of the Zambian CPI is a combined income level and geographical classification. The overall CPI is calculated from three sub-aggregates: Metropolitan High Income, Metropolitan Low Income and Non-Metropolitan. All rural households, regardless of their income, were classified as non-metropolitan group (CSO, 2004b).198 In view of the fact that minimum basic needs do not entail food-energy intake alone, some minimum basic non-food items such as health, shelter, and education are also necessary. The moderately poor category consists of people who can afford to meet the basic minimum food requirements but cannot afford non food basic needs. The overall poverty line is derived from the summation of the food expenditure level that gives the required food energy intake and the mean non-food expenditure allowance. The non-poor category consists of people whose expenditure is equal or more than the overall poverty line (CSO, 2005). As seen from table 6.5 the established poverty lines remain fixed in real terms reflecting a view of poverty as absolute, in the sense that poverty is defined by the ability to purchase a given bundle of goods, a view that is subject to controversy. Deaton(1997) believes that there should be some movement of the line in response to changes in mean level of living (cf. relative concept of poverty). 198 The weight structure for the CPI market basket was derived solely from the 1993/1994 HBS. The results from the 2002/2003 LCMS survey were used to revise the weights for the CPI during 2004. 205 To take into account differentials in household size and age structure of households, total income/expenditure was measured on a per adult equivalent (p.a.e.) basis. This was rationalised on the basis that children are generally considered less ‗demanding‘ in the sense that an additional child requires less additional expenditure to maintain household welfare than would an additional adult. This necessitated assigning a weight of between 0 and 1 to members of the household depending on their age (table A1a, Annex Ch.6). The adult equivalent scale for each member was added to those of other members to give the household total adult equivalent. The total monthly household expenditure divided by the total household adult equivalent gave the monthly household expenditure p.a.e. (Kapungwe, 2004).199 In our analysis we use the total (moderate) and food (extreme) poverty lines calculated by CSO and set at respectively K33,562 (US$15.3) and K24,164 (US$11) (table 6.5). We are mindful that the values were further adjusted by McCulloch et al.,(2001) to account for the difference between the CSO and Latham equivalence scales (table A1b), which give the much higher upper and lower national poverty lines set at K47,187 (US$21.5) and K32,861 (US$15) p.a.e. per month in 1998 prices. However, this 40% difference is accounted for in our sensitivity analysis (tables A.14.a-b). In both cases these two poverty lines correspond to the official basic needs (upper) and food (lower) poverty lines. The lower poverty line satisfies nutritional requirements (corresponds to US$0.50 per day). The upper poverty line adds another 30 percent for basic non-food needs (corresponds to US$0.72 per day in 1998) (table 6.5). The international US$1 per day per capita (PPP) poverty line produces extremely high poverty rates (above 90 percent) (Thurlow and Wobst, 2004b) and is therefore not in our analysis. 199 Adult equivalent scales (weights) were based on calorie and protein requirements for different age groups. Each person was assigned a weight according to his or her age relative to the caloric and protein requirement of an adult person (tables A1a-b, annex Ch.6). 206 6.4.2. Poverty Measures and Trends The relationship between economic growth and poverty critically depends on the nature of poverty, the definition of the poverty line and the level of per capita income in Zambia. There is a large literature on poverty measures (Foster et al., 1984, Ravallion and Bidani, 1994a, Ravallion, 1992). In this section we shall focus on three main measures, all of which are members of the class of measures proposed by (Foster et al., 1984) and used in our analysis. They are: The head-count index H showing poverty incidence, the poverty-gap index PG, and the Foster-Greer-Thorbecke P2 measure (Ravallion, 1992). Ravallion(1992) prefers to interpret them as measures of three different things: The head-count index is a measure of the prevalence of poverty, the poverty-gap index is a measure of the depth of poverty, while the P2 measures the severity of poverty. The data from the series of two national household surveys (LCMSs) in Zambia outlined above are analysed to explore how poverty have changed since 1998 in Eastern Province, which is one of Zambia‘s provinces with the greatest density of poverty. To measure poverty, consumption per adult equivalent (p.a.e.) is used as the index of individual welfare. This index is preferred over other indices such as per capita expenditure measures because it ensures that the differing needs of household members are covered. It is convenient to make all members of the household homogeneous by means of some equivalence scale, and no difference has been attached between male and female adults each have a consumption weight of one. For children less than 12 years different consumption weights according to age-group have been given (table A.1a) (CSO, 2005). The table 6.3 above shows the mean p.a.e. consumption expenditure for 1998 and 2004 —all figures are in 1998 Kwacha. The simplest poverty measure is the head-count index of poverty (P0), given by the proportion of the population for whom consumption p.a.e. y is less than the poverty line z (table 6.5). Suppose q people are poor by this definition in a population of size n. Then the head-count index is: H = q/n = proportion of total population deemed to be poor (Ravallion, 1992). 207 From the datasets in 2004 and 1998, we can see that the incidence of poverty in Eastern Province declined sharply from 1998 to 2004. In 2004 the percentage of people whose p.a.e. consumption was below the upper poverty line ZMK111,747 (i.e. the "total" poverty line) had fallen quite significantly to 67 percent from 90 percent in 1998. The fall in incidence of poverty in Eastern Province was especially noteworthy in the rural areas, where the fall was even more pronounced from 96 percent to 70 percent (table 6.6; cf. tables A.2a-b in Annex Ch.6). An even more important finding is that the change in poverty headcount was almost twice as high in the treatment region compared to the control region (tables A.3a-b in Annex Ch.6). Table 6.6: Headcount Poverty Estimates in Rural Eastern Province, 2004 and 1998 2004 LCMS IV Upper Poverty Std.Err.* Lower Poverty National 0,669 0,010 0,523 Rural (1) 0,743 0,004 0,620 Urban (2) 0,567 0,012 0,389 Eastern Province 0,673 0,017 0,550 Rural (1) 0,699 0,015 0,586 Urban (2) 0,587 0,040 0,432 Type of Household in Rural Eastern Province Smallscale 0,699 0,016 0,586 Mediumscale 0,669 0,036 0,563 Largescale 0,781 0,140 0,567 Nonagricultural 0,759 0,032 0,643 Rural Eastern Province Chadiza (301) 0,731 0,074 0,592 Chama (302) 0,761 0,038 0,629 Chipata (303) 0,655 0,045 0,561 Katete (304) 0,716 0,038 0,623 lundazi (305) 0,543 0,039 0,373 Mambwe (306) 0,749 0,026 0,599 Nyimba (307) 0,886 0,072 0,837 Petauke (308) 0,864 0,022 0,777 1998 LCMS II Std.Err.* Upper Poverty Std.Err.* Lower Poverty Std.Err.* 0,008 0,831 0,013 0,727 0,015 0,004 0,946 0,007 0,904 0,006 0,013 0,719 0,010 0,555 0,011 0,015 0,899 0,017 0,834 0,017 0,010 0,957 0,005 0,919 0,014 0,039 0,731 0,054 0,595 0,062 0,008 0,034 0,150 0,044 0,968 0,942 0,630 0,919 0,006 0,040 0,276 0,028 0,944 0,878 0,630 0,794 0,015 0,061 0,264 0,034 0,069 0,048 0,043 0,037 0,027 0,066 0,071 0,025 0,966 0,974 0,956 0,948 0,940 0,933 0,942 0,988 0,030 0,017 0,004 0,010 0,012 0,024 0,070 0,011 0,931 0,948 0,937 0,911 0,862 0,861 0,898 0,968 0,037 0,026 0,002 0,027 0,033 0,042 0,064 0,014 Note: Standard errors calculated taking into account the survey's two-stage sampling design. Computed from the "svymean" command in Stata 9.2. Code appendix for the STATA code for bootstrap is available upon request. Source: Author's calculations based on LCMS, 1998 and 2004. Table 6.6 also shows bootstrapped standard errors for these poverty measures. These come from 100 bootstrap replications and take into account the clustered structure of the LCMSs. Additionally, "core" (i.e. extreme or severe) poverty rates were determined using a lower core poverty line of ZMK78,223 in 2004, which is defined as the food component of the total poverty line (see table 6.5). The fall in core poverty was even larger, namely 28 percent compared to a fall of 23 percent in total poverty in Eastern Province. Again if 208 we compare the core poverty figures across the urban-rural divide, the decline in core poverty was highest in the rural areas with a fall of 33 percent compared to only 16 percent in the urban areas (table 6.6).200 Similarly, the decline in core poverty was double as high in the treatment region compared to the control region (tables A.3a-b in Annex Ch.6). Furthermore, households within each enumeration area were broken down into household categories. For rural areas there were small-scale farmers, medium-scale farmers, large-scale farmers, and non-agricultural households.201 The household poverty rates by household category are likewise shown in table 6.6. If we focus on the type of households, which lives in the rural areas in Eastern Province we notice that the percentage of small-scale and medium scale household living in extreme poverty fall faster between 2004 and 1998 than those living below the upper poverty line, but also that the percentage of these two strata fell much more than amongst the large-scale farmers and the non-agricultural households (table 6.6; cf. table A.9 in Annex Ch.6). ψy looking at the poverty by household head‘s age we find that the biggest decline in poverty headcount happens in the age intervals from 50-54 years; 45-49 years; and 2024 years respectively, all significantly higher than the average fall of 43.1 per cent (table A.5a). The fall amongst the self-employed was almost as fast as the average fall (table A.6a). Moreover, head of households having attained grade 10 experienced the highest decline in poverty, followed by those with grade 5 and grade 4, whereas grade 1-3 experienced a much slower decline than the average decline in the poverty headcount rate (table A.7a). Household headed by females only slightly experienced a faster decline than that of male headed households (table A.8a in Annex Ch.6). Finally, across districts in Eastern Province there is quite some variation amongst the rural households as shown in table 6.6. Two of the five catchment districts, namely 200 This result does not reflect differences in the cost of living between urban and rural areas. Prices in Zambia vary widely over time and space. Hence, poverty measures between rural and urban areas should therefore be treated with caution. 201 Households were categorized this way in order to stratify the sampling within enumeration areas, to ensure adequate coverage of a diverse set of households. 209 Lundazi and Chipata, have experienced the fastest fall in both total and core poverty rates. Surprisingly one catchment area, Petauke district, experienced a lower or similar decline in both poverty rates than the three control districts: Chadiza, Mambwe, and Nyimba. Ravallion(1992) argues that for certain sorts of poverty comparisons, such as assessing overall progress in reducing poverty, it may be quite adequate (though preferably always calculated for at least two poverty lines). However, for some purposes, including analyses of the impacts on the poor of specific policies, the head-count index has a serious drawback, because the head-count index is totally insensitive to differences in the depth of poverty (Deaton, 1997). A better measure is the poverty gap (P1), based on the aggregate poverty deficit of the poor relative to the poverty line. This gives a good indication of the depth of poverty, in that it depends on the distances of the poor below the poverty line. To see how this measure is defined, let consumptions be arranged in ascending order, the poorest has y1, the next poorest y2, etc., with the least poor having yq, which is (by definition) no greater than the poverty line z. Then the poverty gap index can be defined as follows: PG  1 q  z  yi   n i 1  z  mean proportionate poverty gap across the whole population (zero gap for the non-poor) (Ravallion, 1992, Deaton, 1997).202 Table 6.7 outlines the Poverty Gap Ratios, which in Eastern Province turns out to be 42.7 percent in 2004 up 16% from the 1998, 40 percent in rural areas and 52.2 percent in urban areas up 19% and 5% respectively from 1998. These ratios show that the mean consumption expenditure of these poor rural and urban households are 40% and 52.2% lower than the corresponding poverty line compared to consumption expenditure in these 202 PG also has an interpretation as an indicator of the potential for eliminating poverty by targeting transfers to the poor. 210 areas in Zambia as a whole. Also, they show the fraction of the consumption expenditure that is needed to eradicate poverty. The poverty gap ratio also show that 47% of the adults classified as extremely poor, in Eastern Province, have food expenditures below the poverty line,203 which likewise is an increase of 16% from the lower poverty gap of 31.8% in 1998. Again the biggest increase took place in the rural areas, which in the same period saw a rise of 18% compared to 3% in the urban areas (table 6.7; cf.tables A.2a-b in Annex Ch.6). Table 6.7: Poverty Gaps in Rural Eastern Province, 2004 and 1998 2004 LCMS IV Upper Poverty Std.Err.* Lower Poverty Std.Err.* National 0,473 0,002 0,529 0,004 Rural 0,428 0,004 0,496 0,003 Urban 0,553 0,006 0,600 0,028 Eastern Province 0,427 0,006 0,474 0,003 Rural 0,403 0,004 0,450 0,003 Urban 0,522 0,033 0,581 0,030 Type of Household in Rural Eastern Province Smallscale 0,400 0,007 0,446 0,009 Mediumscale 0,438 0,013 0,510 0,021 Largescale 0,493 0,104 0,474 0,002 Nonagricultural 0,369 0,042 0,392 0,036 Rural Eastern Province Chadiza (301) 0,421 0,033 0,455 0,020 Chama (302) 0,380 0,015 0,406 0,028 Chipata (303) 0,385 0,032 0,437 0,045 Katete (304) 0,412 0,023 0,494 0,026 lundazi (305) 0,496 0,020 0,460 0,031 Mambwe (306) 0,389 0,044 0,417 0,028 Nyimba (307) 0,347 0,029 0,457 0,026 Petauke (308) 0,367 0,010 0,441 0,017 1998 LCMS II Upper Poverty Std.Err.* Lower Poverty Std.Err.* 0,332 0,008 0,375 0,005 0,215 0,003 0,268 0,003 0,483 0,005 0,545 0,007 0,265 0,015 0,318 0,011 0,211 0,012 0,266 0,010 0,469 0,024 0,548 0,018 0,186 0,278 0,217 0,318 0,011 0,063 0,039 0,036 0,244 0,340 0,312 0,350 0,008 0,042 0,069 0,026 0,172 0,184 0,180 0,177 0,301 0,244 0,241 0,200 0,011 0,009 0,014 0,017 0,016 0,016 0,037 0,017 0,212 0,241 0,240 0,218 0,366 0,279 0,308 0,270 0,011 0,014 0,014 0,005 0,008 0,035 0,046 0,024 Note: Bootstrapped standard errors calculated taking into account the survey's two-stage sampling design. Source: Author's calculations. The upper and lower poverty gap of the small-scale farmers in both cases increased approximately 20% to 40% and 45% respectively whereas the medium-scale farmers experienced a slightly lesser increase of around 16% to 44% for the upper poverty gap and 51% for the lower poverty gap. Large-scale farmers saw a much higher increase in the upper poverty gap, 28%, than for the lower poverty gap of only 16%. Nonagricultural household on the other hand hardly experienced any significant change, so that the poverty gap remained much lower than for the other household categories. 203 Consumption of home-produced commodities is an important aspect of household food expenditure (especially in rural areas). 211 Chadiza and Katete districts saw the highest change of the upper poverty gap, contrary to the two control districts Mambwe and Nyimba, which experienced the smallest change. The Chadiza and Katete districts also experienced the largest change in the lower poverty gap, whereas Lundazi district recorded the smallest change from 1998 to 2004 (table 6.7). However the consumption gap ratio defined by PG is not a good poverty measure. It is the average value of the square of depth of poverty for each individual. Poorest people contribute relatively more to the index. Also called the Foster Greer Thorbeke (FGT) poverty severity index (P2), it gives a weight to the poverty gap. It is the average value of the square of depth of poverty for each individual. This poverty measure is sensitive to distribution among the poor, in the sense that the poorest people contribute relatively more to the index (Ravallion, 1992, Deaton, 1997). 1 q  z  yi  P2     n i 1  z  2 The FGT Index, which shows the intensity of poverty, is found to be high for both poor and extremely poor households in Zambia and in Eastern Province in particular. Somewhat surprisingly, according to the data the intensity of poverty is more serious both among poor household living in urban areas, and among extremely poor households living in urban areas both in 2004 and in 1998. In 1998, non-agricultural households show the highest intensity of poverty in rural areas, but in 2004 the few largescale households show a higher FGT index. On the other hand, it is the medium-scale rural households which show the highest lower poverty FGT indices. Finally, catchment district Lundazi has the highest FGT index in both 1998 and 2004, with the control district Nyimba taking the lowest FGT in 2004 and Chipata district in the pre-treatment baseline year 1998. 212 Table 6.8: Squared Poverty Gaps (Severity of Poverty) in Eastern Province, 2004 & 1998 2004 LCMS IV Upper Poverty Std.Err.* Lower Poverty Std.Err.* National 0,287 0,002 0,343 0,003 Rural 0,243 0,002 0,309 0,004 Urban 0,366 0,007 0,418 0,029 Eastern Province 0,249 0,005 0,283 0,003 Rural 0,227 0,004 0,258 0,005 Urban 0,334 0,031 0,400 0,032 Type of Household in Rural Eastern Province Smallscale 0,224 0,007 0,255 0,011 Mediumscale 0,251 0,023 0,317 0,020 Largescale 0,312 0,162 0,225 0,002 Nonagricultural 0,204 0,038 0,193 0,040 Rural Eastern Province Chadiza (301) 0,242 0,038 0,255 0,026 Chama (302) 0,214 0,013 0,224 0,022 Chipata (303) 0,209 0,027 0,243 0,039 Katete (304) 0,224 0,022 0,302 0,027 lundazi (305) 0,336 0,014 0,276 0,035 Mambwe (306) 0,211 0,037 0,227 0,030 Nyimba (307) 0,159 0,026 0,263 0,021 Petauke (308) 0,187 0,011 0,243 0,016 1998 LCMS II Upper Poverty Std.Err.* Lower Poverty 0,179 0,007 0,216 0,088 0,008 0,123 0,297 0,006 0,363 0,125 0,010 0,167 0,084 0,010 0,122 0,280 0,022 0,363 Std.Err.* 0,008 0,003 0,009 0,008 0,008 0,020 0,067 0,127 0,052 0,163 0,008 0,044 0,021 0,024 0,107 0,174 0,107 0,186 0,006 0,066 0,040 0,023 0,062 0,070 0,061 0,067 0,144 0,111 0,101 0,071 0,008 0,010 0,008 0,013 0,011 0,007 0,031 0,012 0,081 0,115 0,100 0,089 0,199 0,128 0,158 0,122 0,010 0,019 0,011 0,001 0,014 0,022 0,043 0,023 Note: Bootstrapped standard errors calculated taking into account the survey's two-stage sampling design Source: Author's calculations. 213 6.4.3. Rural Poverty Analysis by Deciles and Quintiles The fact that the squared poverty gap for most districts wasn't reduced (table 6.8) suggests that the changes experienced haven't been relatively pro-poor.204 To explore this further the growth in total household expenditure for each decile of the expenditure distribution are shown in table 6.9 and table 6.10. Table 6.9: Cumulative Rural total household expenditure by decile, 2004 Deciles decile 1 decile 2 decile 3 decile 4 decile 5 decile 6 decile 7 decile 8 decile 9 decile 10 Districts 301 1,07% 3,36% 6,75% 11,14% 16,63% 23,67% 32,40% 44,50% 64,98% 100,00% 302 1,06% 3,47% 6,97% 11,53% 17,33% 24,51% 33,66% 47,00% 68,49% 100,00% 303 1,05% 3,26% 6,34% 10,53% 15,73% 22,41% 31,05% 42,94% 61,76% 100,00% 304 0,79% 2,89% 5,85% 9,85% 14,88% 21,72% 29,95% 41,66% 59,22% 100,00% 305 1,16% 3,19% 6,40% 10,52% 15,66% 22,37% 30,64% 42,76% 61,30% 100,00% 306 1,04% 3,36% 6,56% 10,59% 15,78% 22,44% 30,72% 42,98% 62,28% 100,00% 307 1,21% 3,75% 7,14% 11,87% 17,82% 24,82% 34,67% 46,42% 63,96% 100,00% 308 0,89% 2,95% 6,10% 10,25% 15,57% 22,29% 30,80% 41,85% 57,55% 100,00% Source: Author's calculations. Table 6.10: Cumulative Rural total household expenditure by decile, 1998 Deciles decile 1 decile 2 decile 3 decile 4 decile 5 decile 6 decile 7 decile 8 decile 9 decile 10 Districts 301 0,30% 1,19% 2,93% 5,54% 9,66% 302 0,36% 1,50% 3,56% 6,81% 11,87% 303 0,24% 1,25% 3,10% 5,97% 10,23% 304 0,37% 1,34% 3,08% 5,82% 9,75% 305 0,28% 1,20% 2,95% 5,58% 9,42% 306 0,29% 1,38% 3,10% 6,03% 10,23% 307 0,45% 1,63% 3,76% 6,88% 11,45% 308 0,28% 1,34% 3,00% 5,89% 9,92% 15,13% 22,82% 33,47% 51,36% 100,00% 18,42% 28,15% 43,66% 63,85% 100,00% 16,05% 24,10% 36,04% 57,64% 100,00% 15,34% 23,00% 34,65% 52,08% 100,00% 14,78% 22,34% 34,88% 58,06% 100,00% 16,25% 24,59% 37,76% 57,12% 100,00% 17,78% 26,87% 40,26% 64,76% 100,00% 15,32% 23,80% 36,69% 56,07% 100,00% Source: Author's calculations. In descriptive statistics, a decile (i.e. a 10-quantile) is any of the 9 values that divide the sorted data into 10 equal parts, so that each part represents 1/10th of the district sample. From the figures in tables 6.9-10 it is seen that the average total household expenditures for households in the top decile range in 2004 ranged from around 31% to 41% of the total household expenditures in the eight Eastern Province districts, whereas in 1998 the share varied from 35% in Nyimba to 48% in Chadiza. The top deciles of household expenditures contain a large fraction of households who either overstated or understated their consumption thereby lending the findings to a good deal of 204 Using different foodbasket / CPI deflators and lower poverty lines we find opposite results for the rural and the overall population of households in Eastern Province (table A.2a), whereas the squared poverty gap also increases for the urban households in table A.2b., Annex Ch.6. 214 measurement error.205 Nevertheless, these figures contrast with the share of the lowestincome households in the first decile, which in 1998 ranged from only 0.24% in Chipata district to 0.45% in Nyimba district. This very unequal expenditure distribution had not improved much in 2004 where the share of the lowest decile only had increased to 0.79% in Katete as the absolute minimum and 1.21% in Nyimba district with the highest share amongst the eight districts. Table 6.11: Cumulative Rural total household expenditure by Quintile, 2004 Quintiles Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5 Districts 301 3,88% 12,38% 27,52% 51,67% 100,00% 302 3,71% 11,97% 25,29% 49,39% 100,00% 303 3,38% 11,92% 25,40% 50,73% 100,00% 304 3,69% 12,49% 26,29% 49,29% 100,00% 305 3,25% 11,33% 23,38% 43,77% 100,00% 306 3,62% 12,23% 26,88% 51,59% 100,00% 307 3,96% 11,71% 24,72% 46,72% 100,00% 308 2,15% 8,21% 17,76% 34,59% 100,00% Source: Author's calculations. Table 6.12: Cumulative Rural total household expenditure by Quintile, 1998 Quintiles Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5 Districts 301 1,51% 6,49% 17,45% 38,55% 100,00% 302 1,34% 6,09% 17,25% 39,82% 100,00% 303 1,37% 6,63% 17,75% 41,30% 100,00% 304 1,14% 5,03% 13,29% 29,50% 100,00% 305 1,44% 6,29% 17,11% 36,40% 100,00% 306 1,74% 6,81% 19,00% 40,16% 100,00% 307 2,05% 7,81% 20,26% 44,10% 100,00% 308 1,52% 6,49% 16,75% 38,16% 100,00% Source: Author's calculations. Due to the small sample sizes in each decile of the expenditure distribution, it is believed that using quintiles at the district level as consumption groups to examine how characteristics vary by consumption within the rural zones in Eastern Province might present a better description of the change in the expenditure distribution between 1998 and 2004. By definition each quintile covers 20 percent of cumulative total household expenditure distribution of the district population (World Bank, 2005d). From our calculations we find that in 1998 for each of the eight districts the two poorest consumption quintiles were lower than the Lower (Food) Poverty Line of ZMK32,861 per month (see table 6.1). In 2004 the mean total household expenditure of the lowest quintile were above the Lower (Food) Poverty Line of ZMK78,223 per month for all districts in Eastern Province, despite the fact the share only ranged from 2 to 4 percent of the total mean (tables 6.11-12). 205 The sample sizes ranged from only 1 to 4, and could to a large extent be considered as outliers. 215 From the ADEPT analysis we find that there is a shift in the distribution of the poor from rural to urban areas as well as from the treatment region to the control region (table A.3a). The distribution of the poor is shifted in favour of the females when using the Upper (total) poverty line of ZMK33,562, whereas the inverse is the case when using the Lower (food) poverty line of ZMK24,164 (table A.8a). When we measure poverty by land ownership we find that the distribution of the poor almost exclusively shifts in favour of the small scale farmers independent of the choice of poverty line (table A.9, Annex Ch.6). Figure A.4 in Annex Chapter 6 plots the Lorenz curve, which sorts the population from poorest to richest, and shows the cumulative proportion of the population on the horizontal axis and the cumulative proportion of p.a.e. expenditure on the vertical axis. From this curve we derive the popular measure of inequality, the Gini coefficient, which ranges from 0 (perfect equality) to 1 (perfect inequality). We find that the Gini coefficient has fallen in rural Eastern Province from 56 in 1998 to 53 in 2004. While the Gini coefficient has many desirable properties it cannot easily be decomposed to show the sources of inequality. Tables A.11a-b in Annex Chapter 6 present the results of Generalized Entropy Inequality decomposition for three GE(α) indexes: GE(0) Theil‘s δ index sometimes referred to as the mean log deviation measure; GE(1) Theil‘s T index, and GE(2). The differences between the treatment and control region represent less than 1% of the global inequality and the contribution within these subpopulations decreased from 69.4% of the overall inequality in 1998 to 53.3% in 2004. The most interesting result is that with parameter α = 0, that is with zero weight given to distances between expenditure at different parts of the distribution, we find that the inequality, which was higher in the pre-treatment region in 1998 compared to the control region is lower in the posttreatment region in β004. The same result applies for α = 1 or β. 216 Table A.13a in Annex Chapter 6 shows the general class of inequality measures: The ratio between 25th and 10th, 50th and 25th, 75th and 50th, 90th and 50th percentiles, quartile ratio (the ratio between 75th and 25th percentiles), tail ratio (the ratio between 90th and 10th percentiles). We find that the biggest decline in inequality in the rural areas between 1998 and 2004 is found in the decile dispersion ratio (i.e. between the richest 90th and poorest 10th percentiles). Finally, table A.16a in Annex Chapter 6 shows how change in inequality holding mean expenditure constant affects poverty. The table is computed based on a simulation of a lump-sum intervention and a subsequent tax, which change the distribution decreasing inequality (as measured by the Gini index) by 10% and the resulting change in poverty. We find that the elasticities of all the FGT poverty measures around actual distribution consistently are higher in 2004 than in 1998 and that the change is more pronounced in the rural areas compared to the urban areas (figures A.2-A.3 in Annex Ch.6). 217 6.5. Conclusion Interesting insights emerge from our empirical analysis of the poverty impact of the changes to the rural transport infrastructure network in Zambia‘s Eastern θrovince. ηne of the chapter‘s important results is that the pro-poor growth rate of rural Eastern province was positive. Moreover, contrary to the frequently expressed concerns about widening disparities across space (e.g., involving urban vs. rural areas) our results actually show an opposite trend in favour of a much faster reduction of the headcount poverty rate from 1998 to 2004 in the rural areas, although bearing in mind that Eastern Province is predominantly a rural largely dependent on agricultural activities. The fact that the squared poverty gap for most districts wasn't reduced suggests that the changes experienced haven't been relatively pro-poor. The share of the lowest-income households in the first decile in 1998 ranged from only 0.24% in Chipata district to 0.45% in Nyimba district. This very unequal expenditure distribution had not improved much in 2004 where the share of the lowest decile only had increased to 0.79% in Katete as the absolute minimum and 1.21% in Nyimba district with the highest share amongst the eight districts. However, using the ADEPT tool we actually find that while the inequality was higher in the pre-treatment region in 1998 compared to the control region this is no longer the case in 2004 after the EPFRP treatment. Additionally the biggest decline in inequality in the rural areas between 1998 and 2004 is found in the decile dispersion ratio. Thus, to the extent that remote households are poorer to begin with, a policy of building rural roads according to the same design as the EPFRP can have desirable distributional properties (see Jacoby, 2002, Jacoby and Minten, 2009). 218 Chapter 7: Rural Growth and Poverty Reduction through Feeder Road Rehabilitation in Chipata & Lundazi Districts of Zambia‟s Eastern Province, 1996-2005 219 Household surveys are essential for the analysis of most policy issues… - Nicholas Stern, senior vice president, the World Bank. 7.1. Introduction In this and the next chapter we turn our attention away from the regional level analysis used in the preceeding chapters 4, 5 and 6 towards two successive sub-regional analyses based on our own primary data. Both chapters focus explicitly on the situation in two districts of Zambia‘s Eastern θrovince, namely ωhipata and δundazi district. Under the old regime, remote farmers in Zambia were subsidised by those close to the line of rail (through pan-territorial pricing) and small farmers by larger ones with storage facilities (through pan-seasonal pricing). By 1996 all of these subsidies had been removed (chapter 8). The deterioration in the situation of remote farmers was substantially worse than would have arisen solely from the removal of pan-territorial pricing. For them, functioning markets had largely disappeared. In fact, the status quo in 1996 was that often there was no buyer at all or, if there was, the terms of trade were so poor that transactions occurred on a barter basis (Winters, 2000a). According to Winters(2000a) it is difficult to disentangle the relative importance of institutional and infrastructural factors in this market failure. There had been such a sharp deterioration in transport infrastructure that it was difficult for traders to reach areas that were more than a relatively short distance from a major route (Chapter 3). The question whether trading would be more active if infrastructure was better, or whether there are also institutional impediments is discussed in chapter 8. But in other areas, there are clear institutional constraints (see e.g. Dorward et al., 1998; Kydd and Dorward, 2001, 2004; Kydd et al., 2002) on top of the logistical ones (Winters, 2000a). The aim of this chapter is to take a closer look at the links from 16 rural community clusters to the nearest real market towns namely the district centres of Chipata and Lundazi districts. More specifically, chapter 7 exclusively explores whether rural consumption levels and consumption growth of rural households in catchment areas are better than for those living in the control areas. If so, whether these outcomes are 220 determined by better accessibility to the ―external‖ market and to services through the impacts of the EPFRP. First, this is done by partly drawing on data from a pseudo-panel consisting of 241 rural households interviewed in 1996 and another 88 rural households interviewed in 2005 in the same 16 rural (SEA (i.e. PSUs)). Measuring the welfare changes between 1996 and 2005 at that sub-district level is not without its problems (Dercon and Krishnan, 1998). Thus, I also explore whether the results obtained are robust to alternative solutions to some of the methodological problems. In line with most studies, I use consumption as the basis for measuring the standard of living. Furthermore, I use a cost-of-basic needs poverty line to calculate poverty measures (Ravallion and Bidani, 1994b). Secondly, I also rely on a comparison of the findings of a Zambian study carried out in 1996 and my own 2005 community follow-up survey, which is more qualitative and descriptive in nature. In other words, this chapter proceeds along a multiple research strategy and method triangulation to investigate the same research question above (World Bank, 2005b, Chung, 2000). The chapter is structured as follows. The next section describes the socio-economic and agro-ecological context of the 16 (20) different PSUs in Chipata and Lundazi districts. The third section presents the conceptual and methodological framework respectively. The following two sections successively present the qualitative and the quantitative findings, the robustness of which are tested in section six. Finally, section seven concludes and put forward some suggestions for public policy. 221 7.2. Context: Chipata and Lundazi Districts In 1996 Eastern province‘s 69,106 km2 was divided into 6 administrative districts including Chipata district and Lundazi district. In 2005 the province had been redivided into 8 districts.206 All the PSUs in Chipata and Lundazi districts fall within the 800 to 1000 mm mean annual rainfall range and are found in the Agro-ecological Zone II, which is considered the most suitable areas for crop production (Jha and Hojjati, 1993, Hazell and Hojjati, 1999). Chipata district is divided into eight agricultural blocks, and it‘s political system is made up of 4 national constituencies, 25 local wards and 22 chiefdoms.207 Chipata is bordering Malawi and being traversed by the Great Eastern Road connecting the capitals Lilongwe (130 km) and Lusaka (550 km).208 Except around the main town Chipata, Eastern θrovince‘s administrative capital, the area is sparsely populated (Zambia Vulnerability Assessment Committee, 2004; CSO, 2001, 2011). Table 7.1: Population Size and Average Growth Rates 1996 Population* 2000 Population Districts Male Female Total Male Female Total Chadiza 37628 38082 75711 40918 41482 82400 Chama 32780 33915 66695 37168 38517 75685 Chipata 157110 160608 317719 179450 182682 362132 Katete 81399 84575 165975 89693 92805 182498 Lundazi 104662 107221 211883 116720 120012 236732 Nyimba 31364 32777 64142 32941 34108 67049 Petauke 109853 114693 224556 119593 122940 242533 Mambwe 22801 23180 45983 25950 25994 51944 Eastern Province 577834 595366 1173206 642433 658540 1300973 2005 Population* Male Female Total 45437 46162 91599 43488 45157 88645 211893 214589 426478 101259 104228 205484 133765 138167 271930 35024 35848 70869 132992 134087 267041 30504 29996 60493 733430 Sourceμ Authors‘ calculations based upon CSO, 2001. Notes: * Extrapolations based upon 1990 Census population. 747016 1480423 Average Annual Growth Rate (pct p.a.) Male Female Total 2,12 2,16 2,14 3,19 3,23 3,21 3,38 3,27 3,33 2,46 2,35 2,40 2,76 2,86 2,81 1,23 1,00 1,11 2,15 1,75 1,94 3,29 2,91 3,09 2,68 2,55 2,62 In 2005 Chipata district had a population of 426,478. The total was divided between 214,589 females, which outnumbered the 211,893 males in the district. In addition, during the period from 1996 to 2005 the total population had increased Up until the early 1940‘s ωhief Sandwe‘s area, known as δusangazi, was administered from Chipata which comprised εambwe‘s area before the creation of Mambwe district. 207 Zambia's first multi-party elections since the 1960s were held on October 31, 1991. MMD candidate Frederick Chiluba resoundingly carried the presidential election over Kenneth Kaunda with 81% of the vote. However, UNIP swept the Eastern Province, gathering 19 of its seats there. 208 ωhipata is surrounded on almost all it‘s sides by hills, hence the name ωhipata, meaning ―gateway‖. 206 222 significantly from 317,719 in 1996, that is more than 108,000 inhabitants due to an annual growth rate of 3.33 per cent (see table 7.1). Lundazi district occupies an area of 14,068 Km2 and is located in the central portion of Eastern Province. Lundazi is divided into four agricultural blocks and it‘s political system is made up of 3 national constituencies, 24 local wards and 11 chiefdoms.209 Lundazi town area is about 650 km east of Lusaka,210 and about 160 km north of Chipata town. In the western part, the area extends to Luangwa River, while the eastern part extends to the Zambian border with Malawi. In 2005 the District was the second largest in Eastern θrovince with it‘s total population of β71,9γ0, of which 133,765 were male and 133,765 were female, equivalent to a population increase of more than 60,000 in the period from 1996 to 2005 (table 7.1). 7.2.1. Field Sites of the EPFRP A fundamental concept in transport supply analysis is the network structure of all transport systems (Banister and Berechman, 2000). Table 7.2: Total Length of Primary Roads Rehabilitated in Chipata District Rd Length (km) Road No. RD118 RD121 RD400 RD401 RD595 RD596 U33 Road Name M12-Tamanda Mission D104 - Chipalamba - D104 D124 - Chiguya T4 - Madzimawe - D124 T4 - Nzamane - Kazimuli RD595 - Sayiri - D128 Link RD 402 - Madzimoyo Total Primary Road Length EPFRP Share of Sub-Total Length of Feeder Roads in Chipata EPFRP Share of Total Length of Feeder Roads in Chipata 6,7 14,7 12,2 15,6 19,3 25,5 8,3 102,3 19,3% 12,7% Map Grid Reference Map Ref. Category (1:250,000) P Chipata P Chipata P Chipata P Chipata P Chipata P Chipata P Chipata P Chipata Out of 529,4 km Out of 804,7 km START VA 83 18 VA 40 7 VV 16 70 VV 42 80 VV 40 79 VV 39 71 VV 56 86 END VA 88 14 VA 48 4 VV 8 73 VV 28 85 VV 36 64 VV 58 64 VV 47 86 Sourceμ Authors‘ based on (Rwampororo et al., 2002). Notes: P = Primary Feeder Roads. The feeder roads in Zambia have been divided into 3 categories: Primary; Secondary; and Tertiary. In relation to the feeder road network in the Chipata district the improvement accounted for 19.3% of the total primary and 12.7% of the total length of 209 Each ward is represented by an elected councilor and the 11 chiefdoms by two representative councilors in the Lundazi District Council. 210 In other words, it was a former colonial headquarters. Today, it is the District Capital -- or "county seat" -- for Lundazi District. 223 feeder road network in Chipata district (table 7.2). In Lundazi district the EPFRP accounted for 28.9% of the 727.6 km of the total length of feeder roads (table 7.3). However, given that the selected EPFRP roads were prioritised and the most important links were identified to be rehabilitated and maintained, the impact within and across the two districts should a priori be greater than the de facto proportion of the network addressed (see chapter 3). 224 Table 7.3: Total Length of Feeder Roads Rehabilitated in Lundazi District Road No. Road Name RD107 D103 - Emusa - Chasefu - Chama Boundary RD110 Lundazi (M12) - Mwase RD110N D109 - Kapachila - Mwase (RD110) R243 Mphamba (D104) - Nyalubanga (D103) R246 Phikhamalaza (R245) - R248 R250 RD110 N - Kanyunya School R251 Mwase (RD 110 ) - Pono R254 RD110 - Gwaba - Kamtande R255 Mwase (RD110) - R254 U16 Gwaba (R254 ) - TBZ - Lumezi (M12) U18 Kapachila - RD110 Total Primary Road Length Total Secondary Road Length Total Tertiary Road Length Total Feeder Road Length EPFRP Share of Total Length of Feeder Roads in Lundazi Rd Length (km) 25,5 27,6 23,9 35,4 10 8,2 17 17,2 9,5 25,1 10,9 154,7 27,9 27,7 210,3 28,9% Map Grid Reference Map Ref. Category (1:250,000) P Nkota Kota Mzuzu P Nkota Kota P Nkota Kota P Nkota Kota T Nkota Kota T Nkota Kota S Nkota Kota P Nkota Kota T Nkota Kota P Nkota Kota S Nkota Kota P Nkota Kota S Nkota Kota T Nkota Kota Nkota Kota Out of 727,6 km START WB 16 78 WB 19 38 WB 22 43 WB 15 39 WB 25 52 WB 37 31 WB 40 28 WB 28 27 WB 37 27 WB 26 18 WB 30 32 WB WB WB END WB 0 WB 38 WB 38 WB 16 WB 34 WB 39 WB 58 WB 23 WB 28 WB 5 WB 20 WB WB WB 83 29 29 62 53 37 31 11 25 14 36 Sourceμ Authors‘ based on Rwampororo et al. 2002. Notes: P = Primary Feeder Roads; S = Secondary Feeder Roads; T = Tertiary Feeder Roads. 7.2.2. Description of the Survey Sites and of the Data Collection Procedures Our data come from two households surveys, the first Living Condition Monitoring Survey (LCMS-I) collected by Zambia‘s ωSη in 199θ and our own follow-up Eastern Province Rural Household Survey (EPRHS) implemented in 2005. They only covered 16 rural communities within the two districts where more than a 1,000 villages are situated.211 Basic characteristics of the 16 PSUs was also collected through our own 2005 community survey. Living Conditions Monitoring Survey I 1996 The nationwide data collection for the 1996 home-grown LCMS-I was carried out from September to October as part of the then Social Recovery Project (SRP) funded by the World Bank (CSO, 1996e, 1996b, 1996c; Moyo and Moyo, 1996).212 It covered both rural and urban areas in 57 of Zambia‘s 7β districts, because the other 15 had not been gazetted at that stage (CSO, 1996a, 1996b; McCulloch et al., 2001).213 Four basic instruments were used in collecting data during the LCMS-I survey. These were the 211 Excluding 3 SEAs situated in another agro-ecological zone (Luangwa Valley) covered by the LCMS-I; 3 pre-test sites and an additional area located along one of the EPFRP sites covered by the EPRHS. 212 Other sources mentioned August to end of September 1996, which matches the EPRHS survey period. 213 A Household Budget Survey (HBS) was carried out in Zambia at the same time as the LCMS was in its planning stage. The HBS was on a rather small sample and designed basically to satisfy the needs for providing updates of the weights for the CPI, and not to provide detailed information for monitoring poverty. The need was therefore felt to include data in consumption in the LCMS, although not as detailed as in a full-fledged HBS. 225 listing form214 and 3 sets of questionnaires. Amongst these, the household questionnaire administered to the head of each household and the individual questionnaire, which was administered to all persons in the sample 12 years and above, were both relevant for our monitoring purpose.215. The district was chosen as the domain of study to allow for the effective monitoring of living conditions as this would allow for coverage of small areas as well (CSO, 1996c). The sample had to be large enough to provide reliable district estimates. The sampling frame for the LCMS-I (1996) was drawn from the 1990 Census of Population and Housing and employed a similar multi-stage stratified random sample selection process to draw households. The CSO delineated the districts into CSAs, which were selected in the first stage. Then in the second stage SEAs were identified within the CSAs. Only one SEA was selected from each CSA. In Chipata a sample of 15 SEAs was selected with 10 SEAs selected from the rural stratum. Likewise, in Lundazi a sample of 11 SEA was selected with 9 SEAs selected from the rural stratum. Finally, the households were selected at a third stage. Table 7.4: Criteria for Stratification of rural Households Agricultural Activity Area under Crop Livestock Poultry Small Scale Less than 5 ha Less than 5 exotic dairy cows No beef cattle No exotic pigs No boilers No Layers Stratum Medium Scale Large Scale Non-Agricultural 5 to 20 ha, inclusive Over 20 ha None 5 to 20 exotic dairy cows, inclusive Over 20 exotic dairy cows None Up to 50 beef cattle Over 50 beef cattle None Up to 10 exotic pigs Over 10 exotic pigs None Up to 6000 broilers Over 6000 broilers None Up to 1000 layers Over 1000 layers None Parent stock of poultry Source: CSO, 1996b. Stratification was done using urban/rural and centrality as stratifying variables.216 Furthermore, within the selected rural SEAs, stratification was done on the basis of the rural household‘s scale of agricultural activity (small scale, medium scale, large-scale, 214 The listing took three (3) days on average. The selection of households was done using the circular systematic random sampling method. The sampling at household level was done by the supervisors. 215 The child questionnaire is disregarded. 216 The relevant centrality classification codes for the rural areas are: 8 = Within 30 kms of provincial capital; 10 = within 30 kms of district centres; and 11 = remote areas (CSO, 1996a, 1996c). 226 and non-agricultural) (see table 7.4) (CSO, 1996a, 1996b, 1996c; McCulloch et al., 2001, 2000). Eastern Province Rural Household Survey 2005 The selection of sites was facilitated by the existing 16 rural survey sites covered by the 1996 LCMS-I baseline survey. Thus, in July 2005 we started a panel survey, which was implemented from August to September. It incorporated all of the 16 SEAs surveyed earlier in the 1996 sample (the remaining 3 rural SEAs in Luangwa valley, of which one was located in Chipata district and two in Lundazi district, could not be revisited again because of logistical reasons). We deliberately selected four additional villages situated along feeder roads, which had knowingly benefited from the EPFRP. Thus, a total of 20 PSUs were sampled by the EPRHS 2005. However, the sampling in the SEAs was not based on the list of all households that was constructed during the 1996 LCMS-I baseline.217 We had to construct our own sampling frame instead. Our own sample was not stratified within each SEA and therefore run the risk of not being as representative of the medium-scale farmers and nonfarm rural households as the LCMS-I. Nor were an exact proportion of female-headed households included via stratification, which is another weakness of the EPRHS in addition to the smaller randomly selected sample size of only 88 rural households selected within the predetermined SEAs due to logistical problems. Table 7.5: Allocations of Standard Enumeration Areas & Weights to compute statistics in Eastern Province 1996 Total Population Rural Population Sample size (rural) Percentage of Population* Weigth (based on population) Weigth (based on sample) Rural Sample SEA Total Sample Size SEA Chipata 317719 236480 151 0,064% 60% 53% 10(9) 15(12) Lundazi 211883 163026 136 0,083% 40% 47% 9(7) 11(8) 2005 Whole Sample Sub-Region 529602 399506 287 0,072% 100% 100% 19(16) 26(20) Chipata 426478 n.a. 73 0,017% 61% 74% 12(9) n.a. Lundazi 271930 n.a. 25 0,009% 39% 26% 8(7) n.a. Whole Sample Sub-Region 698408 n.a. 98 0,014% 100% 100% 20(16) n.a. Sources: Authors‘ calculations e.g. based on CSO. Notes: * Sampling Fraction. Within PSUs the 1996 LCMS-I used stratified sampling, whereas the 2005 EPRHS used Random Sampling. 1996 sample size also include the three SEAs in Luangwa Valley. 217 CSO Zambia had for mysterious reasons lost the list of all households interviewed in 1996, something that we first learned in Chipata after we had bought all the 1996 SEA maps in Lusaka! 227 Table 7.5 provides details of the sampling frame and the actual proportions in the total sample. It shows that in 1996 the population shares in Chipata and in Lundazi within the sample were broadly consistent by being respectively 7 percentage points below and above the actual population share, whereas in 2005 the shares were 13 percentage points above and below. Table 7.6: Survey Sites used by the LCMS I 1996 and EPRHS 2005 PSU Survey Sites 1 Mafuta 2 Chiweteka 3 Kwacha 4 Jelo Farm 5 John Shawer 6 Kalume Kalinga 7 Fulato 8 Zimena 9 Chimanga 10 Chuni 11 Kalunga 12 Mulemba 13 Bila 14 Kapela 15 Chikhumbi 16 Chaloka 17 Kauka 18 Chimazuma 19 Kapaika 20 Kachindila District Chipata Chipata Chipata Chipata Chipata Chipata Chipata Chipata Chipata Chipata Chipata Chipata Lundazi Lundazi Lundazi Lundazi Lundazi Lundazi Lundazi Lundazi Constituency Ward CSA SEA Block Camp Chief Chipangali Rukuzye 6* 2* Chanje Mkanda Mkanda Kasenengwa Chiparamba 2* 4* Chipata Central Mtaya Chikuwe Chipata Central Kanjala 3* 3* Chipata Central Kanyanya Chinyaku Chipata Central Kanjala 97 2 Chipata Central Chinjara Chinyaku Chipata Central Msanga 93 4 Chipata Central Katopola Kapata-Moyo Chipata Central Msanga 95 1 Chipata Central Mtaya Chikuwe Chipata Central Kanjala 109 3 Chipata Central Chisifu Chinyaku Luangeni Khova 104 2 Kalunga (Eastern) Mtowe Sairi Kasenengwa Kwenje 28 4 Kwenje I Kwenje I Mzamane Chipata Central Dilika 129 3 Chipata Central Nsanjika Mpezeni Chipangali Chipangali 142 1 Chankhadze Chankhadze Sairi Luangeni Nsingo 125 3 Kalunga (Eastern) Katambo Mpezeni Chasefu Nkhanga 23 1 Lundazi Central Nkhanga Magodi Lundazi Central Msuzi 55 1 Lundazi Central Mapara Kapichila Lundazi Central Lunevwa 67 1 Mwase Pharaza Mwase-Lundazi Lundazi Central Vuu 30 1 Lundazi Central Kaithinde Mphamba Chasefu Manda Hill 2 1 Emusa Mwata Magodi Chasefu Kajilime 17 2 Emusa Mnuyukwa Magodi Lundazi Central Vuu 28 1 Lundazi Central Vuu Mphamba Chasefu Membe Lundazi Central Phikamalaza Phikamalaza Source: Authors. Communities within Agricultural Blocks and Camps The EPRHS was accompanied by an elaborate community module that allows linking household‘s use of social services to changes in the supply of such services at the community level. In line with the baseline study by Chiwele et al.,(1998) we also relied on the agricultural administrative demarcations that had been adopted by the Ministry of Agriculture (chapter 4).218 The camps within the agricultural blocks overlap with the pre-selected SEAs sites. Thus, a total of 16 (20) camps, considered our PSUs, were included in our 2005 EPRHS, 218 These move from the national to the provincial level and are broken down into districts according to existing political-administrative boundaries. Each district is subdivided into agricultural blocks headed by a block extension supervisor. The block is further subdivided into camps staffed by extension officers. A camp is supposed to have a radius of 15 kms but often tends to be much larger. Within a camp, the extension officer is responsible for several village extension groups (VEGs) (Chiwele et al., 1998). 228 12 of which are located in Chipata district and the remaining 8 are located in Lundazi district (table 7.6).219 As recorded in table 7.7 the topography for all the PSUs were inland plains on the plateau. We deliberately wanted to select villages from a sampling frame where the agroecological zones were as similar as possible. Table 7.7: Geographic & Demographic Characteristic of Communities, 2005 PSU Survey Sites 1 Mafuta 2 Chiweteka 3 Kwacha 4 Jelo Farm 5 John Shawer 6 Kalume Kalinga 7 Fulato 8 Zimena 9 Chimanga 10 Chuni 11 Kalunga 12 Mulemba 13 Bila 14 Kapela 15 Chikhumbi 16 Chaloka 17 Kauka 18 Chimazuma 19 Kapaika 20 Kachindila Latitude -13,28 -13,33 -13,42 -13,68 -13,53 -13,57 -13,72 -13,87 -13,80 -13,68 -13,27 -13,84 -12,15 -12,34 -12,40 -12,32 -11,70 -12,14 -12,27 -12,23 Longitude 32,47 32,21 32,37 32,50 32,61 32,50 32,53 32,50 32,12 32,77 32,84 32,75 33,05 33,22 33,28 33,13 33,25 33,03 33,02 33,28 Altitude Feet Meter 3456 1053 3180 969 3528 1075 3257 993 3191 973 3284 1001 3436 1047 3688 1124 3386 1032 3960 1207 3451 1052 4043 1232 3694 1126 3891 1186 3826 1166 3788 1155 3783 1153 3699 1127 3597 1096 3963 1208 Mean Rainfall (mm) Land Number Number of Number of of people Households female HHs Topography 1995/96 2004/05 Area (ha) Land Use Inland Plains 1124,2 1207,8 Farming Inland Plains 1124,2 1207,8 100 Planned Housing 366 130 25 Inland Plains 1124,2 1207,8 Farming 582 97 5 Inland Plains 1124,2 1207,8 50 Farming 150 70 8 Inland Plains 1124,2 1207,8 50 Farming 150 32 8 Inland Plains 1124,2 1207,8 50 Farming 380 97 28 Inland Plains 1124,2 1207,8 50 Farming 350 91 10 Inland Plains 1124,2 1207,8 150 Farming 1400 160 45 Inland Plains 818,5 1207,8 200 Farming 936 195 30 Inland Plains 1042,2 1207,8 240 Farming 1250 300 100 Inland Plains 1042,2 1207,8 50 Farming 200 35 8 Inland Plains 1042,2 1207,8 2400 Farming 894 148 25 Inland Plains 866 900,2 45 Farming 92 15 5 Inland Plains 866 900,2 450 Farming 210 54 9 Inland Plains 866 900,2 450 Farming 150 31 4 Inland Plains 866 900,2 100 Farming 500 50 5 Inland Plains 866 900,2 400 Farming 250 120 15 Inland Plains 866 900,2 50 Farming 40 16 2 Inland Plains 866 900,2 101 Farming 100 38 10 Inland Plains 1042,2 1207,8 4900 Farming 682 146 35 Note: 1 feet = 0.3048 meter. Sourceμ Authors‘ Survey results & GθS measurements. As recorded in table 7.7 the annual rainfall ranges from 850 millimetres to 1,050 millimetres in the higher –altitude plateau, concentrated between the months of November and April. The farming system in both Chipata and Lundazi was a mixed plough/hoe cereals-farming system (chapters 3-4). The resulting baseline LCMS-I sample can be considered broadly representative of the rural households in the different farming systems in Eastern Province. Notwithstanding, with only 16 rural SEAs, despite the relatively large samples of around 15 rural households within each SEA, the interpretation of the results in terms of rural Eastern Province as a whole has to be done with care. Even more so, with regards to the 219 A number of the Camp Extension positions in the valley had been vacant for extended periods. 229 2005 EPRHS, where only between 3 and 15 households were interviewed per PSU community (table 7.8).220 Table 7.8: Survey Data for Evaluation of the EPFRP, 1996 & 2005 Communities Households Interviewed EPRHS 2005 Stratification PSU Survey Sites Treatment Comparison LCMS-I 1996 EPRHS-2005 Small-scale Medium-scale 1 Mafuta Yes No 0 3 1 0 2 Chiweteka Yes No 0 5 2 2 3 Kwacha Yes No 0 2 0 2 4 Jelo Farm No Yes 15 9 5 3 5 John Shawer No Yes 16 15 14 1 6 Kalume Kalinga No Yes 15 15 13 1 7 Fulato Yes No 15 13 12 1 8 Zimena No Yes 15 7 6 0 9 Chimanga No Yes 15 3 3 0 10 Chuni No Yes 15 3 3 0 11 Kalunga Yes No 15 3 3 0 12 Mulemba Yes No 15 3 3 0 13 Bila No Yes 17 3 3 0 14 Kapela Yes No 15 3 2 0 15 Chikhumbi No Yes 15 3 3 0 16 Chaloka Yes No 15 3 3 0 17 Kauka Yes No 15 3 3 0 18 Chimazuma Yes No 15 3 3 0 19 Kapaika No Yes 15 3 3 0 20 Kachindila Yes No 0 3 3 0 Total 20 7 9 243 105 88 10 LCMS-I 1996 Stratification Small-scale Medium-scale 0 0 0 0 0 0 7 5 14 0 12 0 11 1 15 0 11 1 13 0 7 5 13 0 10 5 12 0 13 0 11 2 9 2 7 5 11 1 3 0 179 27 CSA 6* 2* 3* 97 93 95 109 104 28 129 142 125 23 55 67 30 2 17 28 SEA 2* 4* 3* 2 4 1 3 2 4 3 1 3 1 1 1 1 1 2 1 Sourceμ Authors‘ Survey results. 220 For logistical reasons this target was not possible to interview the same number or rural households. The distance from the most northern locality (PSU17) to the most Southern locality (PSU8) was 253 km. From the most western locality (PSU9) to the most eastern locality (PSU15) the distance was approximately 130 km. In other words the total area covered was no less than 32,890 km2, which is almost as large as Guinea Bissau and much larger than e.g. Equatorial Guinea, Burundi, Rwanda; Djibouti; or The Gambia [or 76% the area size of Denmark]. 230 7.3. Theoretical Framework This section presents the conceptual framework used to investigate whether access to rural feeder roads rehabilitated through the EPFRP is a critical determinant of the treated rural households‘ ability to increase their income and prove their chances of breaking out of the poverty trap. 7.3.1. Conceptual Framework Outcomes, results, effects and impacts are important, but according to Stern(2005) we need to think clearly about why this is so, what methods are appropriate; and in what circumstance. Evaluators have for example been criticized for privileging initial success by looking at outcomes that have happened after a few months or a year (cf. the UNCDF joint final evaluation report by Rwampororo et al.,(2002)), but never follow up in terms of what might be the more important longer term results or effects of particular interventions (Stern, 2005; Center for Global Development, 2006). „Impact‟ has been given an even narrower methods-led meaning, namely as ―a comparison of what happens with what would have happened had the intervention not been implemented‖ (εohr paraphrased by Stern, β00η). From this perspective impact has become identified with attribution and the counterfactual; and experimental methodologies associated with that understanding of science and research. According to Stern(2005) attribution does matter, mainly because we need to disentangle what makes a difference, ‗what works‘ from changes that have nothing to do with our efforts. ζot simply was this initiative successful, but also ‗did this intervention make a difference that would not otherwise have happened?‘ There are two complementary approaches to this attribution problem: There are comparative methods, including: before/after comparisons; quasi experiments; and full (randomized) experiments, all of which essentially depends on comparing something with something else in order to identify what it is that is distinctive in this thing that we are interested in. And the other approach is not looking at comparison across cases but is looking in-depth inside a case. This approach is known as „theory-based‟ methods. The logic here is that by going in-depth in the links of the chain between the different actions 231 that are taken it becomes more possible to understand what causes what (Stern, 2005). The main conclusion by Stern(2005) and others is that we need both approaches. In fact, according to Stern(β00η) we can compare three ‗scenarios‘μ    S1: Standardized interventions in identical settings with common beneficiaries. S2: Standardized interventions in diverse settings, possibly with diverse beneficiaries. S3: Customized interventions in diverse settings, with diverse/diffuse beneficiaries. These different scenarios necessarily pull for different methodologies:    S1: Is better adapted to experiments. S2: Is better adapted to quasi experiments and comparisons (contingent and realist) and combinations of methods. S3: Is better adapted to case studies or narrative/qualitative approaches that build plausible theories. Economists have traditionally used quantitative methods for collecting and analyzing data. As a result, many economists believe that qualitative methods are of limited use in economics or policy research. However, this is untrue according to Chung(2000). New Institutional Economics (NIE) researchers e.g. in the field of agriculture have demonstrated that combining qualitative and quantitative methods has improved the quality of their research and policy recommendations (see Carruthers and Kydd, 1997; Dorward et al., 1998). ―Theory based approaches to understand mechanisms that cannot be fully observed. Distinguishing between causality and explanation and recognizing the limits of proof and certainty. The possibility of building typologise of contexts and the possibilities of trying to understand within what sets of domain it is possible to generalize. Thus, linking process evaluations with outcome/impacts so as to understand: (a) what is being implemented, and (b) what accounts for divergence/diversity (Stern, 2005).‖ There is surprisingly little hard evidence on the size of the impact and nature of the rural transport infrastructure induced benefits and on the contextual factors that influence outcomes at the local level (van de Walle, 2008, 2009). The degree to which an improvement in rural transport infrastructure (i.e. change in accessibility) affects rural 232 development is not independent of the economic and demographic characteristics of the locality where the improvement takes place.221 Hence the analysis must consider the nature of the local economy (section 7.2) and the different actors that make consumption decisions (Banister and Berechman, 2000). A significant feature of all infrastructure projects is their investment multiplier effect which stimulates local use of factors and demands for final goods and is a function of the size of the investment but not of its type. That is, a capital investment in an infrastructure facility of any kind, at the regional level, necessarily infuses the local economy with a substantial amount of funds. In fact, in excess of ZMK2,065 billion was paid in wages within the districts of Eastern Province during the implementation of the EPFRP (Rwampororo et al., 2002). These funds, in turn, stimulated the local economy in terms of farms‘ demand for casual labour and other inputs and in terms of consumers‟ demand for goods and services. These increased demands, further stimulated economic activity – i.e. the multiplier effect, which in essence, are transfer payments (Banister and Berechman, 2000). 7.3.2. Methodological Framework Ex-post Impact Evaluation of Rural Road Improvements This chapter studies ‗small local rural road improvements for which classic evaluation tools are appropriate, i.e. non-assigned units that did not get the intervention represent what would have happened in the absence of the intervention. Thus, through an ex-post counterfactual analysis our impact evaluation will seek to establish causality and net impacts of the EPFRP interventions that are assigned to specific units – namely the PSU communities in the catchment areas as opposed to those in the control areas of the same two districts. It is acknowledged ex-ante that associations in the survey data establish neither causality nor the magnitude of the effects (Deaton, 1997). As a minimum in order to carry out the impact evaluation it is necessary to have panel data with a baseline as well as a comparison group, while allowing for factors that 221 Banister & Berechman(2000:35) regard the change in economic opportunity resulting from accessibility improvements, which is capitalized in the form of a greater use of input factors, expanded output or enhanced welfare, as economic development. 233 influence both programme placement and outcomes. Moreover, we need appropriate controls for exogenous time varying factors (e.g. shocks). It is also important to allow for the necessary time for impact to emerge (van de Walle, 2008, 2009), since it is possible that the short-term impacts recorded by Rwampororo et al.,(2002) might be quite different from the long-term impacts explored in our study. Our analysis builds on the contributions reviewed in chapter 2 and aims to apply them to the study of the EθFRθ‘s impact on consumption growth in two districts of Zambia‘s Eastern θrovince. The contribution to the literature mainly comes from the analysis of new primary survey data. The contribution of different policy factors to the the poverty incidence trend from 1996 to 2005 (see chapter 6), as well as the sub-district distribution of the poverty incidence unfortunately can only be insufficiently explored using cross-section data. In such a turbulent context, a genuine panel would have allowed a direct analysis of factors that contribute to the changes in household‘s expenditure as well as their poverty level. Contrary to a pseudo-panel cohort approach (chapter 5), with our independent cross-sections (LCMS 1996 and EPRHS 2005) we are still able to consider the impact of a much richer set of initial variables, many of which correspond to those used in the cross-country literature, but likely to be much less affected by measurement error and problems of comparability (Deininger and Okidi, 2003). Although, regression analysis only gives the right answers under ideal conditions (Deaton, 1997), the starting point for our non-experimental study of the impact of the EPFRP is a parametric linear regression model, in which the outcome variable y is related to a set of explanatory variables x derived from two successive cross-section data sets that span a medium to long-run period (1996-2005), but also coincide with considerable changes in policy. One of the x-variables is the treatment variable (i.e. accessible feeder roads), while others are ―control‖ variables (Deaton, 1997). 234 The finite number of randomly selected rural households in each of the 16(20) rural PSUs covered by both the LCMS-1996 and EPRHS-2005 varies from 3 to 15,222 which is too few to be used to make comparisons over time across these broad PSU clusters (PSU1 versus PSU2) (see discussion in Chapters 4 and 5). Instead we compare the average of households living in the catchment area with the average of those living in the control area.223 The change in average consumption would be estimated by the difference in average consumptions in 1996 and 2005 (Deaton, 1997). This approach should enable us to address the key policy question by following the same catchment and control areas over time. In this quest it is necessary to verify that the measured changes are statistically significant and thus unlikely to be caused by chance alone (Glewwe and Jacoby, 2000). Problems in questionnaire design and measurement issues Several potential problems with comparing poverty over time exist and have been discussed in the literature. Therefore we only address the potential problems related to questionnaire design and the measurement of consumption. First, the problem of changes in questionnaires over different rounds of a survey needs to be addressed.224 In order not to negatively affect comparability the questionnaire design of the 2005-EPRHSwas adapted to the homegrown 1996 LCMS-I questionnaire, with all the main items prompted for in more or less the same way. The format of the consumption questionnaire is the same in both rounds: Three questions on:  ‗did you purchase‘,  ‗did you consume from own production/stock‘,  ‗did you consume from gift or wage in kind‘, with lists of items for which the interviewee was prompted. However, the difference between the 1996 and 2005 questionnaire was that the list of items used in 2005 was 222 Scott and Holt(1982) and Pfefferman and Smith(1985) show that, although the OLS estimator is inefficient when the explanatory variables are not constant within clusters, the efficiency losses are typically small. These results provide a justification for using OLS, and a means of assessing the maximal extent to which the design effects are biasing standard errors (Deaton, 1997). 223 This data can be constructed for any characteristics of the distribution of interest, we are not confined to means. Medians can be used instead of means a technique that is often useful in the presence of outliers. 224 Grosh and Jeancard (1994) and Lanjouw and Lanjouw (1997) discuss some of the consequences if this were to happen. Appleton (1996) discusses the consequences for poverty comparisons in Uganda. 235 slightly longer, since following the three pilot testing of the questionnaire it was found that more items were commonly consumed than asked for in 1996. Nevertheless, as an additional check on the results, we recalculated the 2005 figures using only the items which were explicitly prompted for in 1996. Constructing Poverty Lines to Analyse Changes in Poverty A description of the structure of expenditure by rural household, will allow us to understand the impacts of improved feeder roads on consumers. In Zambia, as in many low income developing countries, the largest fraction of household expenditure is spent on food. In consequence, the largest impacts of improved feeder roads on the consumption side will be caused by changes in the prices of food items. Expenditures on other non-food items are relatively less important in terms of total expenditure, the welfare impacts being lower as a result (Dercon and Krishnan, 1998). The actual consumption definition used is the sum of values of all food items and non-investment non-food items. The latter was interpreted in a limited way, so that contributions for durables and house expenses were excluded. Although there may be methodological reasons to so measure welfare in practice, excluding these items is also done to avoid further bias due to different prompting of items in 1996 and 2005. However, one would expect that since 1996, and the end of isolation of many remote rural areas, households would be spending more on durables or construction – assets (e.g. a shift away from mud huts to brick houses) which typically are investments that require cash for the purchase transactions. As a consequence, again, we may, if anything, bias the results against reductions in the levels of poverty since 1996 (Dercon & Krishnan, 1998). Another standard problem is related to the valuation of own production or gift consumption. In Chipata district we made an effort to collect data on prices primarily in the two main local markets at the time of the consumption survey itself.225 However, such 225 This proved more difficult than expected. Information about prices was not easy to collect. The enumerators were given a list of well-defined items, and were required to price at two different sites in 236 a local price survey was not available in 1996 (table 7.9). So we unfortunately couldn‘t avoid the problem of using „within survey‟ prices to value the large consumption from own production or from gifts in kind. Table 7.9: Poverty Lines and Implied Inflation Rates Poverty Line Price Index Food Basket (OER adj). Implicit GDP deflator Monthly Food CPI* Basket* CSO Total CSO Food CSO Upper CSO Lower CPI* 1996 26573,54 19132,95 28979 20181 100 100 100 100 100 2005 104066,91 74928,18 113487,15 79032,55 39,70 617,12 111,85 391,62 87,52 Sources: CSO = regional price data based on Central Statistical Office price data collection; CPI = official Consumer Price Index based on urban price data; Food CPI = food Basket Price Index. Poverty lines for CSO data are population weighted averages within the sample. Notes: * July. Consumption data are available only at the household level so further corrections are needed. The same definition of the household concept was used in both the LCMS and the EPRHS questionnaire. Irrespective of the concept of the household, correcting for household size and composition is also an important issue. We calculated per adult equivalent (p.a.e.) units using World Health Organisation (WHO) conversion codes (see table A2 in Annex: Chapter 7). Since data on household size and composition was collected in each period, we adjusted the household size and the p.a.e. units in both periods. In many respects, this remains a relatively arbitrary correction, especially since consumption is not limited to just the intake of calories (Dercon and Krishnan, 1998).226 Household consumption poverty is defined relative to a poverty line. Although alternative methods to define the poverty line are possible (Anand and Harris, 1994; Greer and Thorbecke, 1986), we use the cost-of-basic-needs approach to estimate a poverty line (Ravallion and Bidani, 1994b, Dercon and Krishnan, 1998). A food poverty line is constructed by valuing a bundle of food items providing 2300 Kcal. A specific Chipata district centre. The three enumerators were not given money to make actual purchases, but instead approached the seller, explained that he or she was conducting a survey (which had nothing to do with taxes or law enforcement), and asked the price of an item. Hence, we abandoned this approach even before we continued our survey in Lundazi district. Many items are not standard or available, even on the nearest urban market. These urban markets are often 5-10 km away or located along the tarred roads M12 (Chipata-Lundazi) and T4 (Lusaka-Lilongwe) and prices relevant for the households are not necessarily the same. 226 Ravallion and Lanjouw (1995) provide a careful analysis of the robustness of poverty measures to the weight attached to household size. This is beyond the scope of our current paper. 237 value for this basket was not obtained per survey site. The reason for this was that pricing a basic basket assumes the availability of all these commodities in the local market. Indeed, during our fieldwork in 2005 we encountered problems with finding price data for some commodities in the local markets. However, the rural areas in the two adjacent districts have quite similar farming systems. Their diets are therefore fairly similar, implying similar product availability in markets, which shouldn‘t affect our pricing noticeable. Thus, we settled for a common diet for everyone, to increase comparability across sites. To this fix value of the basket covering all PSUs, an estimated non-food share is added to obtain the total p.a.e. consumption poverty line per month. The issue of prices becomes even more crucial when attempting to do comparisons over time and space. Price dispersion is high in even within Eastern Province, with markets taking considerable time to perform arbitrage (Chiwele et al., 1998; chapter 8). Also, all the rural areas in 2005 were still not well served by rural markets, e.g. due to very poor quality of the road surface in several of the PSUs. Even if markets always clear, price variability over time is high, and is not explained by seasonal factors. Such variability is very difficult to deal with in analysing poverty. Depending on whether consumption was measured when prices were high or low has important consequences for finding whether households were poor or non-poor (Dercon & Krishnan, 1998). We believe that seasonality doesn‘t present a problem, since the two surveys were carried out in the same period. The same poverty lines used for both periods and both the limited and the expanded definition of consumption also uses the same basket of commodities for both periods to increase transparency and comparability in the analysis, but valued at the prices for the survey period. The poverty lines can therefore also be thought of as a price deflator allowing comparisons across PSUs and over time (table 7.9). 238 7.4. Qualitative Findings We start with a qualitative analysis because it is generally difficult to assess the role of complementary policies, such as feeder road improvement, empirically. This participatory assessment approach of the poverty situation as seen by the poor brings us in a position to get some sense of their importance.227 The ex-post final evaluation report of the EPFRP by Rwampororo et al.,(2002) used a mixed method approach based on the concept of 'triangulation‘ to collect information from various stakeholder groups who were invited to agree or disagree with statements provided in a questionnaire administered by the UNCDF evaluation team across three districts visited from the 24th to the 30th of March 2002. This chapter likewise combines several qualitative methods. Group discussion with knowledgeable community representatives was a key approach pursued to obtain information from farmers in the 16(20) PSUs to analyse the EθFRθ‘s impacts on poverty. The qualitative data collection protocol was less structured than the quantitative standardized survey field protocols (Chung, 2000). The qualitative approach relies on triangulation to increase the internal validity of its findings,228 and it also involves combining qualitative and quantitative methods. 7.4.1. 1996 Chiwele‟s Baseline Survey Findings In 1996 six agricultural blocks in Lundazi and Chipata North were visited by Chiwele et al.,(1998). A striking feature of all the sites visited by this team of Zambian researchers was the poor accessibility due to the bad state of the feeder roads prior to the launch of the EPFRP. Distances ranged from 31 kms in Mwase Lundazi east of Lundazi district centre to 92 kms northwest of Chipata North Central. All the agricultural blocks had motorized feeder roads, which were passable during the dry season but were difficult to use during the rainy season. The bridges were particularly unreliable. The situation World ψank, 1999. ―Voices of the θoor.‖ Triangulation is the practice of using several data collection methods to assess whether a given finding is authentic. Denzin (1978), claiming that no single collection method ever provides a complete answer to a research question, conceptualized several different forms of triangulation, including triangulation by: Data source, method, researcher, and theories Chung(2000). 227 228 239 appeared to deteriorate as one moved away from the centre of the districts. Thus, in many areas, agricultural inputs and products were difficult to deliver after December. Chiwele et al.,(1998) observations with regard to the state of the roads confirmed what had been observed for Zambia as a whole at the time of their fieldwork. The ωSη‘s 1995 “Crop Forecast Survey: Supplementary Information” established that 81 per cent of farm households in the country lived within 5 kms of a public road. Only 7.3 per cent lived more than 10 kms from a public road.229 A study by the Institute of African Studies (IAS, 1996) concluded that ―the poor state of feeder roads, impassable at critical times of the agricultural season, was the major problem. Other sources of worry were the lack of bridges and other problems at watercourses (chapter 3)‖ Because of the bad state of feeder roads and perhaps low incomes, the most common mode of transport found in the study areas were ox-carts, particularly for shorter distances. For longer distances, farmers used motorized transport but many found this costly. A few farmers living in the communities have one or two trucks which were hired out to other farmers. Transportation appeared to be a major problem in 1996 for many traders as well, some of whom went to the area and bought grain and then waited for sometimes up to one week for transport to move their purchases (Chiwele et al., 1998; chapter 8). Chiwele et al.,(1998) used largely qualitative methods to obtain the necessary information. One group discussion was held in each camp, averaging about 20 farmers with a total of about 240 farmers participating in the 12 group discussions. After every group discussion, an effort was made to obtain more systematic information about the characteristics of farmers, the factors driving their participation in the market and their experience with the liberalised marketing system. For this purpose, a questionnaire was administered. A total of 188 farmers were interviewed, 98 in Lundazi and 90 in Chipata North. 229 In general of the rural people who live within 2 kilometers of an all season road from their resident around 30 to 40 percent do not have access to roads in general (Fan, 2008). 240 Chiwele et al.,(1998) mention two factors that can explain why there was little flow of maize from surplus areas to rural deficit areas. First, because of the low income prevailing in the deficit areas, it was unlikely that the prices at which maize could be sold would be economical for traders. Second, the population in deficit areas survives through one form or the other of subsistence farming. Chronic deficit areas have diversified out of maize to grow many more food crops than is the case in surplus areas. There was evidence to show that the number of crops grown per household in Lundazi was much less than that grown in Chama, the most chronic deficit area in Eastern Province (Njovu et al., 1995). The average number of crops grown by each household in Lundazi is three, in Chama it is five, with some households growing as many as seven.230 Thus, apart from their lower financial capacity to buy the commodity, farmers in deficit rural areas are also less dependent on maize for household consumption. However, a major means by which grain from surplus areas reached deficit areas was through operations mounted by relief agencies such as the Programme Against Malnutrition (PAM) in Eastern Province. Maize was distributed through food-for-work programmes (FFW). The relief organizations also engaged in the distribution of inputs as a strategy to promote food security in deficit areas (Chiwele et al., 1998). A number of farming systems were identified in 1996, which likewise was identified in 2005. The Semi-Permanent Hoe and Ox Plough Systems were very common in Zone II, with maize, finger millet, sorghum, groundnuts and beans as the dominant crops. Another system is the Semi-Commercial Ox and Tractor Plough Systems, which mostly are identified with Zone II. This system is used mostly by medium-scale and the area cultivated is usually above five hectares. Both draught and tractor ploughing is used. Cash crops, particularly maize and groundnuts, are dominant. Commercial systems, although existing in Eastern Province, were not encountered in the Chiwele et al.,(1998) study and only on one occasion in the 2005 EPRHS just outside Chipata town. 230 Chama district is in the Luangwa Valley, which is part of the Agro-ecological Zone I. The region is prone to drought and part of this diversification has been a coping mechanism against rainfall failure. 241 The results based on questionnaire interviews in Chipata and Lundazi carried out in 1996 by Chiwele et al.,(1998) indicated that small and medium-scale farmers were constrained by a number of factors that had important implications for the emerging marketing channels. Many did not have access to credit. Farmers were also forced to accept low prices for their products. This situation was not helped much by the poor state of feeder roads and the lack of on-farm storage facilities that would enable farmers to sell their produce when the price is highest. Most roads became impassable in 1996 during the rainy season (Chiwele et al., 1998). Chiwele et al.,(1998) also observed that farmers in their study areas had inadequate agricultural price and market information to enable them to make critical decision about the crops they needed to grow. All six blocks reported access to market information but the levels of access varied greatly. Farmers in Chipata had better access than farmers in Lundazi. This difference arose from the fact, that Chipata is the provincial capital of Eastern Province and had better communication facilities than Lundazi in 1996. The emerging picture was that the agricultural price and marketing system was still at a very early stage of transition from the pan-territorial and pan-temporal pricing system under which the government announced the price of maize at the beginning of the planting season and the information was disseminated through the radio and other media. Liberalisation brought about a more complex situation requiring varied information for different areas and for different times of the year (Chiwele et al., 1998). It was apparent that farmers were not empowered in 1996 to take advantage of price variations and market information due to lack of on-farm storage, the prevalent problem of impassable roads in critical times of the marketing season, and the uncertainties in the macroeconomic situation. These constraints generally made farmers feel that better price and market information didn‘t provide any advantage to them and so they took little interest in it (Chiwele et al., 1998). 242 7.4.2. 2005 Eastern Province Community Survey Findings In this section we test the robustness of the preliminary 2002 UNCDF results and extend the period of analysis to 2005. Thus, we relate the changes in feeder road surface characteristics to changes in the behaviours and outcomes of the 16(20) rural communities by conducting an analysis of a community questionnaire adapted from the usually administered LSMS survey prototype (Frankenberg, 2000). Transportation Situation in Survey Communities Full details on the distance to various public services provided in the nearest local market towns as well as road access and road quality is provided in tables 7.10-11. Table 7.10: Distances to Public Services, 2005 Community/Village Centre to District Centre Distance to Distance (km) Mini- Daily Periodic Public Health Primary PSU Survey Sites Village Est.* GPS Time (min) Bus Mkt Mkt ** Phone Centre School 1 Mafuta 32 51.5 60 0,5 0,5 0,5 32 2 7 2 Chiweteka 73,5 73 60 0,5 1 73,5 1 1 1 3 Kwacha 12 12 12 12 12 8 1 4 Jelo Farm 17 17 15 1,5 1,5 17 17 1,5 0,5 5 John Shawer 18 9.7 20 5 5 18 18 5 5 6 Kalume Kalinga 27 43.45 40 2 2 27 2 2 1 7 Fulato 14,5 14.5 12 7 7 14,5 14.5 7 3 8 Zimena 24 46.67 120 10 24 24 24 4 5 9 Chimanga 104 78 185 24 104 38 38 10 10 10 Chuni 12 4 12 12 12 4 3 11 Kalunga 60 53 60 0,2 60 60 60 5 2,5 12 Mulemba 33,5 33.8 5 5 33,5 5 5 5 13 Bila 35 24 45 35 35 35 35 2 2 14 Kapela 8 8.05 23 8 8 8 8 5 0,5 15 Chikhumbi 27 35 69 3 3 27 27 3 0,5 16 Chaloka 8 5 20 8 8 8 8 8 3 17 Kauka 45 15 72.4 45 28 28 45 5,5 2,5 18 Chimazuma 50 53 90 50 50 50 50 5 1 19 Kapaika 30 21 30 30 30 30 10 1 20 Kachindila 16 32 45 16 16 10 16 2 2 32,33 22,56 54 13,3 20,6 26,3 22,0 4,8 2,8 Average Fin. Inst. 32 73,5 12 17 18 27 14,5 24 104 12 60 33,5 35 8 27 8 45 50 30 16 32,3 Feeder Grinding Camp Road Mill Agr. Ext. 0,5 0,5 1 0,5 0,5 0,5 0,5 0,5 0,5 7.5 0,5 2 1,5 7 5 0 1,5 2 0,5 1 .9 0,5 0,5 5 1 0,5 4.82 0 2 3.5 1 1 5 5 1 1 2 0,8 1,8 2,1 Source: Authors. Notes: * Community's perception of distance from PSU to places that community residents use. ** For instance Saturday Market. The distance estimates obtained through the community interviews are verified by the accurate measurements of distance; travel time; and average speed collected by means a Global Positioning System (GPS) technology device.231 In these questions on distance, the term ―community centre‖ is used to denote a reference point from which distance is calculated. In each of the 16(20) rural PSUs a specific GPS reference point is measured. 231 On each fieldtrip a Garmin eTrex Vista C GPS device was used to register our movements; provide guidance to destinations (i.e. navigation); provide tracking of elevation and pressure; marking locations etc. 243 Table 7.10 shows that none of the PSUs in the sample had access to telephone services. Only one PSU1 was situated close to a daily market. None of the PSUs had a periodic weekly market. The local market town (i.e. the district centre), which is the only urban locality containing a wider variety of services (e.g. public phone services; hospitals; formal financial services; post office etc.)232 other than those found within the PSU such as: basic education services; health centre services; grinding mills and camp agricultural extension services. The distance to these latter public services provided in the rural areas, notwithstanding the considerable variation among the 16(20) PSUs, was on average from 1.8 to 4.8 km much closer to the rural communities than the former services, which on average was more than 20 km away. Moreover, 11 out of the 20 PSUs had themselves sponsored literacy programmes and 12 had sponsored health programmes in their own communities. 13 of the PSUs had even been involved in road repair of the local community roads. As recorded in table 7.10 the village households in the PSUs live anywhere from 0.5 km to 73-104 km from these public services. Our community survey instrument also collected data on whether motorized public transportation was available, which was only the case in 5 out of 20 PSUs (table 7.11). Since there were no reliable transport services to and from the majority of the PSUs to the district centres travel is mostly done on foot, by bicycle, by dragging on foot the bike overloaded with merchandises or by oxcart. In 17 of the PSUs the materials from which the community roads were constructed was made of dirt. ηnly γ could truly be considered as lying within the EθFRθ‘s catchment area by benefiting directly from the EPFRP gravel roads.233 On average the community roads hadn‘t been graded more than 10 years prior to the 2005 EPRHS and only 9 PSU had benefited from maintenance of the rural roads. Consequently, 10 PSU scored the quality of the roads to be 5 (i.e. bad), followed by 5 scores on 4 and 2 got the score 3. No roads were considered better than the score 3 (i.e. good). Therefore, in 2005 232 Lundazi district capital - Lundazi town - has government offices, one bank, a gas station, many small shops, a daily market, and a brand new bus station (opened Independence Day 2005). 233 The community perception doesn‘t exactly correspond to reality, since θSU1-3 all were situated along the EPFRP gravel roads. 244 there were several cases (7 PSUs) where the roads only were passable by motorized vehicle 4-8 months per year. Table 7.11: Transportation in Survey Communities, 2004/2005 PSU Survey Sites 1 Mafuta 2 Chiweteka 3 Kwacha 4 Jelo Farm 5 John Shawer 6 Kalume Kalinga 7 Fulato 8 Zimena 9 Chimanga 10 Chuni 11 Kalunga 12 Mulemba 13 Bila 14 Kapela 15 Chikhumbi 16 Chaloka 17 Kauka 18 Chimazuma 19 Kapaika 20 Kachindila Motorized public transportation Available Type Yes Minibus Yes Minibus No No No No No No Yes Van No Yes Motorbike No No No No No No Yes Van No No Road Surface in Community Dirt / Earth Roads Dirt / Earth Roads Dirt / Earth Roads Gravel Roads Dirt / Earth Roads Dirt / Earth Roads Dirt / Earth Roads Dirt / Earth Roads Dirt / Earth Roads Dirt / Earth Roads Dirt / Earth Roads Dirt / Earth Roads Dirt / Earth Roads Dirt / Earth Roads Dirt / Earth Roads Dirt / Earth Roads Dirt / Earth Roads Gravel Roads Dirt / Earth Roads Gravel Roads 4-wheeled motor vehicles Graded Maintenance Passable by vehicle Travel Below 50 VPD Years p.a. Months Yes Yes 8 1 12 Yes Yes 8 1 12 Yes Yes 0 12 No Yes 12 Yes Yes 45 1 5 No Yes 18 6 Yes Yes 1 1 12 Yes Yes 6 4 Yes Yes 20 0 7 Yes Yes 23 6 Yes Yes 2 1 12 Yes Yes 10 1 12 Yes Yes 14 0 12 Yes Yes 17 1 12 Yes Yes 4 0 6 Yes No 4 1 12 Yes Yes 2 0 8 Yes Yes 1 1 12 Yes Yes 5 0 12 Yes Yes 4 0 12 Road quality 3 4 3 5 6 5 3 5 5 5 4 5 5 4 5 5 4 3 5 4 Note: Road: 1= High quality; 2= Good quality; 3= Good; 4= Medium; 5 = Mediocre; 6= Impassable. Source: Authors. The descriptive analysis of the community survey focuses on two dimensions of access – road quality (whether PSU benefited from the EPFRP or not) and distance to district centre and distance to district road (table 7.12). By dividing the sample into two groups I find that: 11 (55%) of the PSUs could be considered as belonging to the „treatment‟ group and 9 (45%) of the PSUs constitute the counterfactual group. Concerning the quality of the community roads only 4 (20%) were considered good, all of which were located in the catchment areas.234 Another 5 (25%) were considered of medium quality spread between the two groups. Half the roads were perceived by the local communities as being of mediocre quality of which several of them were located in the treatment areas. Finally, one community road was considered impassable in the sense that motorized vehicles only could access 5 months during the year. It should be mentioned that some of the roads considered mediocre only were passable between 4 and 7 months as well. The average distance to the two district centres with their weekly local market and public services was approximately 24 km, of which 11 (55%) PSUs were located less 234 None of the roads were considered of either high (1) or good quality (2). 245 than 20 km away from the district centre and the other 9 (45) PSUs were found more than 20 km away. It took on average almost 1 hour to reach the PSUs with a land rover from the outskirt of the district centre. Distance to the nearest district road connecting SubBoma with the Boma (i.e. the district centre) and with the Bomas with each other was on average 6.37 km of which 11 (45%) were less than 5 km away and the other 9 (45%) were more than 5 km away.235 As mentioned earlier not all the district roads had benefited from the EPFRP or had been maintained regularly after the rehabilitation. Therefore access to the nearest district road didn‘t necessarily mean that the road was in good quality. The same applies for the M12 between Chipata and Lundazi, whose segment in Lundazi district until 2004 was in a very sorry state.236 Rural Transport Infrastructure Impact in Survey Communities Tables 7.12-16 show that in 12 (60%) of the PSUs some of the members of the community had been involved in the EPFRP as labourers.237 In compensation for their piecework they received on average ZMK85,000 per month (or ZMK2,833 per day), which is in line with ILO guidelines on recommending wages lower than the local wages in all cases in order not to disturb the local labour market. The variety of work offers weren‘t particularly affected by the EθFRθ. As shown in table 7.13 in all the communities farming was the most important source of employment. In two communities (PSU2 & PSU20) small-scale (gemstone) mining was the second most important source of employment. However, the most widespread nonfarm employment opportunity was in retail trade or craft and small-scale trade. Building construction and other services also appeared as way to diversify the income-generating activities. Nevertheless, only 42% of the PSUs had experienced an enhancement in their 235 Only two district roads (T4), which also is a regional one-lane highway connecting Chipata with Lusaka and Lilongwe, and M12 connecting Chipata and Lundazi, were tarred with culverts. 236 That all changed in 2005 when the potholes had been removed and the road received new asphalt, thereby connecting Lundazi with Lusaka through a new bus service. 237 It could be that some of the villagers had been involved in other labour-based feeder road projects than the EPFRP, such as Smallholder Enterprise and Marketing Programme (SHEMP) - which is financed by the International Food and Agricultural Development Agency (IFAD); Projects under the Zambia Social Investment Fund (ZAMSIF); or The Highly Indebted Poor Countries (HIPC) – Emergency Drought Recovery Programme. See last column in table 4.4 below. 246 quality of life as a direct consequence of the EPFRP. As a consequence basically all the PSUs had been subjected to emigration by community members primarily to urban areas. At the same time all the PSUs also benefited from immigration from poorer areas in e.g. Luangwa Valley or primarily the much more population dense neighbouring Malawi, which were hired to help with the various agronomical practices. However, in terms of improving these agricultural practices through public agricultural extension services (i.e. the Visit & Training system) provided by the agricultural camp extension officers, a high share of the θSUs hadn‘t received any public support.238 This leads us to question to what extent this „partial‟ Train and Visit extension system as used in these two districts of Eastern Province agriculture actually had any measurable impact on agricultural production despite the EPFRP intervention? 238 An explanation for the lack of outreach towards the more remote villages in the agricultural camps was due to the fact that the government wasn‘t providing fuel to the ωamp extension officers‘ (ωEηs) motorbikes, which had been financed by the African Development Bank. Consequently, the CEOs only covered the villages located closest to their homes. 247 Table 7.12: Impact of Eastern Province Feeder Road Project, 2004/2005 Communities Road Quality Distance to District Centre Time (min) Treatment Comparison Good Medium Mediocre Impassable less than 20 km or From DC I I (3) (4) (5) (6) 20 km greater to PSU km PSU Survey Sites 1 Mafuta Yes No Yes 26,18 No Yes 60 2 Chiweteka Yes No Yes 21,7 No Yes 60 3 Kwacha Yes No Yes 7,93 Yes No 25 4 Jelo Farm No Yes Yes 16,62 Yes No 15 5 John Shawer No Yes Yes 11,77 Yes No 20 6 Kalume Kalinga No Yes Yes 17,19 Yes No 40 7 Fulato Yes No Yes 15,39 Yes No 12 8 Zimena No Yes Yes 30,28 No Yes 120 9 Chimanga No Yes Yes 60 No Yes 185 14 10 Chuni No Yes Yes 15,9 Yes No 11 Kalunga Yes No Yes 45,33 No Yes 60 Yes 25,14 No Yes 52 12 Mulemba Yes No 13 Bila No Yes Yes 20,32 No Yes 45 14 Kapela Yes No Yes 6,84 Yes No 23 15 Chikhumbi No Yes Yes 18,13 Yes No 69 16 Chaloka Yes No Yes 4,68 Yes No 20 17 Kauka Yes No Yes 65,89 No Yes 72,4 18 Chimazuma Yes No Yes 36,18 No Yes 90 15 19 Kapaika No Yes Yes 16,61 Yes No 20 Kachindila Yes No Yes 13,9 Yes No 45 Total 16 7 9 2 5 8 1 25,39 9 7 53,28 Pct. 44% 56% 13% 31% 50% 6% n.a. 56% 44% n.a. Distance to District Road less than 5 km or 5 km greater km 0,88 Yes No 4,65 Yes No 3,2 Yes No 1,92 Yes No 9,82 No Yes 1 Yes No 1 Yes No 0,1 Yes No 13 No Yes 4,2 Yes No 4,06 Yes No 11,73 No Yes 14,62 No Yes 5,42 No Yes 14,86 No Yes 0,38 Yes No 1,83 Yes No 13,88 No Yes 9,59 No Yes 11,34 No Yes 6,71 8 8 n.a. 50% 50% Nearest Communities EPFRP District EPFRP Treatment Comparison Road Road II II labourers RD118 M12 Yes Yes No RD121 D104 Yes Yes No U33 D128 Yes Yes No U33 U33 No No Yes RD121 RD121 No No Yes RD121 RD121 Yes Yes No U33 U33 Yes Yes No RD596 RD596 No No Yes RD400 D124/RD400 No No Yes T4 Yes Yes No M12 No No Yes T4 Yes Yes No R243 D103 No No Yes RD110 M12 Yes Yes No RD110 M12 Yes Yes No R243 D104 Yes Yes No D103 No No Yes R107 D103 Yes Yes No R243 D104 No No Yes R246 D103 Yes Yes No 8 8 8 50% 50% 50% Worktype Service Immigration Origin Malawi Malawi Sourceμ Authors‘ Survey results. Table 7.13: Labour Market Issues in Survey Communities, 2004/2005 PSU Survey Sites 1 Mafuta 2 Chiweteka 3 Kwacha 4 Jelo Farm 5 John Shawer 6 Kalume Kalinga 7 Fulato 8 Zimena 9 Chimanga 10 Chuni 11 Kalunga 12 Mulemba 13 Bila 14 Kapela 15 Chikhumbi 16 Chaloka 17 Kauka 18 Chimazuma 19 Kapaika 20 Kachindila 1st Farming Farming Farming Farming Farming Farming Farming Farming Farming Farming Farming Farming Farming Farming Farming Farming Farming Farming Farming Farming Employment Sources for Community 2nd 3rd Mining Large & Retail Trade Large & Retail Trade Other Small-scale Industry Crafts & Small-scale Trade Other Services Crafts & Small-scale Trade Crafts & Small-scale Trade Large & Retail Trade Other Services Other Large & Retail Trade Crafts & Small-scale Trade Large & Retail Trade Building Construction Crafts & Small-scale Trade Large & Retail Trade Building Construction Other Services Other Mining Other Other Crafts & Small-scale Trade Crafts & Small-scale Trade Other Services Mining Other Sourceμ Authors‘ ωommunity Survey results. Emigration Temporary Destination Yes Urban Areas No Yes Yes Yes Yes Yes No Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 248 Urban Urban Urban Urban Urban Areas Transportation Areas Sales Areas Other Areas Construction Areas Other Rural Areas Urban Areas Urban Areas Urban Areas Urban Areas Urban Areas Urban Areas Urban Areas Urban Areas Urban Areas Urban Areas Farming Other Farming Service Other Other Sales Other Sales Other Other Temporary Yes Yes No Yes Yes No Yes No Yes Yes Yes Yes Yes Yes Yes No Yes Yes No Worktype Farming Farming Malawi Malawi Farming Farming Malawi Other Malawi Malawi Malawi Rural Areas Malawi Malawi Malawi Farming Farming Farming Farming Farming Farming Farming Malawi Luangwa Valley Farming Farming Small-Scale PWP Industries 1996-2001 Yes Yes No No No No No No No No No Yes No No No No No No No No Yes No Yes No No Yes No Yes Yes No Yes The most important crops cultivated were maize (local and hybrid); groundnuts and cotton. In 80% of the θSUs the production of local maize wasn‘t sold, but used for domestic consumption. A weakness in the design of the community questionnaire is that it didn‘t explicitly explore whether there were any food deficit cases amongst the individual members of the community or whether they were buying food or working for it e.g. through FFW. In PSU1, PSU4-5 the hybrid maize was sold to private buyers, and in PSU3 the maize was sold to the Food Reserve Agency (i.e. the Government). Most of the groundnuts were sold to buyers outside of the community, although in a couple of PSUs the groundnuts were sold on the local market. Through the out-grower schemes all the cottons were sold to the big private cotton companies (table 7.14). Table 7.14: Agricultural Situation in Communities, 2004/2005 PSU Survey Sites 1 Mafuta 2 Chiweteka 3 Kwacha 4 Jelo Farm 5 John Shawer 6 Kalume Kalinga 7 Fulato 8 Zimena 9 Chimanga 10 Chuni 11 Kalunga 12 Mulemba 13 Bila 14 Kapela 15 Chikhumbi 16 Chaloka 17 Kauka 18 Chimazuma 19 Kapaika 20 Kachindila Most Important Crops Cultivated 1st 2nd 3rd Maize G/nuts Beans Maize G/nuts Maize Maize G/nuts Cotton Maize G/nuts Cotton Maize G/nuts Cotton Maize G/nuts Cotton Maize G/nuts Cotton Maize G/nuts Maize Maize G/nuts Cotton Maize Beans G/nuts Maize G/nuts Cotton Maize G/nuts Soyabeans Maize G/nuts Cotton Maize G/nuts Soyabeans Maize Cotton Soyabeans Maize G/nuts Cotton Maize Cotton G/nuts Maize Cotton G/nuts Maize Cotton G/nuts Maize G/nuts Cotton Sourceμ Authors‘ ωommunity Survey results. To whom the harvest is sold Government Services 1st 2nd 3rd 1st 2nd Private Buyer Local Market None Not Sold Local Market Not Sold Advice None Government Private Buyer Private Buyer None None Private Buyer Private Buyer Private Buyer None None Private Buyer Private Buyer Private Buyer Advice None Not Sold Private Buyer Private Buyer Advice None Not Sold Private Buyer Not Sold None None Not Sold Private Buyer Private Buyer Fertilizer None Not Sold Private Buyer Fertilizer Advice Not Sold Private Buyer Private Buyer None None Not Sold Private Buyer Fertilizer Advice Not Sold Private Buyer Private Buyer Fertilizer Advice Not Sold Private Buyer Not Sold Advice Not Sold Other Private Buyer Advice Advice Not Sold Private Buyer Private Buyer Fertilizer Advice Not Sold Private Buyer Private Buyer Advice Advice Not Sold Private Buyer None None Not Sold Private Buyer Private Buyer Fertilizer Advice Not Sold Private Buyer Private Buyer Money/Credit Not Sold Other Private Buyer New Seeds Varieties Advice 3rd None None None None None None None None Advice None Other Standard of Living Determinants In table 7.12 it was recorded that only 42% of the PSUs had seen their quality of life go up because of the impact of the EPFRP. Two members of a randomly chosen focus group, interviewed in 2002 by Rwampororo et al.,(2002) on the Chadiza – Tafelansoni Road, had started businesses at the edge of the road. This was a result of increased traffic and increased opportunities. New shops had recently been opened by farmers wishing to diversify their income sources. These new retail shops were usually the result of reinvestment in shops from the profits derived from cotton growing. In table 7.15 we take a closer look at the determinants causing the change in life quality in the 20 PSUs. In fact, in 12 (63%) of the reporting PSUs the life quality 249 situation in the PSU was considered better than before the implementation of the EPFRP. 5 (26%) of the PSU the situation had actually turn worse than in 1995, whereas in 2 (11%) the community hadn‘t perceived any noteworthy change for the better or worse. In the 12 PSUs where the quality of life had gone up compared to the baseline situation the major determinant according to the communities perception were in: 5 (45%) due to the feeder road rehabilitation mainly because the sales of crops to traders predominantly of Asian origin had helped increase the agricultural production. It is also worth noticing that in 4 out of 5 (80%) PSUs, where the quality of life either had gone worse or stayed unchanged one of the major determinants was the bad state of those feeder roads, which hadn‘t benefited from the EθFRθ, therefore leading to a lack of marketing support and transportation to the PSUs. Two (17%) PSUs had seen their quality of life go up indirectly due to the feeder road rehabilitation, which provided better accessibility to medical care. On the other hand there was a split decision between whether the changes in agricultural policies actually had helped improve the quality of life in the rural areas. Partly because fertilisers, which in the past had been provided by NAMBOARD, now were considered too expensive. Life quality had also benefited from clean water through construction of new water wells and improved farming methods. However, the fact that all PSUs rely on rainfed agriculture means that drought pose a constant risk to the welfare of the vulnerable communities, especially because the on-going Global Climate Change is expected to bring less precipitation and more extreme droughts to certain parts of the world as argued in the assessment reports by the International Panel on Climate Change (IPCC).239 239 For more info see IPCC website: http://www.ipcc.ch/ 250 Table 7.15: Communities Perception of Determinants of Standard of Living, 2005 Life Quality PSU Survey Sites Compared to 1995 1 Mafuta Better 2 Chiweteka 3 Kwacha Better 4 Jelo Farm Worse 5 John Shawer Worse 6 Kalume Kalinga No Change 7 Fulato Better 8 Zimena Worse 9 Chimanga Better 10 Chuni Worse 11 Kalunga Worse 12 Mulemba No Change 13 Bila Better 14 Kapela Better 15 Chikhumbi Better 16 Chaloka Better 17 Kauka Better 18 Chimazuma Better 19 Kapaika Better 20 Kachindila Better Determinant 1 Determinant 2 Accessibility of Medical Care Feeder Road Rehabilitation Changes in Agr Policies Natural Disaster (drought) Changes in Agr Policies Feeder Road Rehabilitation Feeder Road Rehabilitation Feeder Road Rehabilitation Changes in Agr Policies Weather Natural Disaster (drought) Feeder Road Rehabilitation Accessibility of Medical Care Changes in Agr Policies Other Changes in Agr Policies Feeder Road Rehabilitation Changes in Agr Policies Other Feeder Road Rehabilitation Accessibility of Medical Care Feeder Road Rehabilitation Changes in Agr Policies Feeder Road Rehabilitation Improving Skills Changes in Agr Policies Accessibility of Soc Services Feeder Road Rehabilitation Changes in Agr Policies Other Accessibility of Medical Care Accessibility of Soc Services Accessibility of Medical Care Improving Skills Accessibility of Medical Care Improving Skills Accessibility of Medical Care Other Source: Authors‘ ωommunity Survey results. Determinant 3 Other Accessibility of Soc Services Access to farm inputs through the EPFRP Accessibility of Soc Services Water Wells Weather Lack of GoRZ support. Lack of FRR. Rain Lacking. Changes in Agr Policies Fertiliser too expensive. Marketing Natural Disaster (drought) Marketing and Transportation Accessibility of Medical Care Accessibility of Soc Services Expansion in Non-Agr Empl Changes in Agr Policies Lacking marketing support and transportation Changes in Agr Policies Drought. Fertiliser and ill health. Changes in Agr Policies Changes in Agr Policies Sales of crops to asians increase production Changes in Agr Policies Clean Water Changes in Agr Policies Accessibility of Soc Services Accessibility of Medical Care Farming of Cotton and G/Nuts up Accessibility of Medical Care Improved Farming Methods. Poverty Determinants Finally, the community representatives were asked about what they thought were the three main reasons behind the generalized high poverty levels. There were many different explanations as provided in table 7.7 below. Lack of formal borrowing opportunities e.g. to buy farm inputs, especially fertiliser, and the associated poor land productivity due to lack of fertiliser application was considered the two most important poverty determinants by 4 (21%) of the PSUs. These were followed by the low level of education and skills by the farmers which could enable them to understand and thereby absorb new farming technologies. There was a widespread belief that the communities didn‘t receive any help from the GoRZ due to the lack of interaction with the camp extension officer,240 and as a consequence they didn‘t have sufficient agricultural knowledge. Lack of employment opportunity was also a key determinant because there was no small-scale industry in the rural areas. The community representatives also admitted that laziness and drinking away their earnings from the sales of cash crops partly was responsible for the poverty situation. 240 Another explanation was that the absence of marketing boards was making the remote communities poorer, because the private Asian traders were exploiting them by buying at below market prices. However, others argue that this exploitation didn‘t happened to the same extent as during the previous regime 251 Table 7.16: Communities own Perception of the Determinants of Poverty, 2005 PSU 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Survey Sites Mafuta Chiweteka Kwacha Jelo Farm John Shawer Kalume Kalinga Fulato Zimena Chimanga Chuni Kalunga Mulemba Bila Kapela Chikhumbi Chaloka Kauka Chimazuma Kapaika Kachindila Determinant 1 Determinant 2 Lack of Empl Opportunities Other Lack of Empl Opportunities Other Poor Land Productivity No Way to Borrow Money No Way to Borrow Money Other Poor Land Productivity No Way to Borrow Money Poor Health Poor Land Productivity Laziness Low level of Edu & Skill Poor Land Productivity No Way to Borrow Money Low level of Edu & Skill Laziness Low level of Edu & Skill Poor Land Productivity Poor Health No Way to Borrow Money Poor Health No Way to Borrow Money Other Lack of Empl Opportunities Poor Land Productivity Other Poor Land Productivity No Way to Borrow Money To Old to Work No Way to Borrow Money Poor Land Productivity Lack of Empl Opportunities Low level of Edu & Skill No Way to Borrow Money Remoteness/Isolation Poor Land Productivity Source: Authors‘ ωommunity Survey results. Poverty Determinant 3 Other Poor Land Productivity Poor Health No Way to Borrow Money Poor Health Poor Land Productivity Low level of Edu & Skill Low level of Edu & Skill Other Poor Health Poor Land Productivity Other Poor Health Laziness No Way to Borrow Money No IMT Other Other Other Other No help from GoRZ Drinking No industry. No credit for farm inputs Lack of farming land and funds to buy fertilisr Rainfall Lack of Agricultural Knowledge due to no visit by CEO Bad Roads. Marketing. Boreholes lacking Lack of Land because of Growing Families Lack of water. No boreholes. Lack of info about possibilities Loosing Man Hours by looking after the sick HIV AIDS and Lack of Fertiliser Absence of Marketingboard is making them poorer. Asians buy at low prices Credit for Fertiliser is missing. Don't understand CEO training on crop rotation. Too much drinking. Despite Using Crop Rotation the Soil addicted to fertiliser. Dont receive fertiliser and no CEO to teach them new technologies Soyabean, maize and g/nuts Marketing poor due to low prices Beer Drinking of all Money earned Amongst the second most important determinants the community listed e.g. poor health (e.g. HIV/AIDS pandemic),241 old age and remoteness/isolation as possible poverty explanations. To give an example of the issue of remoteness, the cotton company Dunavant usually buy cotton from different farmers in different areas e.g. using its own trucks although it mostly depends on hiring private transporters to transport the crops to the depot/ginnery. However, due to the bad state of the feeder roads most transporters refuse to cover these cotton farming areas a second time, which makes it difficult not only for Dunavant to collect the cotton but also for farmers to sell their harvest. A couple of PSUs highlighted the fact that the increasing population growth was already causing lack of farming land in addition to the existing problems of lack of water and boreholes and the continued reliance on rainfed agriculture. Consequently in 12 of the 20 PSUs the community had experienced that more inhabitants moved away; whereas 5 PSUs experienced neither arrivals nor departures. These overall community-level findings beg the question whether improved access to the two market towns through the rehabilitation of feeder road network actually led to improved social welfare at the household level in Chipata and Lundazi district e.g. 241 The loss of working hours was common, because the healthy have to look after the sick. 252 through an increase in purchased inputs, and/or greater sales of agricultural and nonagricultural products? Finally, Deaton(1997) critiques the usefulness of the community questionnaire approach. One difficulty lies in the concept of a community. The simplest idea is a village, whose inhabitants share common health, educational, and other facilities, and who buy and sell goods in the same markets. The question is to what extent the communities in the farming scheme outside Chipata and Lundazi district centres conform to this model of having unified social or administrative structure. In consequence, even when the community questionnaire yields data on schools, health clinics, and transportation, we do not always have a clear delineation of the population served by those facilities, or on its relationship to the survey households in the cluster (Deaton, 1997). 253 7.5. Quantitative Findings In this section data from the community questionnaire above will be combined with household survey data. The two independent household surveys allow an assessment of consumption poverty, which reflects the economic opportunities available to the poor. These household surveys remain the basis for documenting poverty in Chipata and Lundazi districts. They are used to keep track of who benefited from the EPFRP. The surveys allow a good deal of disaggregation and allow us to look beyond means to other features of distributions, while distinguishing between households in catchment and control groups. 7.5.1. Changes in Socio-Economic Characteristics, 1996-2005 First the descriptive statistics regarding changes in income, assets and poverty, as well as the distribution of income and assets in the 1996 and 2005 sample is discussed. In the 1996-2005 period, Zambia experienced almost a decade of consecutive economic growth (chapter 2) with real GDP per capita 10.48 per cent higher in 2005 (ZMK91,413.81 per month) than in 1996 (ZMK82,744.07).242 Moreover, the distribution of family income - Gini index was 50.8 in 2004,243 only slightly up from 49.79 in 1996,244 thereby making Zambia the World‘s 4th most unequal society a deterioration from its 11th position in 1996.245 The widening of an already extremely high income inequality suggests that the large majority of the population didn‘t benefit from the continuous economic growth during the period in focus. Table 7.17 both provides information on the head of household and on the community‘s infrastructure and market access situation. Information is provided for the total sample (columns 1 and 2); and for the poor (columns 3 and 4) and non-poor (columns 5 and 6) households in both periods. Table 7.18 contains the same information disaggregated between Treatment areas and Comparison areas. 242 Data source: IMF World Economic Outlook database. Data source: CIA World Factbook. 244 Data source: World Bank World Development Indicator Database. 245 In 1991 with a GINI coefficient of θ0.0η Zambia was the World‘s most unequal society! 243 254 To assess the extent to which rural road infrastructure access affects a household‘s growth opportunities two dimensions of access is considered: The community (i.e. PSU) distance to nearest district road and the community distance to the district centre (of either Chipata or Lundazi); and the presence of a feeder road, which have benefited from the EPFRP, as a proxy for the quality of the main feeder road in the community. 246 Following the identification of ethnic diversity as a factor that is directly or indirectly responsible for much of Africa‘s growth tragedy (Collier and Gunning, 1999, Easterly and Levine, 1997) we construct a social-capital related variable. This will in our case be levels of ethnic diversity in the local community in 2005. Thus, we create another dummy variable equaling 1 if the Ngonis are amongst the two most important ethnic groups in the community and 0 if the Tumbuka‘s are among the two most important groups in the community.247 Information on the economic status of the head of household indicates the critical role of agricultural growth in poverty reduction in these two districts. I find that the majority share of rural households, which draw their main livelihood from farming, even goes up from 75% in 1996 to 8η% in β00η. Although the poor‘ share of above 80% clearly exceeds that of the non-poor, both social groups saw an increase from respectively 83% to 86% and 62% to 66%. Moreover, about 28% of rural households were headed by females in 2005 up from 23% in 1996. However, the increase exclusively took place among the poor female households. From the information on employment status of the rural households I find that self-employment in the rural areas is by far the most important and in many cases the only income generating activity as a result of the IMF-SAP induced roll-back of government institutions and parastatals in the rural areas in the first half of the 1990s. Around 14% 246 From table A6.1 in annex we notice that treatment in the form of feeder road rehabilitation in the period 1997-β001 doesn‘t necessarily mean that the quality of the road is good in 2005, especially if no maintenance has been carried out in between. 247 Ngonis and Chewas mainly speak Nyanja/Chewa or Nsenga, whereas the Tumbukas speak Tumbuka. 255 had a non-farm enterprise in both 1996 and 2005. The major difference was that the percentage was much higher among the non-poor in 2005 compared with 1996. Table 7.17: Descriptive Statistics for the Treatment and Comparison households, 1996 and 2005 (%) Treatment I* 1996 2005 Comparison I 1996 2005 Treatment II** 1996 2005 Comparison II 1996 2005 66,36 33,65 90,32 9,68 82,48 17,52 87,04 12,96 75,63 24,37 86,67 13,33 75,41 24,59 90 10 74,75 25,25 87,5 12,5 90 10 89,09 10,91 80 20 82,61 17,39 86,84 13,16 95,12 4,88 66,35 33,65 24,04 75,96 41,94 58,06 9,38 90,63 68,61 31,39 31,39 68,61 70,59 29,41 18,82 81,82 67,23 32,77 27,73 72,27 52,27 47,73 13,04 86,96 68,03 31,97 28,69 71,31 68,42 31,58 17,07 82,93 49,04 50,96 53,13 46,88 54,74 45,26 54,55 45,45 55,46 44,54 58,7 41,3 49,18 50,82 48,78 51,22 38,46 61,54 52,88 47,12 25,96 67,31 6,73 31,25 68,75 25 75 13,79 82,76 3,45 43,07 56,93 57,66 42,35 27,74 55,47 16,79 52,73 47,27 36,36 63,64 19,61 78,43 1,96 36,97 63,03 60,50 39,5 21,01 68,07 10,92 45,65 54,35 34,78 65,22 14,29 80,95 4,75 45,08 54,92 50,82 49,18 32,79 53,28 13,93 43,9 56,1 29,27 70,73 21,05 78,95 0 21,15 78,85 25 75 24,09 75,91 29,09 70,91 76,47 23,53 73,91 26,09 77,87 22,13 70,73 29,27 52,88 47,12 65,63 34,38 42,34 57,66 74,55 25,45 50,42 49,58 71,74 28,26 43,44 56,56 70,73 29,27 56,73 43,27 40,63 59,38 33,58 66,42 16,36 83,64 49,58 50,42 26,09 73,91 37,7 62,3 24,39 75,61 15,38 84,62 50 50 17,48 82,48 34,55 65,45 10,08 89,92 50 50 22,95 77,05 29,27 70,73 n.a. n.a. 46,88 53,13 n.a. n.a. 30,91 69,09 n.a. n.a. 32,61 67,39 n.a. n.a. 41,46 58,54 56,73 43,27 42,31 57,69 n.a. n.a. 40,63 59,38 27,13 71,88 40,63 59,38 33,58 66,42 56,2 43,8 n.a. n.a. 16,36 83,64 43,64 56,36 49,09 50,91 24,37 75,63 49,58 50,42 n.a. n.a. 13,04 86,96 26,09 73,91 52,17 47,83 62,3 37,7 50,82 49,18 n.a. n.a. 39,02 60,98 51,22 48,78 39,02 60,98 Economic status of Head Farming/fishing/forestry Other Employment status of Head Self-employed Other Education of Head Never Attended School Ever Attended School Attained less than or equal to 5 years of schooling Attained more than 5 years of schooling Age of Head Between 20-40 years More than 40 years Assets No Bicycle Bicycle Ownership No Radio Radio Ownership Walls material: Bricks Walls material: Mud Walls Material: Other Gender Female-Headed Households Male-Headed Households Land Ownership per household Member Less than 0.5 hectare More than 0.5 hectare Lagged rainfall (avg) Below average Above average Access to Credit No access to informal borrowing Access to informal borrowing Agricultural Extension Service No Access to Advice Access to Advice Infrastructure Distance to District Centre more than 20 km Distance to District Centre less than 20 km Distance to District Road more than 5 km Distance to District Road less than 5 km Time more than 30 min from District Centre Time less than 30 min from District Centre Notes: * Based on Road Quality. ** Based on Village Labour participation in EPFRP. Sourcesμ Authors‘ calculations. We use the number of school years completed by the head of household to represent the initial human capital endowment. ωoncerning the εDG β of ―Ensuring all children complete primary education by 2015. Zambia achieved an increase of 19% in primary school completion rates moving from 64% in 1990 to 83% in 2006 (UNDP, 2008).‖ This positive national trend is not captured by the EPRHS, which find an overall 256 decline both in terms of head of households with more than and less than 5 years of primary education. However, we do find a clear increase in the percentage of non-poor rural household with more than 5 years of primary education, which may suggest that education is indeed an important determinant of the ability to escape from poverty (chapter 2). Rural electrification faces many challenges in Zambia such as long distances from existing power stations to targeted rural areas, low population densities, high poverty levels etc (Haanyika, 2008). These and other factors have contributed to the fact that in none of the PSUs visited in 2005 did the households have access to electricity. Finally, concerning the rural transport infrastructure, I find that the treatment areas on average are located closer to the district centre‘s most important market place, than the comparison areas. 257 7.5.2. Linkage between EPFRP and Household Welfare 7.5.2.1. Poverty Levels and Changes from 1996 to 2005 Table 7.19 below provides descriptive evidence on both the growth in per adult equivalent (p.a.e.) expenditure in constant 1996 prices for panels of different household categories. It is noted that despite the overall economic growth in Zambia in this period, large differences persist across these categories. Moreover, the growth is very sensitive to the choice of price deflator (see section 7.5.4.2 below). On the other hand the growth trend is not particularly affected by whether we choose either ‗Total ωomprehensive Expenditure‘ or ‗Total δimited Expenditure‘ in real values (1996 prices). In the first case the average annual growth rates fell as much as -8.91% p.a. compared with -7.57% p.a. in the latter case (table A8a in Annex: Chapter 7). Amongst the 16 PSUs only one (PSU=10), which is even considered a counterfactual, experienced positive annual average growth rate, but the result was based upon only 3 non-stratified observations in 2005 compared with the 15 stratified observations in 1996.248 These sub-district results are contrary to the officially recorded positive growth trend at the national level in Zambia. Another surprisingly result, as recorded in column 7, is that the negative average annual growth rate of -9,88% p.a. was higher in the chosen treatment areas than the 7,99% p.a. found in comparison areas (table A8a in Annex: Ch.7). Using the „limited total expenditure‟ the figures of respectively -9,17% and -6,20% p.a. give the same picture. These results correspond to the negative consumption growth results based on the natural logarithm of the ‗ωomprehensive‘ (or ‗δimited‘) expenditure as recorded in column 8. Similarly these results are also at odds with the officially recorded growth trends in Zambia. The first interpretation of these contradictionary results that comes to mind is that the total household expenditures are underestimated in 2005 because of a likely undervaluation of own production or gift consumption associated with using ‗within survey‘ prices. On the other hand, these quantative findings correspond to the qualitative Using the „Limited Total Expenditure‟ p.a.e. instead we find that both PSU=10 (counterfactual) and PSU=6 (catchment area) have positive annual growth rates. However, the positive growth rate for the latter is much smaller with only 1.05% p.a. compared with 23.27% in PSU=10. 248 258 finding above that only 37% of the PSUs had seen their quality of line going up since 1996, while the small sample size makes it difficult to capture the within PSU variation. The male-headed households outperformed the female-headed households in the sense that they on average had a somewhat smaller negative annual growth rate. The difference in performance was much clearer between the young head of households in the age range from 20 to 40 years compared with the head of households older than 40 years of age, whose ‗ωomprehensive Expenditure‘ fell almost twice as much per year. Yet another result that goes against common sense is that the head of households with more than 5 years of education, which in both samples constitutes the large majority, was outperformed by the head of households with less than 5 or no education. The most probable answer to this is related to the small number of observations of households (13) with less than 5 years of education. Not surprisingly, the large households of more than five family members in p.a.e. terms were significantly outperformed by the smaller households whose ‗ωomprehensive‘ and ‗δimited‘ total expenditure per month fell η.γ% p.a. and 4.ηη% p.a. respectively. Finally, the rural households with a land ownership larger than 0.5 hectare per person at -7.01% p.a. also performed better than those households with less or equal to 0.5 hectare per person, who experienced a fall of -9.17% p.a. In other words, for the panel rural households included in our sample, the level of poverty increased from 1996 to 2005 across the board for all categories included. A graphical illustration of the extent to which overall p.a.e. ‗ωomprehensive‘ expenditure in real terms fell between 1996 and 2005 is provided in figure 7.1 and figure 7.2, which plot the cumulative density of the logarithm of this variable across households for the 1996 and the 2005 survey.249 It appears fairly clear from the two figures that the logarithmic 1996 distribution dominates the 2005 logarithmic distribution in real values. That is, if ranked by the distribution of ‗ωomprehensive Total Expenditure,‘ rural households in the 16 PSUs were unequivocally better off in 1996 than they were in 2005. 249 The 10 highest outlier observations have been trimmed away. 259 In 1996 around 79% of the sampled rural households were living for less than 1 US dollar per day, whereas the percentage had increased to 95% in 2005. Thus, despite the aggregate economic growth experienced in Zambia, almost the entire rural population in these 16 PSUs was worse off at the end of the 9 year period. Again, this finding to some extent corroborates the qualitative findings above (section 7.4). Figure 7.3 and figure 7.4 show that the comparison areas oddly dominated the treatment zones throughout the entire distribution, whereas the disparity of the middle range of the quintiles (q2-q4) in 2005 seems to have narrowed quite significantly. The 199θ distribution‘s domination of the β00η distribution is clearer when looking at the treatment areas compared to the comparision areas (Figures A5.2-5.3 in Annex: Ch.7). The domination is partly explained by the fact that the initial value of the average total household expenditure was higher in the comparison areas than in the treatment areas in 1996. In fact this disparity had increased in 2005 (Tables A8a-b in Annex: Ch.7). Moreover, we also find that the share of food falls from 81.49% for the 2nd decile in 1996 to 68.23% for the 9th decile, whereas the food share in 2005 fluctuates between 60.83% for the 3rd decile and 66.71% for the 8th decile again perhaps due to the underestimation (i.e. measurement error) of the value of the own production (Table A9 in Annex: Ch.7). 260 Table 7.18: Poverty Levels, 16 PSUs (1996: n=241 & 2005: n=88) Average per capita expenditure: Total Per Adult Equivalent Expenditure: Total Food Food Comprehensive Limited Comprehensive Limited Comprehensive Limited Comprehensive Limited Food Poverty Line 19132,95 55,60% 70,12% 74,69% 85,06% 50,62% 63,07% 67,63% 82,57% Poverty Lines (ZMK), 1996 Total Poverty Line Upper 26573,54 28979 74,27% 77,59% 82,16% 85,06% 85,48% 88,38% 93,36% 93,36% 65,98% 71,78% 78,01% 81,33% 80,08% 83,82% 90,04% 92,12% Lower 20181 59,75% 73,03% 75,93% 86,31% 53,53% 65,98% 70,95% 82,99% Food Poverty Line 74928,18 86,36% 89,77% 94,32% 97,73% 82,95% 87,50% 93,18% 96,59% Poverty Lines (ZMK), 2005 Total Poverty Line Upper 104066,91 113487,15 94,32% 95,45% 96,59% 96,59% 97,73% 97,73% 97,73% 97,73% 92,05% 93,18% 94,32% 94,32% 96,59% 96,59% 96,59% 96,59% Lower 79032,55 88,64% 92,05% 95,45% 97,73% 86,36% 90,91% 94,32% 96,59% Notes: Limited definition means that the list of items explicitly prompted for in 1996 is used in 2005 as well [without Own-consumption and excluding durables]. Comprehensive definition uses all data food and non-food consumption items recorded in the survey [including Own-consumption and excluding durables]. Sourceμ Authors‘ computations. Table 7.19: Changes in Comprehensive Total Expenditure and Poverty, 1996 and 2005 (%) Food Pove rty Line 19132,95 Price De fla tor 1,1185 Tota l Ex pe nditure pa e including Ow nconsumption Tre a tme nt I Compa rison I Tre a tme nt II Re gions 1996 2005 1996 2005 Re e l Va lue s Loga rithm Compre he nsive Compre he nsive Compre he nsive Compre he nsive Ex pe nditure , Ex pe nditure , Ex pe nditure , Ex pe nditure , pa e pa e pa e pa e 38034,16 38494,52 10,55 10,56 30096,67 57943,44 10,31 10,97 28457,66 54664,21 10,26 10,91 Grow th (% p.a .) Compre he nsive Ex pe nditure , pa e 0,13% 7,55% 7,52% Consumption Grow th: Δln cons 1996 Compre he nsive Ex pe nditure , pa e 0,012 0,655 0,653 obse rva tions 104 32 135 55 119 46 2005 Compa rison II 38601,18 46442,93 10,56 10,75 2,08% 0,185 120 41 Le ss tha n 5 km 40489,34 58131,08 10,61 10,97 4,10% 0,362 120 54 33 More tha n 5 km 26553,63 38776,84 10,19 10,57 4,30% 0,379 119 Le ss tha n 20 km 37408 56733,76 10,53 10,95 4,74% 0,416 136 65 Distance to District Centre More tha n 20 km 28457,43 33228,15 10,26 10,41 1,74% 0,155 103 22 Eve r a tte nde d school 40865,16 34910,20 10,62 10,46 -1,73% -0,157 78 33 School Attendance Ne ve r a tte nde d school 30006,96 63679,04 10,31 11,06 8,72% 0,752 161 49 Le ss tha n 5th gra de 23273,14 43540,44 10,06 10,68 7,21% 0,626 68 13 Highe r tha n 5th gra de 33207,31 52063,36 10,41 10,86 5,12% 0,450 160 74 Ow ne rship 25894,99 56819,93 10,16 10,95 9,12% 0,786 142 48 Bicycle Ownership No Ow ne rship 44757,89 43368,13 10,71 10,68 -0,35% -0,032 97 39 Le ss tha n 40 ye a rs 34555,97 37953,49 10,45 10,54 1,05% 0,094 115 40 Age Olde r tha n 40 ye a rs 32618,28 61714,35 10,39 11,03 7,34% 0,638 124 47 Ma le he a de d 35326,68 55419,5 10,47 10,92 5,13% 0,450 184 63 Gender Fe ma le he a de d La rge la nd ow ne rship (>0.5/pe rson) Sma ll la ndow ne rship (<=0.5/pe rson) or la ndle ss 27608,98 38636,88 10,23 10,56 3,80% 0,336 55 24 36655,14 66767,43 10,51 11,11 6,89% 0,600 126 25 30088,99 44347,23 10,31 10,70 4,40% 0,388 113 62 36714,17 29373,56 33550,64 56507,7 33896,07 50789,82 10,51 10,29 10,42 10,94 10,43 10,84 4,91% 1,60% 4,71% 0,431 0,143 0,415 136 103 239 65 22 87 Distance to Road Highest Grade Attained Land ownership Rainfall Above a ve ra ge Be low a ve ra ge W hole Sourceμ Authors‘ ωomputations. 261 In table 7.20 we explore the extent to which the relative disparities across the treatment and comparison areas have taken place despite the increased poverty. This is done by comparing inequality in the distribution of p.a.e. comprehensive expenditure by using measures of economic income inequality such as: The Atkinson index; Relative mean deviation; Gini coefficient and the Theil entropy measure. We find that the simplest of these measures of inequality, the relative mean deviation,250 for the whole 1996 distribution fell from 0.42 to 0.36 in 2005 in line with the CDF curves as illustrated in figures 7.1 and 7.2. The fall was very pronounced for head of households older than 40 and head of households with less than 5 years of education. We also see a widening in disparity between the treatment areas and comparison areas with the latter spending 31.32% more than the former in 2005 up from 20.81% in 1996 despite the fall in the relative means in both areas. These trends are similar to those measured by the Atkinson index, which gauges movements in different segments of the consumption distribution. While the relative mean deviation is based on the differences between the x values and the arithmetic mean, the Atkinson measure takes the deviations not from the arithmetic mean but from a quantity, described as the ‗equally distributed equivalent consumption‘.251 The Atkinson measure, which indicates the proportionate loss of consumption, is only interested in the effect of unequal consumption on total welfare, rather than on the inequality of individual welfares (Sundrum, 1992). The Gini coefficient is more intuitive since it is based on the Lorenz curve.252 We find that estimated Gini coefficient of 49.12 for the whole 2005 distribution is slightly less unequal compared to the 1996 distribution of 55.9. 250 It has the simple economic interpretation of being equal to the proportion of total income that must be transferred from the rich to the poor in order to make all incomes equal (Sundrum, 1992). 251 This quantity is interpreted as the income which, if received by every member of the population, would give the same total welfare as the distribution of incomes actually observed (ibid.). 252 The Gini coefficient could also be considered as half of the Relative mean difference. The mean difference is the average absolute difference between two items selected randomly from a population, and the relative mean difference is the mean difference divided by the average, to normalize for scale. 262 Figure 7.2: Cumulative Density of pae expenditure in 2005 0 0 .2 .2 .4 .4 .6 .6 .8 .8 1 1 Figure 7.1: Cumulative Density of pae expenditure in 1996 8 9 10 LNTotexppae96_Com 11 12 7 8 9 10 LNTotexppae05_Com 11 12 Source: Authors. Source: Authors. Figure 7.3: Cumulative Density of pae expenditure in 1996 Figure 7.4: Cumulative Density of pae expenditure in 2005 CDFs for Treatment and Comparison Regions, 2005 0 0 .2 .2 .4 .4 .6 .6 .8 .8 1 1 CDFs for Treatment and Comparison Regions, 1996 0 50000 100000 Totexppae_Com Treatment Zones Source: Authors 150000 200000 0 Comparison Zones 50000 100000 150000 Totexppae_Com Treatment Zones Source: Authors 263 200000 Comparison Zones 250000 Table 7.20: Measures of Inequality of per adult Expenditure, 1996 and 2005 1996 Panel Sites (PSUs) Whole Treatment Comparison by gender of head: male-headed households female-headed households by age of head: Age head (20-40) Age head (>40) by education level of head: head has more than 5 years of education head has less than 5 or no education by household size: large household (>5) Small household (<=5) by land ownership: Large land ownership (>0,5/person) Small landownership or landless 2005 Atkinson 0,4201 0,3284 0,3150 0,2777 0,4472 0,3564 1996 2005 Relative mean deviation 0,4153 0,3567 0,3540 0,3039 0,4277 0,3991 1996 2005 Gini coefficient 0,5590 0,4912 0,4763 0,4210 0,5767 0,5353 1996 2005 Theil entropy measure 0,7848 0,5282 0,4300 0,3143 0,9778 0,6571 0,4449 0,3261 0,3547 0,2189 0,4295 0,3642 0,3823 0,2808 0,5782 0,4831 0,5185 0,3745 0,8725 0,4310 0,5934 0,2326 0,4025 0,4334 0,3580 0,2455 0,3960 0,4246 0,3993 0,2860 0,5412 0,5659 0,5245 0,3866 0,6262 0,9963 0,6075 0,2472 0,4234 0,4107 0,3492 0,1844 0,4162 0,4104 0,3792 0,2155 0,5612 0,5484 0,5158 0,2950 0,7720 0,8166 0,5819 0,1504 0,4890 0,3824 0,3192 0,3012 0,4462 0,4003 0,3565 0,3106 0,6050 0,5320 0,4875 0,4326 1,1261 0,6100 0,6617 0,3417 0,4612 0,3522 0,3117 0,3182 0,4442 0,3828 0,3245 0,3520 0,5911 0,5032 0,4516 0,4866 0,9755 0,4533 0,3795 0,5857 Source: Authors‘ ωomputations. The Theil entropy (T) measure253 can be viewed as another way of comparing a given distribution with the corresponding egalitarian distribution, devised by taking the difference between the logarithms of the observed consumption and their mean value, instead of the difference between their numerical values (Sundrum, 1992). In sum table 7.5 shows a big fall from 1996 to 2005 for all categories. Emphasis on relative rather than absolute welfare confirms the feeling by many of the representatives of the 16(20) PSUs communities acquired through the qualitative community surveys that their (relative) situation has failed to improve along with the absolute fall in mean income in real values (section 6.4).254 This finding of narrowing inequality of rural household expenditure in the period 1996 to 2005 suggests that the diminishing growth opportunities have been shared relatively equitably, which provides a motivation for empirically exploring this issue further in the next sections 253 A Theil index of 0 indicates perfect equality. A Theil index of 1 indicates that the distributional entropy of the system under investigation is almost similar to a system with an 82:18 distribution. This is slightly more unequal than the inequality in a system to which the "80:20 Pareto principle" applies. 254 Given the low overall levels of asset endowments, e.g.: Vehicles; proper farm/non-farm equipments and livestock, but consisting instead mostly of items such as: Radio; bicycle and tools, we chose not to focus on the households‘ ability to over-come long-term poverty by levels of asset ownership. 264 7.5.2.2. Decomposing poverty changes, 1996 and 2005 Rows 1-2 of table 7.21 identify the changes in the incidence of poverty for the cross-section of rural households living in the treatment areas versus those living in the comparison areas. Note that by using the „food poverty line‟ the change in headcount poverty increase by 33.72% in the former case slightly more than in the comparison areas. The highest increase is found in the large households with more than 5 members followed by those with a distance of at least 20 km to the nearest district centre. The smallest increase is found for the smallest households with smaller or equal to 5 household members, followed by head of households with less than 5 years of education (again based on only 13 observations) and households with relative large landownership, which saw an increase in the incidence of poverty of respectively 20%; 25%; and 27%. From a further disaggregation at the PSU-level we find that in 1996 the headcount ratio ranged from 20% in PSU11 and in PSU17 as well as 25% in PSU=5 to as much as 80% in PSU6 and PSU10 and even 100% in PSU8. In 2005 we find that in 7 out of the 19 PSUs all the sampled rural households were living below the food poverty line of ZMK74928.18 per month. Notwithstanding the wide variation found using this type of disaggregation, we conclude that the changes in poverty for the rural households across Chipata and Lundazi districts highlight that despite the rehabilitation of the feeder road network there wasn‘t any pronounced difference in the large increase of rural poverty between the treatment areas and the comparison areas. The initial socio-economic characteristics of the households had a bigger impact on the poverty headcount. 265 Table 7.21: Decomposing poverty changes per adult equivalent, 1996 and 2005 Comprehensive Total Household Expenditure Regions Treatment Comparison by gender of head: by age of head: FGT(0): Poverty Headcount in 1996 0,5 0,51095 FGT(0): Poverty FGT(1): FGT(1): Change in Headcount Change in Poverty Poverty Poverty in 2005 Headcount Gap, 1996 Gap, 2005 Gap 0,83721 0,33721 0,18795 0,47623 0,28828 0,82222 0,31127 0,2016 0,45208 0,25048 FGT(2): Squared Poverty Gap, 1996 0,08863 0,10604 FGT(2): Squared Poverty Gap, 2005 0,31313 0,28912 Change in Squared Poverty Gap 0,2245 0,18308 Total Number of Obs: 1996 104 137 Total Number of Obs: 2005 43 45 male-headed households 0,50538 0,8125 0,30712 0,19507 0,45971 0,26464 0,09742 0,2951 0,19768 186 64 female-headed households 0,50909 0,875 0,36591 0,19785 0,475 0,27715 0,10228 0,31621 0,21393 55 24 Age head (20-40) Age head (>40) 0,48413 0,53043 0,8125 0,85 0,32837 0,31957 0,19331 0,19833 0,4443 0,48738 0,25099 0,28905 0,09852 0,09854 0,2767 0,32984 0,17818 0,2313 126 115 48 40 head has more than 5 years of education by education level of head: 0,50282 0,84 0,33718 0,20335 0,47607 0,27272 0,10449 0,31277 0,20828 177 75 head has less than 5 or no education 0,51563 0,76923 0,2536 0,17457 0,39357 0,219 0,08204 0,23211 0,15007 64 13 by household size: large household (>5) 0,5122 0,90741 0,39521 0,22513 0,55336 0,32823 0,12238 0,37091 0,24853 82 54 Small household (<=5) 0,50314 0,70588 0,20274 0,18053 0,32176 0,14123 0,08623 0,18958 0,10335 159 34 0,44531 0,72 0,27469 0,16465 0,3258 0,16115 0,08059 0,19119 0,1106 128 25 Small land-ownership or landless (<=0.5 pc) At least 5 km Less than 5 km 0,53435 0,47107 0,54167 0,87302 0,81818 0,83636 0,33867 0,34711 0,29469 0,20992 0,17428 0,21731 0,51867 0,45492 0,46926 0,30875 0,28064 0,25195 0,10756 0,08531 0,11186 0,34437 0,29344 0,3053 0,23681 0,20813 0,19344 113 121 120 63 33 55 At least 20 km 0,48571 0,86364 0,37793 0,18813 0,52099 0,33286 0,09086 0,33925 0,24839 105 22 Less than 20 km Whole 0,52206 0,50622 0,81818 0,82955 0,29612 0,32333 0,20156 0,19571 0,44484 0,46388 0,24328 0,26817 0,10445 0,09853 0,28805 0,30085 0,1836 0,20232 136 241 66 88 by land ownership: Large land ownership (>0.5 pc) by Distance to allweather road by Distance to nearest District Centre Sourceμ Authors‘ computations based on Total Comprehensive Monthly Household Expenditure and the National Food Poverty Lines in 1996 and 2005. 266 7.5.3. Regression Analysis This section provides estimates from regressions of household-level expenditure on a p.a.e. basis and the incidence of poverty on the socio-economic characteristics; the distance variables and the EPFRP treatment dummy variable. 7.5.3.1. Consumption Growth Determinants We use our non-experimental household survey data to look at the differences in consumption behaviour between rural households in the 16(15) PSUs, and to try to relate the degree of exposure to the treatment to variation in the consumption outcomes. The starting point for our non-experimental study is the classical linear regression model (CLRM), in which the outcome variable „monthly comprehensive total expenditure p.a.e.‟ (y) is related to a set of explanatory variables (x). The EPFRP dummy variable is the treatment variable. In order to ensure that the results are not biased by the presence of other confounding factors, we control for these other variables. These “control” variables include: Agro-ecological potential via lagged rainfall and soil type. We also include a number of household characteristics that might affect the consumption level: Economic status; employment status; health; gender; age; and education of the household head; as well as landownership per household member; assets; non-farm income generating activities; and access to borrowing opportunities. The error term in the regression captures omitted controls, as well as measurement errors in the outcome y, and is assumed to satisfy: E[u|x] = 0, that its expectation conditional on the x‘s is 0.255 We would like to know whether improvements in access to the PSUs through improved road quality derived from the EPFRP are associated with making the rural households better off in the medium-to long term. We thus examine some of the determinants of growth in living standards between 1996 and 2005. Our focus is on the relative accessibility in terms of roads, and general remoteness (distance), and the 255 These variables play the same role as the control group in an experiment (Deaton, 1997). 267 associated changes in monetary living standards in this period. We found in section 7.5.1 that the mean growth in this period was -8.91% p.a. (growth in consumption p.a.e.), but with high variability between PSUs, with negative growth extremes ranging from -0.12% p.a. (PSU6) to -17.78% p.a. (PSU11), with one outlier (PSU10), which recorded positive growth of +20.57% p.a. in this period. Ravallion and Jalan(1996) found strong evidence that initial conditions matter with generally divergent impacts with respect to community characteristics and generally convergent effects with respect to household-level initial conditions. We explore whether these negative growth rates can explained by the rural infrastructure project (EPFRP) and remoteness (distance). The retained results from a total of 25 (XXV) specifications of the household level equations with „monthly comprehensive total expenditure p.a.e.‟ as the dependent variable are reported in table 7.22. In particular, through the successive OLS regressions we find as reported in columns 3, 8, 13, 18, 22 and 23 that improved access to the district roads through the EPFRP Treatment did in fact have a significant effect on the poverty proxy. However, it is known from the above poverty analysis that the poverty level in 2005 is much higher than in 1996 and therefore this fact could explain the negative coefficient in column 22 (XXII). Moreover, we know that at the PSU-level access to the rehabilitated feeder roads wasn‘t entirely exogenous. To the extent that such a supply was geared towards areas with high potential agricultural production return in a predominantly agricultural local economy, the estimated coefficient would be biased upwards. In 2005 the distance to the nearest district roads as measured in logarithm no longer has a significant and positive effect on the household expenditures. This could simply be due to the fact that we for the 1996 LCMS-I doesn‘t know exactly where the sampled households lived within the PSU/SEA, whereas we for the 2005 EPRHS have the exact GPS coordinates for the community that the sampled households belonged to. Therefore the interpretation of the association between this distance covariate and the household‘s expenditure level should be done with extreme caution. 268 Table 7.22: Determinants of Total Household Expenditure in per adult equivalent, 1996 & 2005 Variable Type Logarithm of Monthly Comprehensive Expenditure pae DV Treatment I DV Treatment II CV Distance to District Road CV Logarithm of Distance to District Road CV DV Distance to District Centre Distance to District Road above/below 5 km DV Distance to District Centre above/below 20 km DV DV Economic Status: Farming or other Employment Status: Selfemployed or not DV Ever Attended School or not CV Highest Grade Attended (education) CV DV Education Squared Attained Higher or Lower than 5 years of edu 1996 I 2005 II -.4742413 (.4952709) .0016589 (.0127423) -.0015936 (.040097) OLS Merged (i) III 2005 IV Merged (i) V .3968411*** (.1232567) .0153651 (.0109888) .3215632 (.2350871) .1925632* (.1149598) .0286421 (.025994) .002883 (.0102207) OLS Merged (i) VIII OLS Merged (i) XIII .3677176*** (.0952221) .3598253*** .4164411*** (.0982635) (.0960602) .0523246 (.0327083) .0040054 (.0031848) .0006889 (.0072788) .0004353 (.0030573) .0047992 (.0081238) .0049177 (.0033984) .2182571** .2084271** (.0962861) (.095739) -.6275409*** (.2203333) .2873274 (.2871824) -.84904 (.6774846) -.6765497 (.7338837) -.2503784 (.2627343) .0437221 (.2801749) 2005 XXII OLS Merged (i) XXIII -.2753647** (.1221995) .4663692*** (.0993153) 2005 XXIV Merged (i) XXV -.0561012 (.199511) -.0817247 (.1026827) .0670558 (.0567008) -.0036415 (.0477536) .0690874** (.0337662) .0141031 (.0719367) .0396156 (.0344621) -.0615164 (.1822479) .204561* (.1056502) .2119388** (.0995133) .3132282 (.2255412) .1195227 (.1071096) -.1115192 (.116713) -.180438 (.1308619) -.1537065 (.1011211) -.2620068 (.1893818) -.0847546 (.1043507) -.2769279 *** (.0926679) .2377376 (.184064) -.1183159 (.0956383) .2098894 (.1838484) -.1060184 (.0991907) -.0053791 (.0125657) -.0217352 (.0157707) -.0220635 (.0177327) -.0219205 (.041047) -.0211428 (.0184344) .0000139 (.0001102) .0001347 (.0001533) .0001647 (.0001855) .0001413 (.0004505) .0001404 (.0001923) .0674394 (.110552) .1332185 (.2167896) -.3816513 (.2653794) .0653528 (.2866297) -.1563691 .0981573 -.1579171 (.0972868) -.0737849 (.11215) .0082952 (.0079541) .5502065** (.213134) -.0447573** (.0175011) -.0254448 (.0693522) .0048745 (.0050081) -.0186519 (.0706518) .00498 (.0051115) .3382108* (.1646906) .196153 (.2357068) -.5104014*** (.1042309) -.1455635** (.055584) .1600416 (.1112317) .4909702 (.1156222) .0351456 (.1265676) -.0539574** (.0209919) .1400069 -.0259248 .1133902 (.0949723) .4945508*** .4723414*** (.1178519) (.094126) .0167833 .1293437 -.0510844** -.0193255 (.0214832) (.0171599) -.0216446 -.0124853 (.0965786) (.0943837) .5441691*** .4694583*** (.0954949) (.0933365) .0001666 (.0001785) -.1913357** (.0901479) .1490832 .1719197 (.1077323) (.1083658) .0670179 .1059377 DV Bicycle Ownership DV Radio Ownership DV Wall Material: Bricks or not CV Age CV Age Squared .0003772** (.0001689) .0014994** (.0006251) .0004989** (.0002287) .0004908** (.0002336) .0001379 (.0001794) DV DV Below or Above 40 years of age Gender: Male or Female headed household .2339179 (.1561088) .2387956 (.2307554) .2676504** (.1350368) .2796017** (.1375795) .1647816 (.1094227) CV Landownership per household member (ha) .0441936 (.0164922) .6339749* (.3480228) .0321864* (.0167565) .0365506** (.017005) .0406256** (.0166959) DV Landownership above or belove 0.5 ha pc -.0060116 (.0059881) -.0008771 (.0101115) .013991*** (.0041137) -.0216284 (.0170599) .0457802*** (.016686) .1817602** (.0904392) Rainfall lagged (mm) DV Above or Below Average Rainfall in period DV Access to Informal Borrowing or not -.1287557 (.1516035) .023512 (.28346) -.1892549 (.1304674) -.2029538 .1332401 Cotton sale .000017*** (2.83e-06) 1.84e-09 (1.79e-07) 1.73e-07 (1.34e-07) 2.10e-07 1.36e-07 10.68554*** (.9002113) 174 0,2501 13.36002*** (2.34374) 37 0,6094 9.052062 (.7933331) 211 0,3344 .0380961 (.1204084) .0918326 (.1476264) .0553547*** (.0184187) .6442336*** .0457175*** .6131341*** (.1473631) (.0173443) (.2027083) .035298 (.11419) .003465 (.0075576) -.0009941 (.0037318) .013933*** (.0032789) 9.707851*** (.8603322) 239 0.1009 10.8213*** (.7087706) 74 0.2521 8.523347*** 10.4955*** 9.337442*** (.5956275) (1.202393) (.6057894) 313 74 313 0,1415 0.2332 0,0807 .0431691** (.0179397) .0113818*** (.0031949) .0090544** .0115366*** (.0038077) (.0032185) CV Constant Observations R-Squared .0593639* (.0325422) 1996 XXI -.0009761 (.0027092) -.1665437 (.1490543) .2380366** (.1141718) -.0532164 (.1800068) -.0357052** (.0141136) CV .0474211 .033047 OLS Merged (i) XVIII .0003945 (.0056254) .0100051*** (.0033126) .1473699 (.1012396) 37 0,6075 9.645674 .7802947 211 0,3087 8.67305*** (.5777955) 313 0,1959 9.278319*** 8.537616*** (.1965425) (.5730284) 326 313 0,1685 0,1791 Notes: Linearized Standard Errors in parentheses. *** Significant at 1%; ** 5%; * 10%. (i) Standard OLS without declaring that the data is complex survey data. Sourceμ Authors‘ computations. 269 At the same time the distance from the community centre to the nearest district centre (i.e. the market place) remained significant in 2005 although the magnitude of the coefficient increased and the signed changed to negative according to expectations. The importance of assets in the form of bicycle ownership is sensitive to the choice of model specification, although it is seem to be significantly robust. Even more importantly in the last specification in columns 7-8 (i.e. VII-VIII) its importance becomes even more pronounced because the sign switch from negative in 1996 to positive β00η, which tells us about the household‘s consumption benefits from owning a bicycle that can be used on the rehabilitated roads to get to the local market. Education measured in terms of number of years doesn‘t seem to have an impact despite the government‘s overall effort to achieve universal primary education (εDG2) during that period. This might be explained by the large amount of missing values. On the other hand, when looking at human capital as „school attendance‟ the sample size increase and this leads to a positive significant impact on the household‘s monthly consumption level in 2005 (column 2) from an insignificant effect in 1996 (column 1). Moreover, the magnitude of the coefficient also increases. Finally, larger households experienced a significant negative impact on ‗expenditure levels‘ in both periods at the 1% level but with a higher coefficient magnitude in 2005. This explains why a higher „landownership per capita‟ had a significant positive effect on ‗expenditure levels‘ in both periods, but with a higher magnitude again in 2005 compared with 1996 (column XXI and XXII) To sum up, both the results for the 2005 and the pooled (i.e. merged) data to some extent, given the caveats mentioned above, support the hypothesis that the delivery of public services, the provision of physical public capital as well as human capital each are positively associated with ‗monthly expenditure p.a.e. levels‘, perhaps more clearly than what would have been the case when relying on cross-country regressions at the macro level. 270 7.5.3.2. Poverty Determinants The Poverty Gap measures the depth of poverty; that is how far, by which households fall below the basic needs poverty line (BNPL), where the non-poor have zero poverty gap). In the OLS fitted model above we throw away all the observations where the poverty gap (yi= 0), which leaves us with a truncated distribution with the various problems that it creates (Baum, 2006).256 If the relationship parameter is estimated by regressing the observed yi on xi, the resulting OLS regression estimator is inconsistent. It will yield a downwards-biased estimate of the slope coefficient and an upwards-biased estimate of the intercept. In other words, conventional regression methods fail to account for the qualitative difference between limit (zero) observations and nonlimit (continuous) observations. In a seminal paper, Amemiya(1973) has proven that the likelihood estimator suggested by Tobin for this model is consistent (Cameron and Trivedi, 2009, Wooldridge, 2002). Thus, to take account of all the information in yi properly, we must fit the model with the tobit estimation method, which uses maximum likelihood to combine the probit and regression components of the log-likelihood function (log L). The distribution that applies to the sample data is therefore a mixture of discrete and continuous distributions, where the two parts corresponds to the classical regression for the nonlimit observations and the relevant probabilities for the limit observations, respectively (Greene, 2000). The Tobit model or “Tobin‟s probit” is a special case of a censored regression model, because the latent (unobserved) variable (i.e. the index variable) y*i cannot always be observed while the independent variable xi is observable (or denotes a vector of exogenous and fully observed regressors): 256 In the truncated case, we observe neither the dependent nor the explanatory variables for individuals whose yi lies in the truncation region. In contrast, when the data are censored we do not observe the value of the dependent variable for individuals whose yi is beyond the censoring point, but we do observe the values of the explanatory variables (Baum, 2006). Thus, values of the dependent variable in a certain range are all transformed to (or reported as) a single value (Greene, 2000:905). We do not observe the exact value of the latent variable, only that it is zero or less, so that the contribution to the log likelihood is the logarithm of the probability of that event (Deaton, 1997). 271 y*i = x‘i + ui, (5.1) where ui ~ ζ(0, 2 i = 1 ,…, ζ, ), and xi denotes the (Kx1) vector of exogenous and fully observed regressors. The observed variable yi is related to the latent variable y*i through the observation rule: (5.2)  y * if y*  L y L if y*  L The probability of an observation being censored is: (5.3)  ( L  xí  )  Pr(y* < δ) = θr(x‘i + < L) =   ,    Where Φ( ) is the standard normal cumulative distribution function (CDF) (Cameron and Trivedi, 2009, Wooldridge, 2002). We first fit the model with OLS, ignoring the censored nature of the response (poverty gap) variable (i.e. if y* were observed). This is in turn followed by a refitting of the model as a tobit and by indicating that „poverty gap‟ is left censored at zero (i.e. the censoring point). We run the linear tobit model without any transformation of the dependent variable, even though it appears that the data distribution in the two samples may be non-normal (see sub-section 7.5.4.3 below). Finally, following the tobit estimation, we first generate the marginal effect of each explanatory variable on the probability that a household will have a positive poverty gap. 272 Table 7.23: Determinants of Poverty OLS I Tobit* II -647.5263 (394.48) 189.3678 (1155.422) 470.2412 (1265.393) 3221.657** (1258.191) 57.10364 (208.1209) Poverty Gap 2005 OLS Tobit* III IV 7695,58 8289,656 5181,857 5684,958 4,182797 113,6646 1694,034 1860,841 -4473,495 -5802,606 5369,332 5898,336 4109,968 5665,165 4756,815 5216,264 -5003,778 -6069,701 4449,615 4863,628 1435,264 1591,569 974,6225 1089,234 1996 Variable Type DV Treatment I CV DV Logarithm of Distance to District Road Distance to District Centre above/below 20 km DV Ever Attended School or not DV Bicycle Ownership CV Age -345.7027 220.7756 11.77719 628.1732 541.3272 686.441 1475.323** (667.6764) 17.79944 114.735 CV DV Age Squared Gender: Male or Female headed household -.1008392 1.190736 -616.5159 757.3785 -.297941 (2.148791) -1276.77 (1398.933) -14,9114 10,73418 -1987,675 5265,674 CV Landownership per household member (ha) -179.9308* 103.1346 -501.563** (245.4533) -16056,23 4862,028 DV Rainfall lagged (above/below avg) 443.0155 664.2466 2763.893 2699.573 239 0,065 691.3791 (1231.931) -1827.863 (4949.898) 239 1974,033 5258,652 6665,833 22688,05 74 0,2652 Constant Observations R-Squared Pseudo R-Squared Log-likelihood Sigma 0.0072 -1346,7919 7748.177 (556.4379) -16,28756 12,05694 -2297,151 5715,119 Merged Squared Poverty Gap 2005 OLS Tobit* IX X 4.87e+08 5.37e+08 (3.34e+08) (3.61e+08) 2.56e+07 4.01e+07 (1.09e+08) 1.18e+08 -2.89e+08 -4.04e+08 (3.46e+08) (3.75e+08) 2.62e+08 3.82e+08 (3.07e+08) (3.31e+08) -3.02e+08 -3.80e+08 (2.87e+08) (3.08e+08) 4.94e+07 5.62e+07 (6.28e+07) (6.96e+07) Merged OLS Tobit* XI XII 6.37e+07 -2.32e+08 1.05e+08 (1.63e+08) -1.22e+07 -7.17e+07 3.56e+07 5.31e+07 9.03e+07 5.06e+07 1.04e+08 (1.58e+08) 1.97e+08* 3.25e+08* (1.06e+08) (1.61e+08) -9.18e+07 2.04e+07 1.01e+08 (1.55e+08) 1.46e+07 2.88e+07 1.85e+07 (2.83e+07) OLS V -1153,699 (1906,517) -732,4871 (645,7109) 995,497 (1887,374) 4126,663 (1915,101) -581,8129 (1823,49) 329,1533 (335,501) Tobit* VI -6654,031** (3027,156) -1795,716* (989,206) 334,0521 (2936,467) 6417,348** (2991,842) 1636,548 (2884,221) 574,678 (525,1718) OLS VII Tobit* VIII -1859721 (2692873) 3400762 (7662034) 1.19e+07 (8372746) 1.72e+07** (8143868) -35852.3 (1399461) -5278246 4664930 6168586 1.37e+07 1.22e+07 1.50e+07 3.91e+07*** (1.49e+07) 439599.9 2462547 -4,102733 (3,511595) 594,8383 (2090,57) -6,316863 (5,487352) -275,5138 (3267,201) 668.5788 (14523.8) -6657097 (9237994) -1784.365 25419.31 -1.48e+07 1.66e+07 -2727,342*** (785,5584) -1877113 (1257966) -5880077** 8.92e+08*** -1.64e+09*** 3.64e+07** (2961329) (3.13e+08) (5.32e+08) (1.81e+07) -1.36e+08*** (4.23e+07) -243,5494 (3110,631) -3668,49 (12508,07) 313 8398792 (8102034) 2.24e+07 (3.29e+07) 239 0.0558 1.24e+07 (1.46e+07) -3.56e+07 5.86e+07 239 3.26e+07 (1.68e+08) -4.86e+08 (6.74e+08) 313 -24826,51*** -841,3006** (7485,779) (327,1613) 1622,518 5706,139 4916,375 25453,77 74 1996 1199,799 (1962,5) 4463,333 (7984,874) 313 0,0693 0,0179 -700,38745 18906,24 (1790,774) 0,0076 -2177,6721 21764,86 (1225,358) 0.0038 -2485.5418 9.12e+07 6402353 -408496.6 (691925.2) -1.20e+08 (3.39e+08) - 2.41e+08 (3.39e+08) 4.73e+08 (1.46e+09) 74 0.2172 -443987.9 (770950.2) -1.43e+08 (3.62e+08) -175763.1 193868.8 7.99e+07 1.15e+08 - 2.23e+08 1.06e+08 (3.61e+08) (1.08e+08) 4.44e+08 -4533322 (1.63e+09) 4.41e+08 74 313 0.0552 0.0083 -1373.0952 1.19e+09 1.12e+08 -302638.5 (295268.2) 3.69e+07 1.76e+08 0,0033 -4162,3676 1,16e+09 6,41e+07 Notes: Standard errors in parentheses. * Since some of the dependent variables are censored, we use a Tobit rather than an OLS framework. CV = Continuous variable. DV = Dummy variable. 1996: Obs. summary: 130 left-censored observations at Poverty_Gap<=0; 183 uncensored observations; 0 right-censored observations. 2005: Obs. Summary: 13 left-censored observations at Poverty_Gap<=0; 61 uncensored observations; 0 right-censored observations. Merged data: Obs. summary: 130 left-censored observations at Poverty_Gap<=0; 183 uncensored observations; 0 right-censored observations. Sourceμ Authors‘ computations. 273 The results from the OLS and Tobit regressions of poverty levels on the same list of independent variables reported in table 7.23 largely confirm the importance of the factors identified above in table 7.22. The inconsistent OLS estimates (i.e. the conditional mean differs from x‘i because of censoring) appear as a basis for comparison with the maximum likelihood Tobit estimates.257 The consistency of the Tobit estimates requires that the distribution of errors be homoskedastic and normal, and biases can occur when it is not (see section 7.5.4.3 below). For our case of left-censored data with the censoring point (L=0), the density function has two components that correspond, respectively, to uncensored and censored observations. Let d=1 denote the censoring indicator for the outcome that the observation is not censored, and let d=0 indicate a censored observation. The (standard normal) density can be written as:  1  1      x'i   exp  2 ( yi  x'i  ) 2    f(yi) =   2  2       2 di (5.4) 1 d i The second term in (5.4) reflects the contribution to the likelihood of the censored observation. εδ estates of ( , 2 ) solve the first-order conditions from maximization of the log likelihood based on (5.4). These equations are nonlinear in parameters, so an iterative algorithm is required (Cameron and Trivedi, 2009, Wooldridge, 2002). The EPFRP Treatment does not have a positive significant impact on the poverty gap in 2005, although this seems to be the case when we pooled the two datasets, thereby creating a larger sample size. The importance of the log distance to the nearest district road seemed to have switched to statistically negative in the pooled dataset in terms of its impact on the poverty gap only. The importance of the distance to the district centre remains statistical insignificance in all specifications both with regards to poverty gap and with respect to the squared poverty gap although the sign does switch. Only the 257 The interpretation of the coefficients is as a partial derivative of the latent variable, y*, with respect to x. It is standard to use the default estimate of the variance-covariance matrix of the estimator (VCE) for the tobit MLE, because if the model is mis-specified so that a robust estimate of the VCE is needed, it is also likely that the tobit MLE is inconsistent (Cameron & Triverdi, 2009). 274 Tobit regressions support the idea of making rural road infrastructure part and partial of rural poverty reduction programme as proposed in the Zambian PRSPs. Although the sign of the coefficient is not the anticipated one, the importance of public services is also illustrated by significant and large coefficient on the „ever attended school‟ co-variate, which is associated with (lower) levels of poverty in the pooled dataset. From our own observations the increased access to education is not directly associated with the improved rural feeder roads, since the path leading to the schools usually are not the same. One co-variate stands out from the rest and that is the higher the landownership per household member, which is statistically significantly associated with the lower the poverty level in both periods. This supports Sachs‘(2009) recent gloomy warning that along with technology, ―steps to stabilize population growth will be needed to make a reasonable – let alone a hopeful – human future possible. We should redouble our efforts to stabilize the human population. Every country should take responsibility to bring their population increases under control through voluntary reductions of fertility...‖258 As illustrated in table 7.23 almost without exception, it is found that the inconsistent OLS estimates are smaller in absolute value than the MLEs.259 Table 7.24: Predictions after Tobit, 1996 & 2005 Year Percentiles Poverty_Gap Merged Data Fitted values Poverty_Gap Fitted values 1996 Poverty_Gap Fitted values 2005 1% 32.95 -29764.78 32.95 -5395.56 146.8837 -3594.736 5% 747.45 -5528.539 433.3992 -2140.477 7758.766 18031.43 10% 2350.175 -2257.014 1132.95 -1557.182 18375.66 19950.11 25% 6042.673 2185.481 4219.207 124.5969 28081.41 24426.31 50% (*) 9845.941 5150.548 7875.267 1598.138 39206.3 32551.08 75% 32647.54 9372.556 9843.735 3037.582 46911.63 37123.41 90% 45625.38 12663.16 12155.76 4321.689 55269.04 42637.34 95% 53869.87 14576.9 14431.46 4784.203 58796.31 46005.36 99% 61883.84 20020.39 15839.39 5803.015 62966.63 52688.28 Mean 18402.34 4850.139 7396.859 1378.658 37050.52 31611.59 Obs 194 183 122 122 72 61 Note: (*) The Median value of y. Sourceμ Authors‘ calculations. Predictions after Tobit If we compare the sample statistics of the in-sample fitted values of the latent variable, y*, for all observations with those for Poverty Gap it shows that the tobit 258 This pessimistic view is in line with the warnings from Dr.Albert Bartlett, who is one of the few voices of common sense when it involves the dangers of overpopulation. 259 A striking empirical regularity is that the maximum likelihood estimates can often be approximated by dividing the OLS estimates by the proportion of nonlimit observations (Greene, 2000:912). 275 model fits poorly in both the lower and upper tail of the 1996 sample, whereas the tobit model for the 2005 sample likewise fits especially poorly in the lower tail but rather much better in the upper tail of the distribution (table 7.24). Since the tobit model has a probit component, its results are sensitive to the assumption of homoskedasticity.260 The tobit model imposes the constraint that the same set of factors x determine both whether an observation is censored (e.g. household is nonpoor) and the value of a noncensored observation (the gap to the BNPL). 7.6. Robustness of the Results In this section we check the sensitivity of the results to different poverty lines. This is followed by a sensitive analysis of poverty measures to changes in the price deflator. 7.6.1. Robustness of Poverty Levels and Changes In tables 7.25-28 we show what a crude device the poverty line concept is as an object of policy. Four poverty measures have been constructed, which remain fixed in real terms. The fact that the lines have changes very little in real terms reflects a view of poverty as an absolute; poverty is defined by the ability to purchase a given bundle of goods. A good deal of the literature on poverty has followed Atkinson(1987) in recognizing that the poverty line is unlikely to be very precisely measured, and trying to explore situations in which poverty measures are robust to this uncertainty (Deaton, 1997). Table 7.25: Food poverty levels 1996-2005; 16 PSUs (1996: n=241; 2005: n=88) 1996 - CSO prices & comprehensive HH, Exp, definition 1996 - CSO prices & limited HH, Exp, definition 2005 - CSO prices & comprehensive HH, Exp, definition 2005 - CSO prices & limited HH, Exp, definition P0 P1 0,50622 0,19571 0,63071 0,29812 0,82955 0,46388 0,875 0,54148 P2 0,09853 0,17249 0,30085 0,37709 Notes: CSO = poverty measure using poverty line valued at CSO national price survey; Comprehensive definition = total food exp + Total non food exp + own consumption; Limited definition = Comprehensive expenditure – own consumption. The standard errors of each measure are not reported. Sourcesμ Authors‘ calculations. Robust standard errors are not available for Stata‘s tobit command (ψaum, β00θμβ66). Maddala and Nelson(1975), Hurd(1979) and others all have varying degrees of pessimism regarding how inconsistent the MLE will be when heteroscedasticity occurs (Greene, 2000). 260 276 Table 7.26: Total poverty levels 1996-2005; 16 PSUs (1996: n=241; 2005: n=88) P0 P1 P2 1996 - CSO prices & comprehensive HH. Exp. definition 1996 - CSO prices & limited HH. Exp. definition 2005 - CSO prices & comprehensive HH. Exp. definition 2005 - CSO prices & limited HH. Exp. definition 0,65975 0,30863 0,17486 0,78008 0,41381 0,26346 0,92045 0,58242 0,41137 0,94318 0,64877 0,48555 Sourcesμ Authors‘ calculations. Table 7.27: Upper poverty levels 1996-2005; 16 PSUs (1996: n=241; 2005: n=88) P0 P1 P2 1996 - CSO prices & comprehensive HH. Exp. definition 0,71784 1996 - CSO prices 2005 - CSO prices & 2005 - CSO prices & limited HH. Exp. comprehensive HH. & limited HH. Exp. definition Exp. definition definition 0,81328 0,93182 0,94318 0,34055 0,44573 0,6105 0,67321 0,19873 0,28998 0,44092 0,51355 Sourcesμ Authors‘ calculations. Table 7.28: Lower poverty levels 1996-2005; 16 PSUs (1996: n=241; 2005: n=88) P0 P1 P2 1996 - CSO prices & comprehensive HH. Exp. definition 0,53527 0,21263 0,10922 1996 - CSO prices 2005 - CSO prices & 2005 - CSO prices & limited HH. Exp. comprehensive HH. & limited HH. Exp. definition Exp. definition definition 0,65975 0,86364 0,90909 0,31634 0,48325 0,55994 0,18613 0,31834 0,39466 Sourcesμ Authors‘ calculations. The tables 7.25-28 show that the headcount ratio (P0), defined as the fraction of the population below the poverty line, in 1996 varies from 50.6% to 71.78% when using the comprehensive consumption measure and P0 varies from 63% to 81% when using the limited consumption measure. Whereas, in 2005 the P0 varies from 82% to 93% and from 87.5% to 94.3% respectively. One way of doing better is to use the poverty gap measure (P1), which can be interpreted as a per capita measure of the total short-fall of individual welfare levels below the poverty line; it is the sum of all the shortfalls divided by the population and expressed as a ratio of the poverty line itself (Deaton, 1997). In 1996 we see the variation of the P1 measure from 0.195 to 0.34 and from 0.298 to 0.445 using respectively the comprehensive and the limited consumption measure. Likewise, in 2005 we find variations ranging from 0.46 to 0.61 and from 0.54 to 0.67 respectively. 277 The final poverty measure the Foster, Greer and Thorbecke(1984) index (P2) is the one that penalize the poverty gap the most like the Sen index, therefore the most used. Again P2 is lowest when using the food poverty line and highest when using the upper poverty line in both 1996 and in 2005. 7.6.2. Sensitivity of Poverty Measures to Price Changes Dercon and Krishnan(1998) point to the dangers if no careful choices are made with respect to price data: If we were to make poverty comparisons simply using the CPI as the appropriate adjustment of the cost of-living over time, then we are likely to underestimate the cost of basic needs, i.e. underestimate the level of poverty in our sample in 2005, in comparison to 1996. Part of the reason is likely to be the fact that the CPI is based on urban data only.261 Stochastic Dominance Testing In order to check whether the over-time pattern is robust to changes in the location of the poverty line and the selection of the poverty index, we applied a first-order dominance test.262 Thus, checking for respectively first-order; second-order and thirdorder poverty dominance and estimating the values of the poverty line at which the two FGT curves for 1996 and 2005 cross, we find no intersection and thereby that the 2005 distribution is entirely to the left and above the 1996 one. Thus, distribution 1 (1996) dominates (in welfare) distribution 2 (2005) indicating an unambiguous increase in poverty and extreme poverty for the entire period 1996-2005. Ideally, one would like to have some dominance results in terms of ranges of inflation rates over which one can confidently predict that the poverty orderings over time (or across space) remain the same. Standard stochastic dominance tests do not allow for this problem. Effectively, the dominance results as in Atkinson(1987) are for poverty comparisons with a common poverty line for “real” consumption, i.e. consumption values comparable over space or over time (Dercon and Krishnan, 1998). 261 Dercon & Krishnan(1998) illustrates the problems with using the CPI within the rural sample as a means of adjusting the poverty line over time. 262 See Ravallion(1992) for a further discussion of dominance. 278 In table 7.29 the formulation of the poverty measure defined over nominal comprehensive consumption in 2005 is corrected for price changes using different price deflators. Given index 1996 equals 100, if we use the food basket (391 in 2005)263 or the monthly Consumer price index (CPI) in July (617.12 in 2005) the annual growth in consumption for the whole sample as well as the treatment and control areas would be negative with double digit negative annual growth using CPI as price deflator. The opposite situation is the case when using the annual inflation rate (39.7 in 2005) and implicit GDP deflator (87.52 in 2005) with significant double digit annual consumption growth, when using the annual inflation rate as price deflator. On the other hand, deflating by using the price index from the food basket price adjusted for US-dollar exchange rate changes (111.85 in 2005) we get single digits positive growth rates. 264 Table 7.29: Sensitivity of Poverty Measures to Price Changes 1996 2005 Nominal Values Regions Price Deflator* 1996 2005 1996 Reel Values 2005 Logarithm Consumption Growth: Δln cons = Growth (% p.a.) ln C2005 - ln C1996 1996 2005 Comprehensive Comprehensive Comprehensive Comprehensive Comprehensive Comprehensive Comprehensive Comprehensive Expenditure, Expenditure, Expenditure, Expenditure, Expenditure, Expenditure, Expenditure, Expenditure, pae observations pae pae pae pae pae pae pae Whole 33594,08 56808,41 33594,08 14506,06 10,42 9,58 -8,91% -0,840 241 87 Treatment 30783,77 47263,83 30783,77 12068,84 10,33 9,40 -9,88% -0,936 104 43 35727,46 66136,07 35727,46 16887,88 10,48 9,73 -7,99% -0,749 137 44 Whole 33594,08 56808,41 33594,08 9205,37 10,42 9,13 -13,40% -1,295 241 87 Treatment 30783,77 47263,83 30783,77 7658,75 10,33 8,94 -14,32% -1,391 104 43 35727,46 66136,07 35727,46 10716,85 10,48 9,28 -12,52% -1,204 137 44 Whole 33594,08 56808,41 33594,08 50788,26 10,42 10,84 4,70% 0,413 241 87 Treatment 30783,77 47263,83 30783,77 42255,15 10,33 10,65 3,58% 0,317 104 43 35727,46 66136,07 35727,46 59127,44 10,48 10,99 5,76% 0,504 137 44 Whole 33594,08 56808,41 33594,08 143084,28 10,42 11,87 17,47% 1,449 241 87 Treatment 30783,77 47263,83 30783,77 119044,19 10,33 11,69 16,22% 1,353 104 43 35727,46 66136,07 35727,46 166578,02 10,48 12,02 18,66% 1,540 137 44 33594,08 56808,41 33594,08 64911,75 10,42 11,08 7,59% 0,133 241 87 Comparison Comparison Comparison Comparison 3,916 6,171 1,119 0,397 Whole Treatment Comparison 0,875 30783,77 47263,83 30783,77 54005,70 10,33 10,90 6,44% 0,133 104 43 35727,46 66136,07 35727,46 75569,94 10,48 11,23 8,68% 0,133 137 44 Notes: * See Table 3.2: Poverty Lines and Implied Inflation Rates. Sourcesμ Authors‘ computations. 263 264 Which is equivalent to the JCTR price deflator of 3.0178 in Lusaka (see Table A13.2). When using the Lusaka data from the JCTR dataset the price deflator is 57.45 in 2005. 279 Similarly, in tables 7.30a-b we see how sensitive the three FGT-poverty measures are to the choice of price deflator. Table 7.30a: Poverty Comparisons with a common Food Poverty line for „real‟ comprehensive household Expenditure 2005 Comprehensive HH Expenditure, Definition CPI* 0,03448 0,0085 Monthly CPI* Food Basket* Implicit GDP deflator Food Basket (OER adj). P0 P1 1996 - CSO prices & comprehensive HH, Exp, definition 0,5062 0,1957 0,93103 0,61865 0,82759 0,45772 0,13793 0,04772 0,2069 0,0709 P2 0,0985 0,0027 0,44805 0,29282 0,02287 0,037 Notes: FGT(0): Headcount ratio (proportion poor); FGT(1): average normalised poverty gap; FGT(2): average squared normalised poverty gap. Sourceμ Authors‘ computations. Table 7.30b: Poverty Comparisons with a common Total Poverty line for „real‟ comprehensive household Expenditure 2005 Comprehensive HH Expenditure, Definition P0 P1 P2 1996 - CSO prices & comprehensive HH, Exp, definition 0,6598 0,3086 0,1749 CPI* 0,0690 0,0181 0,0078 Monthly CPI* Food Basket* Implicit GDP deflator Food Basket (OER adj). 0,9655 0,7102 0,5553 0,9195 0,5776 0,4046 0,2069 0,0817 0,0431 0,3793 0,1280 0,0667 Sourceμ Authors‘ computations. 7.6.3. Robustness of Regression Analysis The Tobit model remains the standard approach to modelling a dependent variable of a linear regression that displays a large cluster of limit values (Greene, 2000), in our case zeros over some interval of its support.265 In other words, the sample is a mixture of observations with zero (i.e. censored) and positive (i.e. uncensored) values (Cameron & Trivedi, 2009). In the 1996 sample of 241 observations, there are 119 (49.4%) observations with zero values of the poverty gap dependent variable. In the 2005 sample of 87 observations, there are 15 (17.2%) zero values of the response variable. The tobit ML estimator (MLE) is consistent under the stated strong assumption above, which are likely to be violated when applied to our case study. A detailed summary of Poverty Gap provides insights into the potential problems in estimating the parameters of the tobit model with a linear conditional mean function. In both the 1996 and the 2005 sample the Poverty Gap variable is heavily skewed and has negligible non265 One interpretation is that zero poverty is a left-censored observation, that equals zero when y* < L = 0 (i.e. lower or left cutoff or censoring point). 280 normal kurtosis at respectively 2.378 and 1.932.266 This feature of the dependent variable should alert us to the possibility that the tobit MLE may be a flawed estimator for the model. In the 1996 sample the skewness is almost removed but the kurtosis is reduced only a little if the zeros are ignored. In the 2005 sample skewness is likewise almost removed, but kurtosis goes up a little from 1.93 to 2.78. Therefore, in this section we will test the assumptions of normality and heteroskedasticity. Heteroscedasticity in the Tobit Model In the case, where the normality assumption is correct, and the disturbances homoskedastic, maximum likelihood overcomes the inconsistency of OLS. However, it is a fact that regression functions estimated from survey data are typically not homoskedastic. Since there is heterogeneity between PSUs, there will be heteroskedasticity in the overall regression function (Deaton, 1997). In the presence of heteroskedasticity, OLS is inefficient and the usual formulas for standard errors are incorrect. In cases where efficiency is not a prime concern, we may nevertheless want to use the OLS estimates, but to correct the standard errors by using a formula that is robust to the presence of both heteroskedasticity and cluster effects (idem). Serious difficulties occur with heteroskedasticity when analyzing censored regression models where the inconsistencies are a good deal more troublesome (Deaton, 1997).267 In the face of the heteroskedasticity, the Tobit procedure doesn‘t necessarily yields estimates that constitute an improvement compared to the biased OLS estimates. Tobit for lognormal data The tobit model relies crucially on normality, but household expenditure data are often better modeled as lognormal (cf. section 7.5.1.1 above). A tobit regression model for lognormal data introduces two complications: A nonzero threshold and lognormal y. We observe that: (6.1)  y * if ln y*   y  0 if ln y*   266 Higher kurtosis means more of the variance is due to infrequent extreme deviations, as opposed to frequent modestly-sized deviations. 267 Deaton(1997:86) illustrates this using the censored regression model, or Tobit. 281 Here it is known that y = 0 when the Poverty Gap data are censored, and in general ≠0. In the 199θ sample of β41 observations, there are 1ββ (49.4%) zero values of the Poverty Gap and in the 2005 sample of 87 observations, there are 15 (17.2%). From a detailed summary of the Logarithm of the Poverty Gap data (i.e. LN(Poverty_Gap)) we find that LN(Poverty_Gap) is almost symmetrically distributed, with the median approximately equal to the mean, except for the lowest 5% of the distribution, which perhaps explains why the nonnormal kurtosis oddly has gone up from 2.298 to 10.39, since we anticipated that the tobit model is better suited to modeling LN(Poverty_Gap) than Poverty_Gap. Similarly, the mean is almost equal to the median in the 2005 sample distribution, while the kurtosis goes up from 2.78 to 24.09 due to the logarithmic transformation of the dependent variable. Table 7.31: Tobit on Log Poverty Gap Transformation Tobit Log Poverty_Gap Treatment I Logarithm of Distance to District Road Distance to District Centre above/below 20 km Ever Attended School or not Bicycle Ownership Age Age Squared Gender: Male or Female headed household Landownership per household member (ha) Rainfall lagged (above/below avg) Constant Observations R-Squared Pseudo R-Squared Log-likelihood Sigma 1996 I OLS -.5052327** (.2561626) -.0569292 .7480934 -.1625307 .8205228 1.954393** (.8140271) .0256532 (.1348608) 6.23e-06 (.0013927) -.7368695 (.9055042) -.3287446** (.1556123) .0811483 (.7977804) 2.927465 (3.204579) 239 2005 Merged Data II III .7775339 -2.584663*** (.8724706) (.6676083) -.1638016 -.6500449*** (.2854106) (.2188208) -.4421384 -.4834681 (.9042622) (.6468324) .7536443 1.052062 (.8018895) (.6595603) -1.042457 .8537683 (.7490805) (.6352962) .3106117* .090577 (.1662419) (.1154838) -.0035605* -.0008955 (.0018373) (.0012066) -.1604823 -.4038906 (.8811694) (.7195722) -3.347386*** -.5739552*** (.9912937) (.165722) -.521276 -.8644657 (.8794178) (.6840597) 4.873355 5.700471** (3.872478) (2.751121) 74 313 0.0215 -453.8846 5.044257 (.3729538) 0.0734 -167.13868 2.920005 (.2804938) 1996 IV -.2917407** (.1375869) -.1118441 (.3914763) -.0618417 (.4277887) .947969 (.4160946) .0090789 (.0715026) .0000332 (.0007421) -.3741694 (.4719968) -.1315106** (.0642733) .0426217 (.4139572) 5.48182 (1.682369) 239 0.0777 2005 Merged Data V VI .7231276 -1.427685*** (.7850139) (.3958467) -.157148 -.4182965*** (.2566339) (.1340678) -.3229115 -.3426131 (.8134151) (.391872) .5687252 .6452391 (.7206231) .3976289 -.9058189 .3859534 (.6740846) .3786079 .2802116* .0532711 (.1476483) .0696594 -.0032352* -.0005885 (.0016262) .0007291 -.1303823 -.2295941 (.7977116) .4340611 -2.485589*** -.2229827*** (.7365622) (.0679279) -.4514179 -.4409437 (.7966478) .4074703 5.316521 6.971846*** (3.437076) (1.657885) 74 313 0.3017 0.1120 0.0314 -651.54129 4.854826 (.2869562) Notes: Obs. summary: 131 left-censored observations at lny<=2.910502; 182 uncensored observations; 0 right-censored observations. Sourceμ Authors‘ estimations. In table 7.31 we obtain the tobit MLE, where now log Poverty Gap is the dependent variable. In 1996 the estimated coefficients of log-distance to district road; distance to the district centre; bike ownership and landownership are statistically 282 significant at the 0.05 level albeit only with the latter having the expected sign. In the 2005 sample the coefficients of age; age squared and landownership were all statistically significant at the 10% level. To assess the impact of using the censored regression framework instead of treating the zeros like observations from the same data-generating process as the positives, in table 7.31 we compare the results with those from the OLS regression of LN(Poverty_Gap) on the regressors. All the OLS slope coefficients are in absolute terms smaller than those for the ML Tobit, but the OLS intercept is larger. The impact of censoring (zeros) on the OLS results depends on the proportion of censored observations, which in the case of the 1996 sample is 49.8% and in the 2005 sample is 20%. Two-limit tobit In less than 95% of the 1996 sample (13 observations) and the 2005 sample (3 observations) does the nominal Poverty Gap value exceeds respectively 12,156 and 58796. We estimate a two-limit tobit version of the tobit model to exclude these high values that contribute to the nonnormal kurtosis. This is done by choosing 12,156 (1996) and 58796 (2005) as the upper censoring points. We see that the impact of dropping 13 observations in 1996 is relatively small.268 The signs are exactly the same; the coefficients have almost the same magnitude; and the standard errors are also similar. Log-likelihood likewise goes up whereas Pseudo R2 goes slightly up. The same results are found when removing the 3 largest observations in the 2005 sample. Table 7.32: Tests of Normality and Homoskedasticity, 1996 & 2005 Test of Normality 1996 N R^2 p-value 2005 I II 190.18318 40.613189 0.000 0.000 Sourceμ Authors‘ estimations. Test of homoskedasticity Merged Data 1996 2005 Merged Data III 236.30433 0.000 IV 224.29202 0.000 V 67.542042 0.000 VI 287.54769 0.000 268 1996 Obs. summary: 120 left-censored observations at lny<=2.91; 108 uncensored observations; 13 right-censored observations at lny>=9.4056. 283 Test of normality To apply the LM test for normality, we need the likelihood scores. The components of the scores with respect to are  i times the relevant component of x, i.e.,   i xi. To execute the regression-based test of normality, we regress 1 on scores and  compute the N R2 statistics. The outcome of the tests on the 1996 and 2005 samples is a very strong rejection of the normality hypothesis, even though the expenditure (Poverty Gap) was transformed to logarithms. However, Skeels and Vella(1999) found that using the asymptotic distribution of this test produces severe size distortions, even in moderately large samples. This is an important limitation of the test (Cameron & Trivedi, 2009).269 Test of homoskedasticity Testing for the homoskedasticity outcome also leads to a strong rejection of the null hypothesis of homoskedasticity against the alternative that the variance is of the form specified as provided in table 7.32. Specification tests for the Tobit Model Once we move into models with censoring or selection, it is much less convenient to start with linearity according to Deaton(1997). It is therefore worth considering alternative possibilities, such as starting by specifying a suitable functional form for the regression function itself, rather than for the part of the model that would have been the regression function had we been dealing with a linear model. It is often possible to finesse the functional form issue altogether by recognizing that the functional form is as much an unknown as is the error distribution (ibid.). Two-part model in logs The diagnostic tests reveal weaknesses. The failure of normality and homoskedasticity assumption has serious consequences for censored-data regression that do not arise in the case of linear regression. Following Cameron & Trivedi(2009) we consider two approaches directed to arrive at a more general model. The two-part model specifies one model for the censoring mechanism and a second distinct model for the 269 Drukker(2002) developed a parametric bootstrap to correct the size distortion by using bootstrap critical values. His Monte Carlo results show that the test based on bootstrap critical values has reasonable power for samples larger than 500, which is not our case with sample sizes of only 241 and 74 observations. 284 outcome conditional on the outcome being observed.270 In contrast to the tobit model, which introduces only one equation, the two-part model can allow different covariates to enter the decision to use equation (i.e. determining whether the outcome is positive or zero) and level-of use equation (i.e. specifying a continuous distribution over strictly positive values) and does not impose restrictions on parameters across equations. This added flexibility is often argued to be beneficial for capturing key features of the censored variable (Koop et al., 2007). The tobit regression makes a strong assumption that the same probability mechanism generates both the zeros and the positives. According to Cameron & Trivedi(2009) many applications have shown that an alternative model, the two-part (hurdle) model can provide a better fit particularly in situations where the model is characterized by a large faction of zero responses (Koop et al., 2007). This is why we apply this modelling strategy in logs rather than in levels, although we acknowledge that a logarithmic transformation is not an appropriate method for dealing with bad data Table 7.33: Two-part model in logs, 1996 & 2005 Variable type DV dy = Poverty_Gap>0 Treatment I DV Logarithm of Distance to District Road Distance to District Centre above/below 20 km DV Ever Attended School or not DV Bicycle Ownership CV Age CV CV Age Squared Gender: Male or Female headed household Landownership per household member (ha) DV Rainfall lagged (above/below avg) CV DV 1996 Part 1: Probit Part 2: Linear Regression: dy Regression: lny Constant Log likelihood Number of obs LR chi2(9) Pseudo R2 F( , ) R-squared Adj R-squared Two-part model log likelihood Sourceμ Authors‘ estimations. -,1146305* (,0630952) ,0272094 (,1738777) -,0519368 (,1906864) ,5070368*** (,1839742) ,0124245 (,0318437) -,0000578 (,0003321) -,1795352 (,2109737) -,0615948* (,0346906) ,0937161 (,181448) -,4647567 (,7445344) -155,971 239 19,28 0,058 -,0916806 (,0754064) -,3268754 (,2379265) ,1093193 (,2626027) -,1886563 (,2695827) -,0270374 (,0412171) ,0002609 (,0004189) ,0125033 (,2883283) -,1089563** (,0548932) -,3072329 (,2637985) 9,844646*** (,9910259) 122 2005 Part 1: Probit Part 2: Linear Regression: dy Regression: lny ,4165187 ,3793139 (,5792985) (,2362141) -,1340065 (,1761066) ,2055733 (,6201662) ,5320191 (,4936833) -,5220814 (,4417997) ,1461157 (,0900539) -,0016907* (,0009924) ,1690494 (,5932811) -,9643701** (,4437748) -,5698837 (,6238048) -,9918816 (2,018256) -25,774 74 17,24 0,251 ,0338212 (,079257) -,3906851 (,243584) ,0009274 (,2234135) -,1839482 (,2119373) ,0474808 (,0486724) -,0005212 (,0005446) -,3014383 (,2365214) -1,411924*** (,4141103) ,174462 (,2357174) 9,980987 (1,133643) 61 0,82 0,062 -0,014 Merged Dataset Part 1: Probit Part 2: Linear Regression: dy Regression: lny .8406244*** 1,529387*** (.3116408) (,2914358) -,1099118* (,0564548) ,098163 (,1567566) ,0472788 (,1675651) ,2632193* (,1572838) ,0201892 (,0287992) -,0001595 (,0003015) -,0489699 (,181538) -,0823105** (,0353755) ,0435796 (,1608034) -,3914037 (,6728578) -197,547 313 29,8 0,070 -,0359855 (,0663172) -,1356878 (,2047647) ,0043496 (,2175769) -,3874924* (,2084745) ,0032051 (,0366589) -,0000655 (,0003796) ,0377429 (,2297357) -,1597634*** (,0554028) -,0079742 (,219172) 9,445541*** (,8743519) 2,58 0,341 0,209 -349,856 -88,275 270 183 5 0,225 0,180 -496,111 The sample-selection model, presented in the subsequent section, instead specifies a joint distribution for the censoring mechanism and outcome, and then finds the implied distribution conditional on the outcome observed (Cameron & Trivedi, 2009:538). 285 The first part is modelled through a probit regression.271 From table 7.33 we see that for the pooled sample, the probit regression indicates that log Distance to nearest district road; bike ownership; and landownership are all statistically significant determinants of the probability of positive Poverty Gap expenditure. The second part is a linear regression of lny (i.e. LN(Poverty Gap)), on the regressors. The coefficients of the regressors in the second part of the pooled sample as displayed in table 7.33 in four cases do not have the same sign as those in the first part. Moreover, the log distance to the nearest district road is no longer statistical significant in part 2. The joint likelihood for the two parts, -496.11, is the sum of two log likelihoods in pooled dataset (table 7.33). In comparison, the log likelihood for the tobit model is 651.54 (table 7.31). The two-part model fits the 2005 and pooled data considerably better, even if AIC or BIC is used to penalize the two-part model for its additional parameters. Table 7.34: Heteroskedasticiy & non-normality, 1996 & 2005 Breusch-Pagan / Cook-Weisberg test for heteroskedasticity Skewness/Kurtosis tests for Normality (var:rLNPov_Gap) chi2(1) Prob > chi2 1996 42.45 0.0000 0.458 2005 Merged Data 138.35 0.0000 0.000 27.07 0.0000 0.211 Sourceμ Authors‘ estimations. Pr(Skewness) Pr(Kurtosis) adj chi2(2) 0.097 . . Prob>chi2 . . 17.87 0.0001 . Finally, in table 7.34 we show the test as to whether the two-part model eliminates the problems of heteroskedasticy and non-normality. The tests unambiguously reject the homoskedasticity hypothesis in 1996 and 2005 and for the pooled dataset, as well as the normality hypothesis in 2005. However, unlike the tobit model, neither condition is necessary for consistency of the estimator. The key assumption needed is that E(ln y | d=1, x) is linear in x. On the other hand, it is known that the OLS estimate of the residual variance will be biased in the presence of heteroskedasticity. This deficiency will extend to those predictors of y that involve the residual variance (Cameron and Trivedi, 2009, Wooldridge, 2002, Greene, 2000). 271 One has the flexibility to change this to logit or cloglog. Comparing the results from the tobit, two-part, and selection models is a little easier if we use the probit form. 286 7.7. Conclusions and Policy Discussion Through our household survey data collection we have identified which areas (PSUs) benefited the most and which areas benefited the least from rural road transport infrastructure expenditures in Chipata and Lundazi districts as part of the overall EPFRP. It is not possible to make the desired inferences with the far from perfect data at hand, especially given the small sample sizes. Bearing in mind this caverat, we conclude thatthe survey data can still to some extent be used to assess the effectiveness of government infrastructure expenditure both through careful inspection and description of data (Deaton, 1997) from household surveys as a basis for discovering how people respond to changes in the economic environment in which they live. In practice road project selection often favours poor areas not only with poor road conditions, but likely with poor conditions of everything else. Thus, if poor road conditions coincides with other bottlenecks to local economic development such as low population density, poor ago-climatic endowments, mal-functioning markets, a lack of access to credit, low education and health care access, this following van de Valle(2008, 2009) begs the following question: Are rural roads going to have impacts in those circumstances? As a matter of fact we do find considerable heterogeneity in impacts across the 16 PSU communities. These impacts are highly context specific, and therefore it seems to be the case that the factors that explain the heterogeneity are not consistently the same across these PSUs. There are some combinations of attributes that are good for impacts and other combinations that lower the outcomes from the EPFRP. This in turn depends on how the attributes interacts with the road improvements, which seems to be key for understanding the impacts (van de Walle, 2008, 2009). Moreover, the paradoxical outperformance of the catchment areas by the control areas is most likely very context specific.This surprising finding further complicate policy prescriptions based on this far from perfect dataset. 287 At the same time, the 1996-2005 period analysed in this chapter is ideal for an exercise seen in the context of the trade liberalization reforms in Zambia (WTO, 1996a; chapter 7). The first LCMS I survey, conducted in 1996, namely provides a picture of the (pre-intervention) baseline situation in two rural districts four years after the initiation of an intense reform period simultaneously marked by continued drought. The year 2005 marks the fourth year after the completion of the EPFRP, that is, the post-project followup. Consequently, we have tried through a series of rigorous test of project impacts on the heterogeneous PSUs to address the change in the period after the end of the EPFRP to give an indication of the long-term sustained impact. Our two households surveys vary in finite sample size from less than a hundred observations in the EPRHS (88 observations) to a few hundred in the LCMS I (241 observations), which makes inference the hostage of sample size. This is why our quantitative analysis is complemented by qualitative information originating from a community survey based on participatory poverty monitoring and carried out simultaneously with the EPRHS. Further, the policymaker is interested not only in the effect on those who are poor, but also in the effect on those who are non-poor. The policy question being addressed in the chapter is therefore about the effect of rural transport infrastructure on average monthly p.a.e. consumption, averaged over poor and nonpoor rural households alike (Deaton, 1997; Grootaert, 1995). Our results, however, do not prove causality between the feeder road project and the change in poverty, but it does indicate a few things. First, the incidence in poverty in a country or region can change very dramatically over a relative short period of time. This suggests that there can be a great deal of mobility into poverty and possibly out of it as well as indicated by the figure 7.2, where the 2005 CDF curve is more steep than the 1996 CDF curve depicted in figure 7.1 above. Concerning the determinants of the standard of living amongst the surveyed communities through our qualitative analysis we find that only 42% of the PSUs had seen 288 their quality of life go up because of the impacts associated with the EPFRP. However, in 12 (63%) of the reporting PSUs the life quality situation in the PSU was considered better than before the implementation of the EPFRP. Amongst the 12 PSUs where the quality of life actually had gone up compared to the baseline situation 5(45%) of those communities considered that the major determinant was directly due to feeder road rehabilitation. Our regression results lend support to the latter qualitative finding by suggesting that the EPFRP treatment does have an impact on poverty; that distance to the market negatively affect poverty and that access to education had a significant poverty effect. Rural households‘ asset endowment is a significant determinant of poverty reduction. These large significant impacts of households‘ endowment of assets (e.g. bicycle) and education (i.e. school attendance), as reported in table 7.32 above, support the emphasis of expanding opportunities for basic education in Zambia in order to achieve the MDG2 as a basis for a sustained increase in Zambia‘s human capital base. The main policy issues are whether the government should be rehabilitating and routinely maintaining the feeder roads in sparsely populated rural areas characterized by generalized poverty and secondly, whether the government should introduce complementary policies? Further efforts to enhance quality of service delivery are likely to have a large payoff. Unless sound policies are complemented by provision of the public goods (e.g. feeder roads) needed for sustained consumption growth, the opportunities opened up by those policies may be utilised mainly by the more affluent (e.g. traders and transport companies, cf. chapter 8), and may exacerbate pre-existing inequalities between the traders based in the urban areas and the farmers based in the rural areas. Ensuring the equality of opportunity needed to avoid such an outcome is an important challenge according to Deininger and Okidi(2003) for both the central and local government. 289 Chapter 8: The Rural Transport Infrastructure and Marketing Linkage within the Context of the Sub-Regional Zambia-Malawi-Mozambique Growth Triangle 290 ―…Increasingly, infrastructure needs to match new demands as developing countries become more closely integrated into the global economy. Infrastructure…is front and centre in development.‖ World Bank, World Development Report, 1994, p:11f. 8.1. Introduction Escaping the poverty trap is highly unlikely without integration into a wider international economy. The lack of surplus resources for financing investment implies that external finance usually plays a critical role in generating the big push which is necessary in order for LDCs to move to a virtuous circle of economic growth and poverty reduction. But international trade is equally vital (UNCTAD, 2004). In fact, trade policy has been a dominant element in determining the pattern of Zambia‘s modern export-led economic growth due to the limited domestic market size and low domestic purchasing power. Due to Zambia's small domestic market, the GoRZ intents through the fifth national development plan (FNDP) 2006-2010 to use its trade policies to take advantage of privileged external market opportunities granted by developed countries and to promote the diversification of the economy. In addition to active participation in multilateral trade, regional markets also provide outlets for Zambian goods and services. In particular, the geographic proximity of regional markets makes them attractive export destinations for products from small and medium enterprises, as well as from smallholder farmers, of which the latter constitutes the most important category of the 12 socio economic groups in Zambia (GoRZ, 2006a).272 Notwithstanding, the Zambia-Malawi-Mozambique-Growth Triangle (ZMM-GT) countries poor performance in the international market place can be attributed to a number of factors, such as, the lack of sufficient high grade skills, the low level of technological know how, the low level of domestic savings, the inability of their economies to attract sufficient foreign capital, etc. In addition to these constraints however, there is empirical evidence to suggest that the poor state and inadequacy of the transport facilities and services in these countries impacted on the performance of their 272 The socio economic grouping was based on main current economic activity, occupation, employment status and sector of employment (GoZ, 2005a). 291 economies and contributed to their inability to compete effectively in foreign markets (UNECA, 2000). It is therefore appropriate, to assess the role and relevance of the transport sector in the ZMM-GT Initiative. In the urban context, the price and quality of transportation does affect decisions by firms regarding where to locate and to a large extent their productivity. Transportation costs are a major part of total costs, affecting especially small firms and the entry of new firms into an industry or market (UNECA, 2000). The geographical movements of firms, together with firm formation and expansion, decline and closure, influence the geographical distribution of economic activity at any point of time. Three main categories of factors influencing firm migration can be found in the literature: (i) internal factors; (ii) location factors (site and situation); (iii) external factors (Brouwer et al., 2002). One topic that needs more attention is the study of firm (i.e. agribusiness) mobility on the local scale. We aim to contribute to the firm relocation literature by exploring the effects of these factors (i.e. determinants of firm migration) on the decision to relocate to Eastern Province by employing data on firms‘ relocation behavior in the same two Eastern Province districts – Chipata and Lundazi – surveyed in chapter 6. This is done by using individual data on firm and (re-) location characteristics from a small sample of 50 firms drawn from our 2005 Agribusiness (Transport) Survey. We concentrate on only one of the firm demographic key events, that is, firm migration.273 The dynamism behind the expansion of regional trade in ZMM-GT and its concentration in certain agglomerations in Zambia‘s Eastern θrovince will also be explored by answering the following key questions: Did the improvement of the feeder roads from 1996 to 2001 reduce the marketing costs? Is the expansion of trade in Eastern Province the result of implementation of rural transport infrastructure development projects such as the EPFRP or vice-versa? The findings of a benchmark study carried out by Chiwele et al.,(1998) are compared with our 2005 Agribusiness survey results. 273 The five demographic key events of firms (birth, growth, shrink, relocation and death) are to be understood in relation to a multitude of factors internal and external to the firm (van Dijk and Pellenbarg, 1999). 292 This chapter doesn‘t capture how rural roads improvements may have affected the size of markets (i.e. volumes of trade). Nor does the chapter aim to sort out any road effects from trade and market policy effects. Consequently, the argument about the needed complementarity of policy and roads is not tested. Furthermore, the analysis of firm movements does not take account of firms that might have moved out of the area. Hence, since our 2005 survey only captured firms existing in the district centres at the time of the survey administration, the estimates of additional firms in the region must be considered as gross rather than a net estimate of the actual change in firm numbers. The following section provides a brief background on the motivations behind the ZMM-GT initiative. Section 8.3 presents our framework. Section 8.4 looks at the data by identifying the users of the EPFRP and presenting the most relevant cash crops in Chipata and Lundazi districts. The role of regional agreements is briefly examined in section 8.5 by respectively focusing on the impact of the SADC on the market participants and that of the ZMM-GT. Section 8.6 explores the nature of the feeder road constraint, and thereby explains the role of the EPFRP as an engine of regional trade. Section 8.7 concludes the study by providing some policy recommendations. 293 8.2. Background: Sub-Regional Zambia-Malawi-Mozambique Growth Triangle In theory, regional economic cooperation generates economic growth by making maximum use of the inherent but under-utilized economic potential in the proposed development area (ACP, 2009). Growth Triangles are transnational economic zones spread over geographically neighbouring areas, in which differences in factor endowments of three or more countries are exploited to promote external trade and investment for the mutual benefit of the concerned countries (UNECA, 2001b). ηf all the world‘s developing regions, East Asia has arguably made the most significant progress on deepening regional economic co-operation and integration since the mid-1990s. One of the most important work by the Asian Development Bank (ADB) under its Regional Co-operation Policy (RCP) introduced in 1994 was the promotion of various sub-regional „growth triangle‟ projects (Dent, 2008).The UNDP Resident Representative in Zambia, Ms. King-Akerele, through her experience in Asia had witnessed the successful growth triangles in South East Asia. It was because of that experience that she in 1999 decided to see whether growth triangles could be replicated successfully in Southern Africa. A ‗ψrainstorming Technical Workshop‘ resulted in the formation of a „Private Sector Forum‟ and a ‗Steering ωommittee‘ with the specific duty of ‗promoting‘ the Zambia-Malawi-Mozambique Growth Triangle (ZMM-GT) (UNECA, 2001a). Thus, it is against this background that Zambia, in close cooperation with its Eastern neighbours, Malawi and Mozambique and within the framework of the Preparatory Assistance of the Programme for Innovative Cooperation Among the South (PICAS) utilized Technical Cooperation Among Developing Countries (TCDC) as a vital complementary instrument for promoting sub-regional cooperation and economic integration (Muchanga, 2001).274 274 The Special Unit for TCDC facilitated the participation of the private sector from the three countries as well as the Asian Private Sector and Chambers of Commerce and Industry at the inception meeting of the ZMM-GT held in November, 2000 in Blantyre, Malawi at which the three governments had shown their commitments by signing a ‗letter of intent‘ and a ‗εemorandum of Understanding‘ to operationalise the ZMM-GT—signed by the end of 2001. The project entered into its implementation phase with the United 294 The strategic objective is to complement and further enhance national efforts at economic development, particularly of the somewhat marginalized and economically depressed areas that constitute much of the economic space of the proposed ZMM-GT. The area of coverage of the Growth Triangle comprises some 301,000 square kilometers and includes: Eastern and Northern Provinces of Zambia, the Central and Northern Regions of Malawi, and Tete Province in Mozambique (UNECA, 2001b).275 The results of apre-feasibility study identified the following areas of cooperation to be implemented in defined phases: Agriculture, agro-industries, transport, tourism, communications and information technology (Muchanga, 2001). The Concept and its adaptation in Africa, in essence, represents an opportunity to translate the "building block" approach of the African Economic Community (AEC) operationally on the ground (UNECA, 2001b). Nations Economic Commission for Africa (UNECA) being the executing agency and the Zambia Investment Centre (ZIC) being the implementing institution (UNECA, 2001a; Muchanga, 2001). 275 The project is also known as the Chinyanja Triangle project after the predominant language spoken in these three areas, Chinyanja (Patel, 2006). The areas/districts of the proposed ZMM-GT e.g. encompasses: Chipata, Chama, and Lundazi within Zambia. It also incorporates the Nacala transport corridors. 295 8.3. Framework In this section we explore the links between transport investment and economic development. This is done by first outlining the conceptual framework to test in a coherent and systematic way the hypothesis that transport infrastructure investments in the rural areas of Zambia‘s Eastern θrovince promote agricultural trade through operation decision-making by startups and relocation decision-making by existing agribusinesses.276 In addition to their productivity impact (chapter 4), rural transport infrastructure investments directly increase the connectivity between the district centres and/or the rural areas in the hinterland of the centres. Then in sub-section 8.3.2 we present the methodological framework within which the analysis of this chapter will take place. 8.3.1. Theoretical Effects The most fundamental outcome of an investment in transport infrastructure is the changes in the relative prices of accessibility of various locations. Since the network structure of transport systems makes accessibility spatially non-uniform, an investment in a new facility, or the improvement of an existing one through rehabilitation or routine maintenance, necessarily alters the present equilibrium structure of accessibility prices. This price change, in turn, implies changes in the relative advantage of spatially located activities and the economic opportunities both for the production and consumption sectors. The main reason being that the costs of inputs and the prices of outputs at alternative locations changes as a function of costs of accessibility to these locations. Furthermore, the extent and strength of various scale, scope and network economies, which affect the location decisions of firms, may become less pronounced as relative accessibility improves (Banister and Berechman, 2000). In response to improvements in accessibility from infrastructure investment firms can increase their demand for infrastructure facilities (i.e. the trip generation effect). They may also change their trip pattern (i.e. the trip distribution effect); their choice of travel 276 When using this generic term for the various businesses involved in food production, we exclude farmers and contract farmers covered in the previous chapters. 296 mode and their travel route (i.e. the modal split and trip assignment effects). They may relocate (i.e. the spatial location effect), or they may adopt all of these options. In turn, each of these effects will influence the degree of use (and the quality level) of transport which is in existence or which is being newly constructed (as well as accessibility) (Banister and Berechman, 2000). 8.3.1.1. Theoretical Effects of the network performance  Four principal determinants characterize the network performance. These are:  accessibility and travel flows;  network effectiveness; and  savings of vehicle operating costs; inter-modality (i.e. the interaction with other transport facilities and technologies). Beginning with the accessibility and travel flows determinant it is obvious that, given the particular infrastructure investment, the performance of the network is measured primarily by such factors as travel time savings between any pair of locations, by the resultant volume of traffic on each link and by changes in the relative accessibility of locations i and j (i, j = 1, ... , N). Similarly, users‟ savings in vehicle operating costs such as in fuel and maintenance costs also constitute key indicators of network performance (Banister and Berechman, 2000). 8.3.1.2. Location theory: Explaining the firm relocation process The land use transport links were explicitly included in Von Thunen‘s classic (1826) study on the impact that transport has had on patterns of agricultural development. This theory simplifies reality and promotes transport as the main determinant of land value and hence uses (Banister and Berechman, 2000). Christaller(1933) demonstrated the links between transport costs and the spatial distribution of economic activity. His central place theory proposed that the improvement of transport infrastructure strengthened the accessibility and dominance of the central city (Banister and Berechman, 2000). 297 The industrial location theory, which admittedly does not fit well with agriculture that has to cover a lot of ground, was formulated in the beginning of the 20th century.The theory focuses on the location factors determining the attractiveness of a site for firm location (pull factors). Relocation theory also takes into account the ‗push out‘ of the present location (push factors). Relocation approaches are treated as a special case of location theories or are based on empirical analysis (Brouwer et al., 2002). The neo-classical location theory focuses on the premise of the rationale firm that maximises profit in choosing the optimal location. The main forces driving firm relocation are transportation and labour costs. On the other hand, the behavioural location theory claims that the idea of ‗optimal‘ decisions, and minimising and maximising, is a theoretical abstraction. Simon (1959) replaces this picture of the firm with the firm as a learning, estimating, searching, information-processing organism. Decision-makers act without perfect knowledge and settle for sub-optimal outcomes. Location theory argues for the strengthening of the centre with a concentration of economic activity. Yet, much of the historical evidence suggested that there was substantial variation between different cities and a weakening of the influence of the city. Regional development policies also assumed that investment in transport would help alleviate depressed industrial regions and open up rural areas by increasing their share of economic activity. Even here, Hirschman‘s concentration arguments were being replaced by those (UNCTAD, 2004) promoting the spread of growth (i.e. balanced development) from the more prosperous regions (Banister and Berechman, 2000). Opposite these two approaches the institutional location theory views the location behaviour as the result of the outcome of a firm‟s negotiation with suppliers, governments, labour unions and other institutions about prices, wages, taxes, subsidies, infrastructure, and other key factors in the production process of the firm (Pellenbarg et al., 2002). Notwithstanding this criticism, most well known migration studies are 298 primarily based on behavioural principles (Brouwer et al., 2002, van Dijk and Pellenbarg, 2000; Assink and Groenendijk, 2009). Location theory is incomplete, and the empirical evidence about agro-industrial location appears paradoxical. Kilkenny and Coleman (2001) attempt to fill this gap in location theory in a way that reconciles the paradoxes. This is done by developing a new location theory for agro-industrial plants in two stages: macro-spatial then micro-spatial. They directly test the establishment location theory and show it‘s positive validity. Another paper by Mccann and Sheppard (2003) argues that the microeconomic foundations of industrial location theory must be reconsidered. In particular, the methodological basis of traditional industrial location models needs to be reconciled with recent models of clustering, the new economic geography literature, and also more aggregate systemic levels of analysis.A recent paper by Assink and Groenendijk(2009) argues that due to socio-economic developments like globalization spatial quality may be considered to be the dominant location factor of our time. The majority of the evidence to support the arguments comes from surveys of firms about their intentions to expand or relocate as a result of a new transport link (Banister and Berechman, 2000). This is also the approach that we will follow. 8.3.2. The 1996 and 2005 Agribusiness Survey Methodologies The aggregate view simplifies the more interesting actions of individual firms and households in their own decisions. Various other methods can be used to determine private sector responses to agricultural market liberalization and/or transport infrastructure improvements. Invariably, however, most researchers have employed survey methods that have identified traders and analysed the factors that stimulated their entry into the market. This survey approach is sufficient whereby a sample of traders can be framed and the study is aimed at identifying the characteristics of the traders, assessing the 299 constraints they face in taking advantage of the opportunities presented by either trade liberalisation and/or infrastructure accessibility improvements (Chiwele et al., 1998).277 We consider how the reductions in the cost – broadly defined – of movement has affected the economic activities of agribusinesses (excluding the farmers) in Chipata and Lundazi districts. To this effect, we will use the study by Chiwele et al.,(1998) as baseline, because it offers insights into the structure of the emerging private marketing system in Chipata and Lundazi districts in 1996 prior to the implementation of the EPFRP. 8.3.2.1. 1996 Baseline Survey Methodology The starting point of the Chiwele et al. baseline study was the recognition that little was known about the factors that influence the capacity and willingness of private traders in Zambia to rapidly and efficiently enter into the markets from which government parastatal agencies had withdrawn, especially in those remote areas that required government subsidies to induce cooperatives to market maize. Nor was it known whether private traders were able to undertake all the functions performed by state agencies such as input supply (fertiliser and seed), provision of credit and grain marketing. Furthermore, questions about the nature of the market networks that had emerged – e.g. the way various players in the market (producers, traders and millers) relate to each other for storage, finance and transport – remained unanswered. Hence, ωhiwele et al.‘s study objectives were to provide answers to these questions. The Chiwele et al.,(1998) study combines both questionnaire surveys and qualitative methods. The authors conducted fieldwork in Eastern Province selected as the surplus province and Lusaka Province selected as the deficit province. 277 There can, however, be other factors that play equally important roles in the development and evolution of marketing networks that survey methods do not capture. Such factors include according to Chiwele et al.(1998) the nature of the producer-trader relationship; the interaction between large-and small-scale traders as regards storage, finance and transport; and the ethnic, social and political affiliations of different actors. These affiliations may be important in explaining why the structure of incentives associated with liberalization favours those traders who are able to overcome various constraints. This type of New Institutional Economics (NIE) analysis is, however, beyond the scope this paper. 300 Two approaches were used to survey the traders. First, a list of traders was obtained from εinistry of Agriculture, Food and Fisheries‘ (MAFF) and traders operating in the survey areas were picked from the list.278 Due to the small number of active more established traders that could be found, it was decided to interview all the traders. A total of 25 were interviewed (16 in Lundazi and 9 in Chipata North, see table A3 in Annex 8). This sample was supplemented by interviews with any trader the group encountered, a strategy that enabled researchers to capture the activities of small traders. Where possible, group discussions were also held with traders, these totalled four (three in Lundazi and one in Chipata) (Chiwele et al., 1998; Chiwele, 2005). A list of 28 traders operating between Eastern Province and Lusaka was obtained from MAFF. Only ten of these could be identified and interviewed for reasons similar to those in Lundazi and Chipata North. In addition, 39 small traders were interviewed in the various markets of Lusaka Urban. In addition, two large millers were selected and interviewed. A questionnaire interview was conducted with all the traders, large and small (Chiwele et al., 1998). Chiwele et al. conducted their fieldwork from 3 August 1996 to 24 August 1996.279 There were four checklists targeted at key informants, farmers and traders. 8.3.2.2. 2005 Follow-up Survey Methodology In our overall quest to evaluate the long-term direct benefits of the EPFRP, in the 2005 follow-up study we investigate whether the accessibility improvements have translated into increased marketing activities within the zone of influence of the rehabilitated feeder roads. The long-term effect of the EPFRPis explored through a agribusiness questionnaire survey approach. 278 It was discovered that many of them had either not participated in marketing activities as they had originally intended or had stopped trading. 279 A pre-field trip was undertaken from 20 December 1995 to 8 January 1996. 301 Fifty questionnaire interviews were carried out from August to September 2005 in respectively ωhipata and δundazi‘s urban district centres, after we had listed all the relevant organizations in those two areas. The focus of the interview was on: The identification of the organization through a small number of background variables; the impact of the EPFRP; the vehicle operations costs and travel issues; socio-economic and stimulus issues; firm migration issues including decision to operate in, relocate to, expand activities in, and stay;280 travel and economic improvement issues; business environment issues; and finally spill-over effects. The master dataset contains 135 variables. For some cases: The Vehicle Operation Costs (VOC) (questions 11-13) and Total Present Values (PVs) (questions 19a-19b) information is not available due to the fact that insufficient number of ‗firms‘ answered these particular money value questions. Therefore, we chose to disregard these variables from the sample in order to keep the number of observations high enough to carry out a simple empirical regression analysis. 280 The ‗firm migration‘ has taken place somewhere in the past. 302 8.4. Identification of Geographical Movement of Firms due to the EPFRP The agribusiness migration event is analysed based upon the 2005 survey. A firm chooses a location from a number of alternatives. In doing so, it takes economic and/or non-economic factors, such as the MMD government reneging on its commitment to withdraw from maize and fertilizer marketing (McPherson, 2004), into account (De Bok, 2004). Mobility profiles Individual firms varies widely in terms of size, nature of its activity, region of origin and spatial relations. To account for these relations the mobility characteristics of firms can be used as dimensions for categorising all firms in groups with similar mobility characteristics (i.e. mobility profiles). This approach assumes that these mobility profiles are at least to some extent of influence on the location preferences of a firm (De Bok, 2004). The mobility profiles presented in table 8.1 are a result of an analysis on a number of mobility characteristics for all industry branches (stratum). Table 8.1 indicates that private companies mainly concentrate their activities on: Agricultural Trade / Marketing (74%); Agro-Industry (13%) and Agricultural Processing (9%). In order to carry out their daily travel and operations activities in Chipata and Lundazi districts they use primarily: A Medium-to Heavy Truck (87%); Passenger cars (personal vehicles) (74%); or a Pickup van (70%). Land rovers (35%) and motorbikes (35%) are also widely used by the private companies. 303 Table 8.1: Mobility profiles in Chipata and Lundazi Districts, 2005 Other Landrover Passenger Car (personal vehicle) Motorbike Pickups (van) small (mini-) bus Tractor Medium-to Heavy Truck Bus (Coach) 74% 4% 9% 100% 0% 0% 83% 0% 0% 0% 0% 83% 0% 100% 0% 0% 50% 0% Agro-Industry (food, beverages etc.) Agricultural Processing Agricultural / Rural Development Project Mobility Profile Private Company Individual Trader Input Supplier Miller NGO Other Mode of Transport for Daily Travel Activities and Operations Transport of Agricultural Produce / Inputs 1 2 3 4 5 6 Agricultural Trade / Marketing Main Activity 0% 0% 17% 0% 0% 0% 13% 0% 0% 17% 0% 0% 0% 0% 0% 0% 0% 50% 35% 20% 17% 50% 71% 50% 74% 60% 33% 50% 14% 0% 35% 40% 17% 0% 86% 0% 70% 60% 50% 50% 71% 100% 4% 20% 0% 17% 0% 0% 9% 20% 0% 0% 14% 0% 87% 80% 50% 100% 43% 100% 9% 20% 17% 0% 0% 0% Source: Authors‘ ωalculations. The „Individual traders‟ are exclusively involved in Agricultural Marketing activities. They mainly travel around in the districts by means of: Medium-to Heavy Trucks (80%); personal vehicles (60%); Pickup vans (60%). However, 20% of the individual traders use either land rovers; minibuses; tractors or coaches. The „Input suppliers‟ are involved exclusively in agricultural trade (83%) and transport of agricultural produce/inputs (17%). These two activities are carried out by means primarily of pickup vans (50%) and medium-to heavy trucks (50%). A third use passenger cars; and less than 20% use motorbikes (17%) or coaches (17%). The „Millers‟ are exclusively involved in agricultural processing (83%) and agroindustry (17%). They all possess medium to heavy trucks. And half of them possess land rovers; passenger cars; and pickup vans. Finally, the „NGOs‟ are exclusively engaged in Agricultural / Rural Development projects. They travel around mainly by means of the following modes: Motorbikes (86%); land rovers (71%); and pickup vans(71%). 304 Thus, from all these „mobility profiles‟ we expect a strong focus on road transport, given their high use of motorized vehicles. Moreover, the organisation of trade flows is determined in part by the type of product in question and the demand for it. There is, therefore, a big difference in trade participation as between food crops (maize and groundnuts) and cash crops (tobacco and cotton) (Chiwele et al., 1998). The food crop marketing channel comprises many actors undertaking particular or several roles which together form a complete marketing chain. In the following subsections the characteristics are highlighted of the major actors in the liberalized agricultural marketing system and of the main users of the feeder roads in Eastern Province, namely traders, millers and NGOs. We deliberately do not focus on farmers for the simple reason that “the African farmer largely inhabits a walking world” (Porter, 2002:285),281 although Chipata district is synonymous with the influx of bicycle traffic. The characterisation of these three actors is based partly on information obtained in 1996 by Chiwele et al., and partly from our own 2005 field visits. 8.4.1.1. Produce Traders Profile The 1996 field results and observations by Chiwele et al., revealed a significant level of participation by traders in maize and groundnut marketing whereas the market for cotton and tobacco was dominated by much fewer traders. In 1996 the cotton market was very concentrated and could easily be satisfied by a few large firms. The need to process and store cotton in specialised places (i.e. ginneries) meant that the trader who controlled these facilities had a competitive edge over those who did not. After the privatisation of the Lint Company of Zambia (LintCo) in 1995 through the sale of its ginneries to Lonrho Cotton and Clark Cotton, the facilities came to be controlled by these two firms (Chiwele et al., 1998). ―ηff-road‖ is defined by θorter(β00βμβ8η) as areas away from a good graveled or paved road which, for at least part of each year, are inaccessible or accessible only with difficulty by motorized transport. 281 305 During the late 1990s Lonrho was compelled to move to a Distributor System while improving the quality of assistance to farmers to overcome a period of severe credit default by farmers as competitors to the two dominant firms first emerged. Clark developed its own approach to this problem with some success. Yet these companies were competing in the export market with enormous subsidies from developed countries and, unlike West and Central Africa, were doing so until 2002 without any state support (Tschirley and Zulu, 2003).282 In 1999 Lonrho, citing input credit losses of US$2m left Zambia. Its assets were purchased by the private company Dunavant Zambia Limited, which possessed a worldwide network and a leading industry position.283 Dunavant (under same management) continued to develop the Distributor Scheme launched by Lonrho with a credit recovery over 60%. From 2000 to 2001 Dunavant fully developed its private Distributor Scheme thereby improving its credit recovery to 85%. In May 2006 the multi-national affiliate Cargill Cotton bought Clark Cotton with former parent company AFGRI citing low profit margins and insufficient global reach in marketing (Tschirley and Kabwe, 2007a). In the past decade many traditional cotton traders have taken a step back up the supply chain to become involved in ginning, the process of removing cotton lint from the seed. Cargill Cotton was one of the first cotton traders to invest in the ginning to augment its trading activities and strengthen its sourcing of quality cottons from Zambia and four other countries in Southern Africa including Malawi, Tanzania, South Africa, and Zimbabwe. The cotton is exported to South Africa, Europe and the Far East. Dunavant likewise has ginning operations in Uganda, Mozambique and in several locations in Zambia, including Katete, Petauke and Lundazi in Eastern Province. 282 WTO members on 19 November 2004 set up a body to focus on cotton. The agreement to create a body to focus specifically on cotton is part of WTη member governments‘ response to proposals from four African countries — Benin, Burkina Faso, Chad and Mali — to tackle the sector (WTO, 2010). 283 Dunavant S.A. is one of the leading companies merchandising cotton from all West and East African producing countries. 306 Tobacco is one of the biggest export earners in Zambia and is able to realise approximately $63 million annually. The Tobacco Association of Zambia (TAZ) is concerned with the increase in floors across the country which, if not controlled or monitored would render farmers vulnerable to merchants who may have misunderstood the free market system in Zambia. In 2008/2009 agricultural season the four major tobacco buying merchants in Eastern Province included: Sun Bridge, Africa Leaf, Pemba Leaf and Alliance One International, which bought the tobacco in four floors in Chipata and Chadiza districts.284 The largest volume of tobacco was bought by Alliance One.285 Marketing of food crops is unlikely to be as specialised as with these cash crops. Consequently, the marketing of food crops has attracted different categories of traders. Historical factors are also important in explaining the differences between food and nonfood crop marketing. In 1996 the market arrangement for maize was found to be in a state of continuing flux.286 The MMD government has always found some pretext to continue to intervene. The government established a food reserve and passed legislation for the formation of the Food Reserve Agency (FRA). The agency was expected to operate through market mechanisms to absorb or release food in times of surplus or shortage. It was not long, however, before the FRA became yet another disruptive force in agricultural marketing (McPherson, 2004; Ngoma, 2005). On the other hand, the situation in the groundnut market, where there was little government intervention before 1992, was much more stable because the impact of liberalisation on the old marketing arrangements appeared minimal (Chiwele et al., 1998). 284 In 2006, according to sale records from the Tobacco Association of Zambia (TAZ), six companies sold their tobacco at the Lusaka floors at an average price of 1.95 per kg. The companies included: TAZ, Zambia Leaf Tobacco(ZLT), Tombwe, Alliance One, Association Central and T L Brokers. 285 In 2005 Dimon Zambia and Standard Tobacco Company (stancom) merged. The merger, which saw the establishment of a new company called Alliance One International, followed a decision by the two companies' boards in November 2004 to merge their operations in an effort to reduce costs and enhance efficiency. 286 It was only during the 1994/95 season that grain marketing was completely liberalized. 307 8.4.1.2. Millers Profile The end-buyers of food crops are more diverse and range from individual households to food processing firms. Maize and groundnuts can be sold within communities to deficit households, at central markets in rural and urban centres, or directly to food processing firms which include well-established millers, hammer millers and confectionary factories. Unlike traders, millers had been in operation for between six and eight years already in 1996 (Chiwele et al., 1998). Apart from milling plants, millers also had their own transport, although this was supplemented by hired transport at peak times in the marketing season. Millers, therefore, had a good capacity to purchase, transport and store maize meal in large quantities. εillers didn‘t rely much on traders bringing maize grain to them but actively went out and purchased it through: Direct purchases from farmers; contract farming; the direct production of maize; and from December onwards, millers began to buy massively from traders who had stored the grain (Chiwele et al., 1998). In 2005 Tombwe Processing, a subsidiary of the TAZ, was processing Tobacco, all of which was sourced from the local buyers, among them Tombwe Leaf, Zambia Leaf Tobacco and Standard Commercial. The amount of processed crop represented the allocations for both export and the local market. In 2004 the GoRZ was considering putting up a tobacco processing factory in Chipata to stop taking it outside the country as a raw material. In October 2009 the mayor of Chipata disclosed that the Tobacco Board of Zambia (TBZ) had applied for a plot to establish a Tobacco processing plant and blending. The tobacco plant will make it easier for the crop vastly grown in Eastern Province to be quickly processed, blended and marketed to the local and international consumers. 308 8.4.1.3. NGOs Profile There is a large NGO sector supplying a variety of services to the poor (Hills, 2004). Moreover, the GoRZ and the NGOs continue to dominate the supply of seed to farmers in marginal areas through drought relief programs (Rusike et al., 1997). This is illustrated in the monthly Crop monitoring report dated 11th of June 2004. It revealed that of the rural household population of 49,563 in Chipata District 45% (22,500) benefited from the input distribution programmes, of which PAM-FSP targeted 10,000 rural households with seeds and fertilizers and GoRZ-FSP targeted 12,500 rural households with the same inputs, and Lutheran World Federation (LWF) provided seed maize. According to the same Crop monitoring report, there was a rural household population of 37,623 in Lundazi District of which 49% (18,565) benefited from the input distribution programmes. The Adventist Relief Agency - Emergency Drought Recovery Project (ADRA-EDRP) targeted 6,129 rural households with seeds (maize, groundnuts, beans, soya, rice), tuber and root crops, and fertilizers; The Church of Central Africa Presbyterian (CCAP) targeted 300 rural households with seeds; GRZ-FSP targeted 10,000 rural households with the maize seeds and fertilisers; LWF targeted 300 rural households with fertiliser, seed, tools and storage binds; PAM-FSP targeted 1,536 rural households with seeds (maize, groundnuts, beans, soya, rice) and fertilisers and finally Wildlife Conservation Society (WCS) provided fertilisers and seeds to 300 targeted rural households. Moreover, in Eastern Province the Smallholder Enterprise and Marketing Programme (SHEMP) was collaborating with the NGO – Africare – on how to link entrepreneurs to the market. In 2004 Africare was responsible for training 150 enterprise groups, which in the end would amount to 3,000 trained farmers doing business based upon real market demand. 309 8.5. The role of Regional Trade Agreements on the Market Participants This section considers a number of contextual issues that helped shape the regional setting in which the users of the EPFRP found themselves. Most of Zambia‘s trade is with non-COMESA countries(Hill, 2004). Zambia has preferential access to developed country markets under the Generalized System of Preferences (GSP), the African Caribbean and Pacific (ACP) countries and the European Union Co-operation Agreement, the Cotonou Agreement ratified in April 2002, the EBA initiative (to the EU) and AGOA (to the US) (table 8.2). The country was ranked 2nd by the Openness to Trade index and 1st for its Average Tariff Rate in the African Competitiveness Report 2000 (UNCTAD, 2006). Table 8.2: Overlapping Membership in Regional Integration Groups, 2009 SADC COMESA EAC SACU CMA ECCAS EU EPA EU EBA Zambia YES YES NO NO NO NO YES NO Malawi YES YES NO NO NO NO NO YES Mozambique YES NO NO NO NO NO YES NO Unit. Rep. Of Tanzania YES NO YES NO NO NO YES NO Dem. Rep. Of the Congo YES YES NO NO NO YES NO YES Angola YES YES NO NO NO YES NO NO Namibia YES NO NO YES YES NO YES NO Botswana YES NO NO YES NO NO YES NO Zimbabwe YES YES NO NO NO NO YES NO South Africa YES NO NO YES YES NO YES* NO * Its own free trade agreement (FTA) with the EU, the Trade, Development and Cooperation Agreement (TDCA). ** AGOA-eligible US AGOA** YES YES YES YES YES YES YES YES NO YES Source: Author. 8.5.1. The Southern African Development Community In 1992 while taking into account the Lagos Plan of Action and the Final Act of Lagos of April 1980, and the Treaty establishing the African Economic Community signed at Abuja, on the 3rd of June, 1991, it was decided to establish the Southern African Development Community (SADC) as successor to the Southern African Development Coordination Conference (SADCC) (SADC, 2001, 1992). The SADC is the only regional trade agreement (RTA), which Zambia, Malawi and Mozambique are all signatories to (Table 8.3). However, the viability of SADC as an economic community was enhanced by the accession to membership in 1994 of South 310 Africa, which generates around three-quarters of Southern Africa‘s combined GDθ (Dinka and Kennes, 2007). On the 24th of August 1996 the SADC Trade Protocol was signed. This was done in the recognition that the development of trade and investment is essential to the economic integration of the Community and that an integrated regional market will create new opportunities for a dynamic business sector (SADC, Art.2, 1996a; Goldstein, 2002). The implementation of the SADC FTA began in 2000 following the signing of the SADC Trade Protocol. The liberalization of tariffs has taken place at different rates by adopting an asymmetric approach (Metzger, 2008; Dinka and Kennes, 2007). However, most products that have some intraregional trade potential, such as consumer products (e.g. tobacco, foodstuffs, textile and clothing), have been declared import-sensitive and their trade liberalization have been post-poned (Kalenga, 2005 referred to in Metzger, 2008).287 None of the FTA objectives were achieved in the three ZMM-GT members during the implementation phase of the EPFRP from 1996 to 2001 nor at the time of our fieldwork in 2005. Hence, we can disregard the external regional forces of change derived from the SADC FTA in our trade analysis, and instead focus at the dynamics at the sub-regional level between the members of the ZMM-GT. 287 From January 2008, when SADC attained the status of a FTA, producers and consumers do not pay import tariffs on an estimated 85 per cent of all trade in Community goods (i.e. all ‗non sensitive‘ products) in the initial 12 countries. On the 17th of August 2008 the SADC FTA was officially launched, which was the achievement of the first important step towards deeper regional integration of the ambitious SADC programme for regional integration, which in 2010 should lead to a customs union, in 2012 trade will be fully liberalized, in 2015 a common market and finally in 2016 a monetary union, which will introduce a single common currency by 2018 (SADC, 2008; table 8.3). 311 Table 8.3: Major Regional Integration Initiatives Major Regional Integration Initiatives AEC Economic and monetary union Single Currency Trade in Free Trade Customs Market Union Area Union Goods Proposed for Proposed for proposed for Proposed for 2019 2019 2023 2028 Trade in Services Labour Mobility & Intra REC EstablishInvestment Migration4 Member States ment signed Rest of Africa (ROA) Export Import Export Import 51 0603-1991 8.5 10.8 5.6 11.0 Proposed 11 1018-1983 0.7 3.8 2.2 14.0 Proposed 19 0511-1993 8.7 11.1 8.6 17.2 14 0401-1980 / 0817-1992 19.9 33.1 2.3 2.6 5 0606-1967 / 3011-1999 12.6 18.7 7.2 9.9 Pillars for the AEC ECCAS COMESA Proposed for Proposed for Proposed for 2007 2011 2007 Proposed Proposed proposed for Proposed for 2008 2018 Proposed Proposed In force Proposed Proposed for Proposed for Proposed for Proposed for SADC 20083 EAC In force 2010 2015 2016 in force 0101- Proposed for proposed for 2005 2010 2010 Proposed Other Active RECs SACU In force in force De-facto in force 5 Notes: Not all e ers parti ipati g yet. Tele o u i atio s, tra sport a d e ergy – proposed. 3 Sensitive goods to be covered by 2012. 4 Free movement of peoples. 5. Based on the texts of the Cameroon-EU interim agreement and Interim agreements concluded between the EC and Cote d'Ivoire and Ghana 6. Algeria, Tunisia, Nigeria, Ghana, South Africa, West African Economic and Monetary Union, Common Market for Eastern and Southern Africa (COMESA). 7. in some cases can lead to a Free Trade Agreement (FTA). Source: Author. 8.5.2. The Sub-Regional Zambia-Malawi-Mozambique Growth Triangle One important element of the ZMM-GT development strategy is the complementary nature of ZMM-GT to the development process of both COMESA and SADC. This complementary nature is explained by the fact that the concept adopts a pragmatic, "bottom-up", flexible and market-oriented approach aimed at external markets, and thereby speeds-up and deepens the process of regional cooperation. The concept helps to realize both SADC and COMESA goals at the local level (UNECA, 2001b). The ZMM-GT development strategy is oriented primarily to the promotion of specialization and regionalization of economic activity to achieve increasing returns or economies of scale. The creation of an increasingly integrated ZMM-GT economy should enhance and hasten the development of export industries oriented beyond ZMM-GT (UNECA, 2001b) especially the Cotton and Tobacco industries. 312 Creating the right policy framework and initial infrastructure setting is essential for ZMM-GT. A key aim of the strategy is to empower the private sector.288 The initiative will also establish linkages and synergies with existing development and transport corridors such as the Nacala and Beira Corridors as well as the Spatial Development Initiatives (SDIs) (UNECA, 2001b). 8.5.3. From the Growth Triangle to the Nacala Development Corridor One of the main priorities of NEPAD in support of regional integration is to bridge the infrastructure gap. The development of regional infrastructure is seen to be critical for sustaining regional economic development and trade (NEPAD, 2002:42). The Nacala Development Corridor (NDC) was one of the first development corridors identified as being a regional development priority.289 The concept of the NDC has been derived jointly by the governments of Malawi and of Mozambique in order to exploit the significantly under-utilised natural resources present in the two countries. Geographically the Nacala Development Corridor covers the central and southern part of Malawi and extends eastwards into Mozambique to the coast at the sea port of Nacala. Subsequently, the GoRZ indicated its desire to join the initiative, when the idea of a Mchinji-Chipata railway was conceived in 1982 as part of a bilateral project between Zambia and Malawi. The Malawi section of the railroad was completed already in 1984, whereas Zambia did not actively pursue the project until 2006. As such the Eastern, Northern and even parts of the Central Province of Zambia through the future linkage of the Nacala sea port railway line to the Tazara Railway line should be regarded as part of the NDC in that they constitute an important hinterland to the NDC. At the Third Annual Poverty Review Conference in August 2004 the Permanent Secretary of Eastern Province mentioned with regards to commercial farming in Tobacco 288 Among the countries/institutions that have shown interest in the initiative include Senegal that has taken the lead in undertaking preliminary work for a Senegal-Mali-Guinea Growth Triangle; African Capacity Building Foundation (ACBF); African Development Bank; Japan; and the European Union Delegation in Ethiopia for possible replication of the concept in IGAD. 289 The NDC initiative is one of a number of regional development corridor initiatives that form part of the Spatial Development Initiatives Programme that is funded via the South African Ministry of Trade and Industry. Source: http://www.nacalacorridor.com/ 313 and Cotton that Eastern Province aspired to endeavour into moving higher up the value chain. To this effect, the implementation of the ZMM-GT, especially the completion of the remaining 44 km between Chipata and Mchinji in Malawi, which will be part of the NDC, would potentially have a tremendous boost on the economy by attracting more investors. However it was not until August 24, 2007, that the late Zambian President Levy Mwanawasa signed a deal with Malawi to extend the railway line to Chipata, which was expected to start operation in the first quarter of 2010.290 290 The first locomotive train that moved on the Chipata-Mchinji railway under construction for the past 27 years arrived in Chipata on 10 Dec. 2009. On August 27, 2010 the railway was officially commissioned. 314 8.6. Model/Empirical Strategy and Main Results In this section we seek to identify and measure the benefits derived from the EPFRP by the road users in the form of journeys, which are faster, more fuel efficient or more predictable in terms of journey time.291 The benefits of EPFRP are considered to be time and cost savings and any reduction in accidents for individuals and businesses. 8.6.1. Model/Empirical Strategy We develop a model for the explanation of firm migration in two districts of Zambia‘s Eastern θrovince. Our analysis of firm location can be carried out with a limited number of alternatives, facilitating data collection and processing, provided the choice process is described by the multinomial logit model (MNL) (McFadden, 1978). „Discrete choice models‟ attempt to explain a discrete choice or outcome. There are at least three basic types of discrete variables, and each generally requires a different statistical model. We focus on binary dummy variables that take on a value of one or zero depending on which of two possible results occur (Johnston and DiNardo, 1997b). Our model is empirically tested on micro survey data for agribusinesses operating in respectively Chipata and Lundazi urban district centres in 2005 (see section 8.3.2). We deal with the case when the dichotomous variable (relocation) is on the lefthand side of the relationship, i.e. when the dummy variable is an endogenous variable, which poses special problems.292 Thus, our approach develops a location choice model that first of all represents the firm as the decision making unit and secondly applies a level of spatial detail (District). To accurately describe the behavioural context of the location decision an approach has been applied in which an individual firm makes a decision to relocate and chooses a unique location from a limited set of feasible alternatives. Furthermore it is acknowledged that firms varies hugely in their mobility 291 Improvements in transportation performances, such as the improved safety, connectivity, reliability, or speed stemming from transportation investments are measured and categorized as user benefits. 292 The linear probability model (LPM) approach is generally not appropriate when the dependent variable is binary. A major weakness of the LPM is that it does not constrain the predicted value to lie between 0 or 1 (Johnston & DiNardo, 1997:415, 417). 315 characteristics and hence have a large variety in location preferences with regard to infrastructure related location characteristics. To account for this heterogeneity in preference, the concept of mobility profiles will be used to address the mobility characteristics of agribusinesses following De Bok‟s(2004) approach. Model The theoretical model is based on a behavioural approach describing the spatial decision making of individual firms in a disaggregated physical environment.293 At a certain point in time this firm has various characteristics that determine their preferences, such as its size (Stratum_b) or its mobility profile (Table 8.1). The firm is located at a unique location in the physical environment. The location alternatives are described by the location characteristics, such as location type (district) and accessibility. The migration behaviour of an individual firm within this physical environment is regarded as choice process that consists of a sequence of considerations and decisions as visualised in figure 8.1 below. First of all the decision to move (relocate) is a result of the relative satisfaction at the current location within both the ZMM-GT and SADC. The factors influencing this satisfaction are referred to as push factors (section 8.5). These push factors influence the propensity of a firm to move. The decision to relocate is mainly determined by firm internal factors relating to the lifecycle of firms and to a lesser extent by site related factors (Louw 1996, van Dijk and Pellenbarg, 2000). The tendency to migrate furthermore shows large differences between sectors (Stratum). The approach consists of two decisions and the formation of a choice set. The actual decision for an alternative location in a choice set Pj|Ci(2) is a conditional choice probability which will be modelled using a spatial preference model in the form of a MNL. The joint decision Pij of firm i to move and to relocate to location j is the product of the probability Pi(1) firm i will move and the conditional choice probability Pj|Ci(2) of firm i choosing location j from a subset of alternatives Ci: 293 The behavioural approach, founded by Simon (1955), assumes satisfying behaviour and is based on the idea of decision making in a context with limited information and bounded rationality. Pred (1969) was one of the first to introduce the behavioural approach in location theory (De Bok, 2004). 316 P = Pi(1) x Pj|Ci(2) (6.1) The influence of accessibility can be of importance in each decision, but from the literature it seems that accessibility mainly expresses itself as a pull factor rather than a push factor. The focus is therefore on the estimation of the choice model for moving firms (De Bok, 2004:8). Figure 8.1: Conceptual model of an individual firm in a physical environment Source: De Bok, 2004:7. Location choice The location decision (operate relocate) of each individual moving firm i is modelled with the MNL, based on random utility theory (McFadden, 1974, 1978). This implies we assume that firm i attaches some utility Uij to each alternative j in a set of alternative locations that are considered. Furthermore, the firm will choose the alternative that yields the highest expected utility. In the presented model we assume a linear additive utility function of the form: 317 (6.2) U = Vij + ij = 0ij + 1ix1ij +…+ nixnij + ij, where Uij is the expected utility of location j and Vij is the deterministic part of the expected utility of location j. The deterministic component of utility is specified as a function of alternative specific variables (x1j,…,xnj) multiplied by estimable coefficients (β1i,…, βni). And because we apply a labeled experiment an alternative specific constant βoij is added to the utility function for the business environment of the location. If the random unobserved component of utility εij is assumed Gumbel distributed, it can be eliminated from the utility function (McFadden, 1974). The resulting MNL describes the probability Pij|Ci(2) of choosing location j from a unique subset Ci with k alternative locations: Pij(¦2C)  e Vij e k C Vik (McFadden, 1978:6; De Bok, 2004:8). Analysis of categorical data Moreover, there are a number of statistical procedures which can be used for our analysis of the 2005 categorical data, also known as data on the nominal scale, e.g:294  Categorical distribution, general model.  Binomial regression.  Generalized linear models (McFadden, 1978:13; cf. chapter 4). 8.6.2. The Impact of the EPFRP on the Market Participants An important change during the 1990s was the growth of nontraditional exports, including agricultural products, processed foods and textiles (WTO, 1996; McPherson, 2004). Moreover, SADC intraregional exports increased fourfold between 1990 and 1995, and almost doubled between 2000 and 2005 (IMF, 2006). The latter increase has 294 Nominal scales are mere codes assigned to objects as labels, they are not measurements. The only kind of measure of central tendency that remains invariant under one-one transformations is the mode. 318 been attributed to accelerated trade liberalization within SADC. However, the share of intraregional exports in total exports has risen only slightly since 2000 (AfDB, 2004). Since many affiliates of South African firms, including Shoprite located in Chipata town, use most of their inputs from South Africa, the increase in intraregional trade in SADC is partly linked to South African FDI in the region benefiting from bilateral trade agreements, which, in principle, could lay the basis for the development of regional production networks (Metzger, 2008; Visser and Hartzenberg, 2004). 8.6.2.1. 1996 Baseline Findings A preliminary investigation by the MAFF indicated that there had been a better response by private traders in Eastern Province to liberalisation than elsewhere in the country following the period between late 1991 and August 1996, when Zambia fundamentally changed its trade and economic policy (Chiwele et al., 1998; Hill, 2004). By 1996 the trade regime was considerably liberalized and there was substantial decentralization and deregulation in other spheres of economic activity (WTO, 1996). Rusike et al.‘s (1997) observations about the poor rural network match the striking features of all the sites visited by Chiwele et al. in 1996. All the blocks had motorized feeder roads, which were passable during the dry season but were difficult to use during the rainy season. The situation appeared to deteriorate as one moved away from the centre of the districts. Thus, transportation appeared to be a major problem in 1996 for many traders, some of whom went to the area and bought grain and then waited for up to one week for transport to move their purchases (Chiwele et al., 1998).295 It was, however, observed that those areas with relatively well serviced roads tended to have private transporters who moved farmers‘ inputs and products and charge prices that many farmers were able to afford. 295 The vast majority of motorized vehicles in rural areas are based in settlements along the paved roads and there is widespread reluctance among vehicle owners to take their vehicles on unpaved roads unless the rewards are high (Porter, 2002). 319 Chiwele et al. suggest that it seemed very likely that poorly maintained roads and the absence of storage facilities meant that the new private sector-led agricultural marketing system marginalized rather than integrated farmers located in remote areas. It was a generally shared view that the government had withdrawn too rapidly from agricultural marketing (Chiwele et al., 1998). In all the blocks visited by Chiwele et al. in 1996, the major concern of the farmers was that private traders were not always coming to their areas to purchase grain. This problem arose from the poor state of the roads. A characteristics of the small and medium traders found by Chiwele et al., was their relative youth. The modal age for small traders was 15-25 years while that of medium traders was 26-35 years. Medium-scale traders were on average slightly better educated than small-scale traders. More than 80 per cent of both the medium- and small-scale traders started trading after the 1992 liberalisation of the market was initiated. In fact, 34 per cent of the traders went into maize marketing for the first time in the 1995/96 marketing season (Chiwele et al., 1998). From the responses to their questionnaire, the major factors determining choice of market seemed according to Chiwele et al. to be the following:     the rate of turnover, the nonexistence of other markets, lack of transport and the proximity of the market. About 60 per cent of traders entered into contracts with suppliers whereas the other 40 per cent didn‘t. The most common contracts included agreements on price (12 per cent), quantity to be supplied (4 per cent), both price and quantity (44 per cent) and price and credit arrangements in the case of contract farming (4 per cent). 320 In Eastern Province, large-scale trading was dominated by Asian-Zambians based in Chipata, which has a large Asian population. Most of the large-scale traders went into crop marketing after 1992 when the market was liberalised. The responses indicated that large-scale traders were largely businessmen who engaged in other business activity besides trade in agricultural commodities because of the need to diversify their business activities. All three categories of traders trade in the same crops. It is apparent that ready access to transport was the most important constraint. Only one trading firm, which had always operated a transport company, had its own transport. All the other traders relied on hired transport. Transport charges were normally based on per kg (per ton) load, distance or hours spent in transit or a combination of these. The price per ton was higher for remote areas. Close to 50 per cent of traders were located within 10 km of their main point of sale (Chiwele et al., 1998). From ωhiwele et al.‘s baseline findings, it can be concluded that transport development in 1996 could be regarded as a constraint on the attainment of economic opportunities in the rural areas of Eastern Province. Hence, the rural transport infrastructure development can be regarded as a necessary condition for achieving the ZMM-GT growth potential. 321 8.6.2.2. 2005 Follow-up Survey Findings The 2005 cross-sectional Agribusiness Survey is based on a standardised questionnaire administered to the potential users of the Eastern Province Feeder Road Network based in Chipata and Lundazi district respectively. The questionnaire was addressed to the most informed person about and/or in charge of day to day operations of the agribusiness. Firms of all scales, that is, small-scale (less than 5 employees); mediumscale (between 5-20 employees) and large- scale (more than 20 employees) are analysed here. In total our sample only contains 50 observations. Descriptive Statistics with Categorical Variables We restrict ourselves to the variables which were available for as large a part of our sample as possible. This implies that we only make a limited use of the Vehicle Operation Costs (VOC) (questions 11-13) and Total Present Values (PVs) (questions 19a-19b) variables. Descriptive statistics of the variables can be found in Annex 8. The dependent variable From the dataset we only know the actual behaviour of firms for the year 2005. However, we also have information about the past (the baseline account in sub-section 8.6.2.1) and present location characteristics of the agribusiness. It is believed that this information can be used to partially explain the past migration. If the organization moved in previous years the present location is the result. We focus on the stated preference of the firms with regard to migration. Firms were asked to indicate whether „the EPFRP had a sufficient effect on the feeder road network to induce the organisation‟s decision to „relocate to‟ (relocate) (question β8.β) or to ‗operate in‘ (operate) (question β8.1)ν to ‗expand activities in‘ (expand) (question β8.γ) or to ‗stay‖ (retention) (question 28.4) in Chipata/Lundazi districts that may not have been considered otherwise? The respondent could choose from the following two categories: 0 = Yes or 1 = No. This implies that the respondent 322 expresses a preference with a nominal scale (also denoted as discrete) variable.296With these two categories the dependent variable y = Migration can take values 0 and 1. For the analysis of this type of dependent (categorical) variable a number of statistical procedures can be used. The binomial regression model is a suitable tool for our analysis (see section 8.6.1). The distribution of the propensity to migrate is shown in Table 8.4. From this it is clear that almost half of the firms in the small sample started operations due to the EPFRP. 24 per cent relocated due to the EPFRP. Moreover, 87 per cent expanded their activities, and 45 per cent decided to stay influenced by the EPFRP. Table 8.4: Frequency of the propensity to migrate (Migration) Y=Migration Yes = 0 Operate No = 1 Yes = 0 Relocate No = 1 Yes = 0 Expand No = 1 Yes = 0 Retention No = 1 Frequency 22 23 10 32 42 6 19 23 Percent 48,89 51,11 23,81 76,19 87,5 12,5 45,24 54,76 Cumulative Percent 48,89 100 23,81 100 87,5 100 45,24 100 Obs (Total) 45 42 48 42 Sourceμ Authors‘ ωalculation. The explanatory variables The goal of our chapter is to find the explanatory variables that determine the stated probability of firm migration. According to economic theory a firm will move if the benefits of moving to another location exceed the costs in a certain period of time. Although the assumptions and goals (optimising versus satisfying) of both the neoclassical and the behavioural approach differs substantially, we follow van Dijk and Pellenbarg‟s eclectic approach for the operationalization of the empirical model and the interpretation of the results. It implies that both variables reflecting the costs of moving as well as variables that reflect the benefits should be taken into account. In each stage of the complicated decision process with regard to a change of location another set of variables can be the most important factor (figure 8.1). In our approach we do not account for the different stages in the decision process, but we only 296 At the Nominal scale, i.e., for a Nominal category, one uses codes assigned to objects as labels. 323 look at the outcome of this process: The stated preference to move to another location. In principal there are three categories of explanatory variables: Firm internal factors, Location factors (site and situation), Firm external factors (van Dijk and Pellenbarg, 2000). Firm internal factors297 In this category we use the following variables: Economic sector (Stratum), and firm size (i.e. number of employees) (Stratum_b). The data contain information about the economic sector from the agribusiness identification. We distinguished only six sectors related to rural development: Private Company; Individual Trader; Input Supplier; Miller; NGO; and Other (figure 8.2). Figure 8.2: Distribution of firms in the 2005 Agribusiness Survey dataset Frequency Total 49 Other 2 NGO 7 Miller 6 Input Supplier 6 Frequency Individual Trader 5 Private Company 23 0 10 20 30 40 50 60 Sourceμ Authors‘ ωalculation. With regard to the size of the firm we distinguished three scales: Small-scale, Medium-scale, and Large-scale. We use the number of employees as indicator for firm size. We expect that the costs of moving and the organisational problems for small firms are much less than for large firms. 297 E.g. quality of management, organisational goals, ownership structure, growth rate of turnover, employment and profits (van Dijk and Pellenbarg, 2000:7). 324 Location factors (site and situation)298 This type of variables is most important for the location decision process.299 Because our limited dataset doesn‘t permit this, we will not pay attention to this type of variables. Firm external factors300 With regard to these factors we will take into account differences in economic performance and the regional labour market situation by means of a set of district dummies and the opinion of firms about (local) government policy. The dataset does contain information about the location (District) in one of the two Eastern Province Districts – Chipata or Lundazi. Concerning the other important external factor - „local government policy„ (localgov) - we expect that firms may be stimulated to move to another location when „local government policy‟ creates attractive locations to move to. In the questionnaire firms are asked to rank on a scale from 1 to 5μ ‗Adequate transportation” (Transportation); ―utilities” (Utilities); “workforce skills” (Skills); ―Adoption of local government policies that motivate response to transit improvements” (Localgov); and “Tax structure” (Taxstructure) as site location requirements that affect Chipata/Lundazi district business costs, markets and overall competitiveness for attracting business investments. Thus, the explanatory variables are the following: Stratum_b (number of employees), Main activity (consumer market), Stratum (type of organization), District (location). The data in table 8.5 indicate that 33% of the private companies relocated and as much as half of the input suppliers relocated and therefore they are considered the most mobile, whereas 14% of the NGOs relocated to one of the districts. On the other hand, the individual traders and the millers were local organizations. 60% of the small sample of 298 E.g. absolute and relative characteristics of the location site, e.g. lot size and size of possible expansion space; distance to customers and suppliers (van Dijk and Pellenbarg, 2000:7). 299 Type of area; Type of enterprise zone/industrial site; Infrastructural facilities; Ownership of the building; and Opinion about the present location. 300 E.g. government policy, regional economic structure, etc. (van Dijk and Pellenbarg, 2000:7). 325 individual traders (5) started operation due to the EPFRP. This percentage was equivalent for the millers (5). Relocation was not common in Chipata with only 15% of the firms deciding to relocate due to the EFPRP, on the other hand, notwithstanding the small sample size (8), more than 60% of the firms relocated to Lundazi. Table 8.5: Firm relocation by stratum, 2005 Strata 1 2 3 4 5 6 Total 33 35 Total Mobility Profile Private Company Individual Trader Input Supplier Miller NGO Other Total District Chipata Lundazi Total Number of observations 21 4 4 5 7 1 42 Relocated firms (%) 33% 0% 50% 0% 14% 0% 24% Number of observations 21 5 6 5 6 2 45 Start up firms (%) 43% 60% 50% 60% 33% 100% 49% 34 8 42 15% 63% 24% 34 11 45 53% 36% 49% Sourceμ Authors‘ ωalculation. Hypothesis Testing One sample t-test We wish to test whether “the average percentage decrease in the cost of doing business in Chipata and/or Lundazi” (pctbusicost) differs significantly from 0. The mean of the dependent variable pctbusicost for this particular sample of organizations based in Eastern Province is 44.16,301 which is statistically significantly different from the test value of 0. We conclude that this group of small-scale(19), medium-scale(12) and large-scale(16) organisations have a significantly higher mean on the percentage decrease in the cost of doing business than the two organizations that hadn‘t experienced any decrease. One sample median test - Wilcoxon signed-rank test We will also test whether the median “percentage decrease in the cost of doing business in Chipata and/or Lundazi” (pctbusicost) differs significantly from 0. The results indicate that the median of the dependent variable pctbusicost for the sample is statistically significantly different from 0. 301 The organisations are distributed between the following strata: 1) private companies; 2) individual traders; 3) Input Suppliers; 4) Millers; 5) NGOs; and 6) other kind of organizations. 326 Two independent samples t-test We want to compare the means of a normally distributed interval dependent variable for two independent groups (Chipata and Lundazi districts) by using independent samples t-test. The results indicate that there is a statistically significant difference between the mean “percentage decrease in the cost of doing business in Chipata and/or Lundazi” (pctbusicost) for Chipata and Lundazi (t = -1.5597, p = 0.0631). In other words, Lundazi has a statistically significantly higher mean “percentage decrease in the cost of doing business” (52.69) than Chipata (40.69). The Wilcoxon-Mann-Whitney test is a non-parametric analog to the independent samples t-test and can be used if we do not assume that the dependent variable (pctbusicost) is a normally distributed interval variable (we only assume that the variable is at least ordinal). The results suggest that there is a non-statistically significant difference between the underlying distributions of the pctbusicost of Lundazi and the pctbusicost of Chipata (z = -1.635, p = 0.1020). However, we can determine which district has the higher rank by looking at the how the actual rank sums compare to the expected rank sums under the null hypothesis. The sum of Lundazi(35) ranks was higher while the sum of the Chipata(33) ranks was lower. A one sample Binomial test We wish to test whether the proportion of „goods and service prices purchased by respondent‟s organisation‟ (goodsservices) differs significantly from 50%, i.e., from 0.5. The results indicate that there is no statistically significant difference Pr(k <= 22 or k >= 24) = 0.883. The same applies for other measures of „seasonal changes in transport quality during the rainy season‟ including „Decreased social travel‟; „Decline in school attendance‟; and „Decline in health visits.‟302 On the other hand, the result for „lost productivity of (business) activities‟ (lostproduc~y) shows that there is a statistical significant difference Pr(k <= 9 or k >= 39) = 0.000015 (two-sided test) in the sense that the proportion of lostproductivity does significantly differ from the hypothesized value of 50%. 302 Question 17 in our Agribusiness Survey Questionnaire. 327 A Logit model We will now estimate an explanatory model for the stated preference of the firms with regard to migration (i.e. Question 28).303 Firms were asked to indicate the decision toμ ‗operate in‘ν ‗relocate to‘ν ‗expand activities in‘ and/or ‗stay (retention)‘ in Chipata/Lundazi as a discrete category. With these two categories the dependent variable y (migration) can take the values 0 (=Yes) and 1 (=No). The decision to relocate activities is modeled within a logit model relating the probability to relocate to a set of explanatory variables Xi. The probability of relocation is F(X‘i ), where F is the logistic distribution, which yields the logit model:304 (6.4) and Prob(yi = 1) = F(Xi ) = exp(Xi )/[1+exp(Xi )], is the vector of coefficients. The choice of F returns a value between 0 and 1. The formulation of the model ensures that the predicted probabilities lie between 0 and 1 (Brouwer et al., 2002, Johnston and DiNardo, 1997b). In the discussion about the results we will use three conventional levels of significance: if t > 1.66, 1.96 or 2.33 the coefficients are significant at, respectively, the 10%, 5% or 1% level. Most of the categories (Stratum and Main activity) have a variable that represent ‗other‘. Although these variables are not discussed in the chapter, they are taken into the analysis to avoid missing cases. The tables 8.6a-b present the empirical results. The estimated parameters for DISTRICT show that the organization‘s location in either Chipata or Lundazi district hasn‘t any probability of affecting the organization‘s decision to ‗relocate‘ to that district. 303 The estimation of the Logit Model has been done with STATA version 9.2. For details about the estimation procedure and the interpretation of the results see Johnston & DiNardo(1997, chap.13). 304 A natural choice of a function F that translates X into a number between 0 and 1 in a sensible way is a distribution function, or the cumulative density. In fact, binary response models can be defined this way (Johnston & DiNardo, 1997:418). 328 Table 8.6a: Empirical results Table 8.6b: Empirical results relocate Chipata Private_Co~y NGO Small_scale Large_scale relocate District Coef. Operate Coef. -1.214284* District .2598594 (.4939683) (.3719428) Stratum .3004319 Stratum -.0601546 (.2957639) (.202225) stratumb .3421194 stratumb .5461258 (.5084106) (.3558211) mainacti .1215341 mainacti -.263613 (.3678684) (.2497303) _cons 40.30256 _cons -9.03512 (16.20487) (12.49195) Number of obs. 41 Number of obs. 44 LR chi2(4) 8.97 LR chi2(4) 4.49 Prob > chi2 0.0619 Prob > chi2 0.3432 Log likelihood -18.293651 Log likelihood -28.205705 Pseudo R2 0.1968 Pseudo R2 0.0738 Sourceμ Authors‘ ωalculations. Agr_Trade Agr_Industry _cons Coef. -37.33617 Operate Lundazi Coef. -.5192594 (.9377194) -19.2523*** Private_Co~y .4600117 (1.265691) (1.26952) -44.61702 Individual~r .6240815 (1.35689) -17.8858*** Miller 18.26497*** (1.42214) (2.34219) -1.229367 Small_scale -18.79169*** (1.474231) (.9512165) 10.68342*** Medium_scale -18.25015 (1.265691) 29.24257 Large_scale -19.97427*** (1.071763) 35.16297 Agr_Trade -.4304713 (2.00524) (1.437127) Agr_Projects -.2171959 (2.273725) Agr_Proces~g -19.06374 _cons Number of obs. 29 LR chi2(7) 19.20 Prob > chi2 0.0076 Log likelihood -9.0801226 Pseudo R2 0.5139 Number of obs. LR chi2(10) Prob > chi2 Log likelihood Pseudo R2 38.88029 (5.53137) 42 11.21 0.3418 -23.318896 0.1937 The covariate variable STRATUM shows that private companies are more willing to ‗relocate‘ than e.g. NGOs due to the EPFRP. On the other hand, Millers decision to ‗operate‘ in ωhipata/δundazi was more influenced by the EθFRθ than the other types of organizations, which all had small estimated parameter values. The parameters for the variable Stratum_b were expected to give an indication that a firm with a 'larger' market has higher chance of relocating. The results are quite different. The probability of Small-scale organizations deciding to ‗relocate‘ was significant unlike that of large and medium-scale firms. ηn the other hand, ‗small-scale‘ and ‗large-scale‘ organizations probability of deciding to ‗operate‘ in either of the two districts were more or less equal with similar estimated magnitude of parameter values. 329 Finally, the variable Mainacti presents six activities: Agricultural marketing (1); Agricultural/Rural Development projects (2); Agricultural Processing (3); Transport of Agricultural Product (4); Agro-Industry (5) and other (6). The estimates show that firms involved in Agricultural marketing have the highest probability of moving. Nevertheless, in 2004, there were still just a few buyers of which most were located in Chipata. Because of the poor road network only a few could afford to send trucks to pick up commodities. Hence, the MoAF was still of the opinion that the private buyers hadn‘t filled the gap left by the Marketing Board and was therefore still buying the staple crop maize, through its Food Reserve Agency (FRA) and through provision of cheap fertilizers and seed support. However, the GoRZ had only put one buying point in Chipata to cover the entire district (bearing in mind that there is 180 km the furthest point from the district centre). The buyers of Tobacco and Cotton hired their own transport some of which they even got from South Africa. They sent smaller trucks deep into the village stations depending on the nature of the feeder roads. 330 8.6.3. Robustness Checks by Comparing Models Next we will discuss the estimation results for the various specifications presented in table 8.6. above. Table 8.7: Comparing Full with Reduced Logistic regression Models Relocation Decision Predictor Variable Name Variable LOG1 LOG2 LOG3 LOG4 LOG5 LOG6 LOG7 Categorical Type of Organization Stratum 1.41 Categorical Size of Organization (Scales) stratumb 1.04 dichotomous (33=Chipata; 35=Lundazi) District 0.32*** continuous pctage decrease of cost of doing business pctbusicost 0.97 continuous pctage increase in output pctoutputi~r 1.02 continuous VOCs in past 12 months VOC12months 1.00 continuous Adequate transportation transporta~n 1.03 Categorical Main Economic Activity mainacti Constant Constant 1.51 2.87 1.05e+17*** 11.54** 1.31 6.01** 2.83 Number of Obs 42 41 42 37 37 23 42 Log likelihood -22 -22,8 -19,49 -19,27 -20 -11,31 -23,1 df_m 1.00 1.00 1.00 1.00 1.00 1.00 1.00 chi2 2.04 0.01 7.13 2.51 1.13 5.65 0.01 LR chi2(3) 7.16 8.96 2.07 (ii) (ii) (ii) 8.94 Likelihood-ratio test Prob > chi2 0.0670 0.0299 0.5581 0.0302 LOG8 LOG9 0.36* 0.99 LOG10(iii) 1.35 1.41 0.30** 1.02 1.13 1.31 5.23e+15** 3.19e+17** 37 37 41 -20 -17,29 -18,29 1.00 2.00 4.00 1.13 6.47 8.97 (ii) Notes: (i) df(unrestricted) = df(restricted) = 2; (ii) observations differ: 41 vs. 36; (iii) Full Model. Sourceμ Authors‘ estimations. In table 8.7 we compare a number of different logistics models consisting of either a single continuous predictor or a single dichotomous predictor with the full model (LOG10) used in table 8.11a above (i.e. the basis for comparison).305 In column LOG9 we have a model with two variables in it, both a dichotomous and a continuous predictor. We ask if the „full model (LOG10)‟ is "better" than the models LOG1-LOG3 with just one of the variables in it. To do this, we carry out a likelihood ratio test, which tests the null hypothesis that the coefficients of the variable(s) left out of the reduced model is/are simultaneously equal to 0. In other words, the null hypothesis for this test is that removing the variable(s) has no effect; it does not lead to a poorer-fitting model (Chen et al., 2010). Thus we compare the full model to reduced models with respectively: ‗type of organization‘ (categorical variable)ν ‗size of organization‘ (categorical variable)ν ‗district‘ 305 Given the difficulty in interpreting the results of models with two categorical predictors with an interaction, and models with continuous and categorical predictors with an interaction we omitted them (table A3 Annex 8). 331 (binary variable) and find that only for the first two independent variables is the chisquare statistic statistically significant. This means that the variable that was removed to produce the reduced model resulted in a model that has a significantly poorer fit, and therefore the variable should be included in the model. In other words, it seems that the full model is preferable.306 Logistic regression uses a maximum likelihood to get the estimates of the coefficients. Many of desirable properties of maximum likelihood are found as the sample size increases. The behavior of maximum likelihood with small sample sizes is not well understood. According to Long(1997), 100 is a minimum sample size, and you want at least 10 observations per predictor (Chen et al., 2010). Since our full model only consists of categorical predictors, it would have been desirable to have more observations to avoid computational difficulties caused by empty cells. 306 A test of nested models assumes that each model is run on the same sample, in other words, exactly the same observations. The likelihood ratio test is not valid otherwise. 332 8.7. Conclusions and Policy Implications The main arguments for liberalization rest upon the ineffectiveness and inefficiency of state service provision (Dorward et al., 2004b, Dorward et al., 2004c). The policy agenda addressing these problems has focused on the intrinsic problems of state failure. Thus, from 1991 to 1996 GoRZ moved decisively away from import substitution to outward orientation as the base of its growth strategy. The trade regime was considerably liberalized and tariffs became the main instrument of its trade policy (WTO, 1996). From 1996 to 2002 the economic reforms consolidated the substantial liberalization efforts made during the 1990s with an emphasis on poverty reduction (WTO, 2002). In general, the reforms created an environment conducive to economic growth. 2003 marked a turnaround with above 5% growth rates (OECD et al., 2010). The GoRZ expected that by the end of 1996 the new private-sector led marketing system initiated in 1992 would be able to stand on its own. However, the study by Chiwele et al.,(1998) showed that the new grain marketing system still was in transition and remained undeveloped at the end of 1996. The nature of the infrastructure and the road network connecting the surplus remote areas and the deficit regions in 1996 was one of the key constraints on the attainment of economic opportunities in the rural areas. A key observation made by Chiwele et al.,(1998) was that rather than stimulate production through seasonal and regional price variations, market liberalisation tended to marginalise farmers in remote areas. It was also found that trade flows and marketing channels depend on the type of commodity. Finally, ωhiwele et al.‘s examination of the key players in the emerging marketing channels showed that most had few facilities that enabled them to carry out their functions effectively. These findings lead us to conclude that an essential component in our understanding of the links between transport and economic integration must be the microeconomic conceptualization of the processes at work in the location decisions of firms (Banister and Berechman, 2000). By employing our own 2005 survey data on agribusinesses‘ relocation behaviour in Chipata and Lundazi districts, it was found that primarily small private companies were 333 more likely to have moved into these two districts as an outcome of the EPFRP. In addition, it was mainly the ‗small-scale private companies‘ engaged in ‗agricultural marketing‘ which had made a positive relocation decision. However, its is important to emphasis that results only captured road users and firms that existed in Chipata and Lundazi districts at the time our survey was being administered in 2005. Evidently, this implies that any firms that had failed or moved out of these two districts of Eastern Province between 1997/1998 and 2005 were not being captured. This means that the growth estimate of the number of (relocating) firms are in gross rather than a net estimate of the actual change in firm numbers. Nevertheless, it is anticipated that with the completion of the Chipata-Mchinji rail line project in 2010, the NDC rail line will to a much larger extent change the economic face of Eastern Province, by boosting both the intra-regional trade among the three members of the ZMM-GT as well as getting export crops to the international markets at a much lower price through the port of Nacala. Notwithstanding, from our research on the local moves we have nevertheless showed that a lot can be learned about:  the basic causes of firm relocation, and  the course of the inherent decision process. In almost all cases, firms consider a local move before eventually deciding upon a move over greater distances e.g. through the NDC. Furthermore, firm relocations over short distances can be very important in order to facilitate adjustment processes in the local economy (Pellenbarg et al., 2002, van Dijk and Pellenbarg, 2000). Policy Implications One view is to argue that lack of success is not the result of the liberalization agenda, but a failure to implement it thoroughly (World Bank, 2000). Chiwele et al.,(1998) on the contrary argue that a more rational approach would have been to introduce liberalization after macroeconomic stabilization had been achieved, while mounting an aggressive rural road rehabilitation programme to open up remote areas to 334 the market. In 2004 one of many outstanding issues was still to put mechanisms in place to ensure utilization of fuel levy for the development of the feeder road network. The efforts of promoting and implementing regional integration using the innovative private sector driven “Growth Triangle approach” should therefore with the completion of the NDC rail line in 2010 seriously be enhanced and reinvigorated with a view to contributing to the acceleration of the implementation of the Abuja Treaty.307 307 At the EAC-SADC-COMESA Summit on Wednesday October 22, 2008 a historic agreement to create an Africa Free Trade Zone (AFTZ) consisting of 26 out of 53 countries was announced. Ultimately, the AFTZ is considered a major step in the implementation of the Africa Economic Community (AEC). 335 Part Three Conclusions and Policy Implications 336 Chapter 9: The Modus Operandi of the InvestmentTrade Nexus 337 "To end poverty, build a road" A Chinese saying. 9.1. Introduction Analysis of the factors at the micro-level underlying the rural growth performance in Zambia‘s Eastern θrovince can help to better understand rural growth and poverty reduction in an overall Sub-Saharan African context, and thereby produce policy-relevant insights that go beyond what is known from the traditional neo-classical and endogenous cross-country growth literature at the macro-level. By providing a better understanding of the challenges ahead, the thesis also aims to help formulate policies that might address these challenges. The application of project analysis to new productive investments in developing countries found a theoretical base in an area of development economics, which specifically debates over the choice of technology (Curry and Weiss, 1993). Casley and Kumar(1987) and Keddeman(1998) describe the general practice adopted by the Panel on Monitoring and Evaluation of the U.N. ACC Task Force on Rural Development as follows: Ex post evaluations are designed as in-depth studies of the impact of an intervention and usually are done five to ten years after the completion of its funded implementation.308 Van de Walle(2009) suggests that it is useful to start by thinking about the nature of roads and the ways in which they may differ from other policy investments. To justify public sector investments in feeder roads, it is useful to show that they also have productive effects. Project analysis can be used, therefore, not simply to analyse the effects of a project, but to justify it in relation to the alternative available (Curry and Weiss, 1993, Roemer and Stern, 1975). Three main methods are commonly used to rank Rural Transport Infrastructure (RTI) investments: (a) Multi-criteria analysis (MCA); (b) Cost-effectiveness analysis (CEA); and (c) cost-benefit analysis (CBA) (Van de Walle, 2002, Lebo and Schelling, 2005).309 However, because traditional CBA approaches do not account for many of the benefits of RTI investments, extending the framework of CBA as we have done in this thesis holds promise for improved analysis. In short, appraisal needs to be holistic in nature, in the sense that it needs 308 The reasons for conducting ex post evaluations are two-fold. First, much of the lasting impact will not be visible at the time the project comes to an end. Second, such impact as is detected at the time of the terminal evaluation might prove transitory. 309 CBA methods for appraising investment in the road infrastructure sector were first developed for roads in more urbanized, high-traffic density areas, drawing on methods from a developed country literature on road appraisal (see Banister & Berechman, 2000). 338 to cover economic, social and (environmental) impacts of the project in a coherent and consistent manner (Institute for Transport Studies University of Leeds, 2003). For a project such as the EPFRP that brought about significant changes in levels of accessibility in five of Eastern θrovince‘s eight rural districts, and that broke new ground through rehabilitation of low volume rural feeder roads, there is a case for examining whether additional measurable economic benefits exist to those measured within the transport market (chapter 8). In order to gain a better explanation of the key research question and hypotheses of this thesis we used a conceptual and methodological framework provided by Banister and Berechman(2000) as illustrated in figure 1.2 (chapter 1), which captures the methodological issues discussed in the core chapters 4-5 and 7-8, while facilitating the measurement of the impact of RTI investment in a coherent and systematic way. Chapter 9 provides a discussion of the lessons learned from the thesis. Section 9.1 discusses the lessons that have been learned about the methods of measuring the impact of roads compared with other similar studies. Section 9.2 compares the empirical findings on the impact of the EPFRP with the results of other studies. Section 9.3 presents some of the most important transport policy decisions in Zambia in the previous 40 years. Section 9.4 draws a number of policy implications based upon the empirical findings and the status quo of Zambia‘s transport policy. Section 9.5 finally provides some future research directions. 339 9.2. Methods of Measuring the Impact of Rural Feeder Roads Van de Walle(2009) proposes that the distinguishing features of rural roads suggest a number of researchable evaluation questions, which have important implications for evaluation design and methods, as well as for data requirements. Thus, an important objective of the thesis is too a large extent to test and assess the validity and rigour of various methods of measuring the wider medium- to long-term socio-economic impacts of rural roads interventions supported by a range of actors such as the Zambian Government (Ministry of Local Government and Housing (MoLGH) and the District Councils) and foreign donors (UNDP, UNCDF and ILO). These methods can be applied to retrospective project evaluations to explain the success or the failure as well as to help provide guidance on how to undertake such studies. Prior to carrying out our field work in 2005 there was not much documented evidence of the extent of these changes in Zambia‘s Eastern θrovince and the mechanisms that brought them about besides a few studies by researchers affiliated to the University of Zambia (UNZA), IFPRI and the Zambia Food Security Research Project (FSRP). Nor had there been any systematic study of the extent to which feeder road access permitted or facilitated these changes. Fortunately, a significant amount of CSO data exists on the socio-economic status and agricultural production of households in the Eastern Province. Supplementing these secondary datasets with our own primary follow-up survey data presented an opportunity to measure changes over the period from 1996 to 2005 using a triangular research design. 9.2.1. Generic issues in assessing the impacts of rural roads The prospective benefits of rural roads are derived and conditional, which constitutes a distinguishing characteristic of roads. One only obtain utility indirectly from a road through its accessibility to opportunities for extra consumption. The nature and extent of impacts are also likely to be heavily dependent on interactions e.g. with other social and physical infrastructure, and the geographical, community and household characteristics of where they are located. For a new road to enhance mobility, people must have access to private intermediate (e.g. bicycles, motorcycles) or public (e.g. minibuses, taxis) means of transport along the road (Ali-Nejadfard, 2000; Banjo, 2005; Barwell, 1996). Van de Walle(2009) argues that for impacts to be felt on input and output prices there must be some means of freight transport and local production for which external demand exists. In the absence of such complementary factors, new or improved roads may simply facilitate and hasten population out migration. In evaluating impacts it is clearly important to control for 340 conditioning factors that can be expected to interact with a road intervention and their heterogeneity across road projects (Van de Walle, 2009). Roads are clearly not randomly placed, and it is likely that the factors that matter to road placement will also affect outcomes. Roads are built (or rehabilitated) in certain locations and not in others, for reasons that tend to have a lot to do with the attributes of those locations. Van de Walle(2009) provides the example that policy makers might assign a new road to a specific region because it is deemed to have economic potential or because it has a strong political constituency. In other words, Van de Walle (2009) warns that there is a clear risk of selection bias ─ also known as the problem of ‗endogenous road placement‘ (see chapter 3). Understanding the potential sources of endogeneity is critical to collecting the right data and choosing the right methodology for estimating unbiased impacts. Thirdly, rural roads and road networks are widely expected to have geographically dispersed effects on numerous outcome variables. Through various linkages, an improved rural road could conceivably have community-wide impacts on livelihoods and socioeconomic development and impacts that extend beyond the community the rural road links up to. Van de Walle(2009) lists a number of implications of this: First, she believes that traditional evaluation methods that use a single cross-section cannot be relied upon. ηbservable local area attributes ─ that would typically be considered as potential controls for programme placement and used to ensure conditional exogeneity in a single difference model ─ may have themselves been determined or contaminated by the road investment project. This leads Van de Walle(2009) to be pessimistic about conducting a worthwhile rural road evaluation without adequate baseline (pre-intervention) data. The fact that roads may have dispersed effects also has implications for how one defines the road‟s zone of influence. The EPFRP intervention passes through a community (i.e. SEA) which is identified as the zone of influence for data collection purposes. If there are externalities from the EPFRP to a neighboring community, the evaluation would miss these impacts as they are outside the defined zone of influence. It also has implications for identifying appropriate comparison areas that are close enough to the intervention areas ─ and hence similar in various agro-climactic, soil, altitude and other respects, which was the case for the 16 PSUs analysed in chapter 7 ─ and yet sufficiently distant to avoid being contaminated by spillover effects as was the case by the three remote control districts compared in chapters 4-6. 341 Van de Walle(2009) through an extensive literature review also remarks that little is known about the distributional impacts of rural road investments. There are likely to be both vertical impacts (between ‗rich‘ and ‗poor‘) and horizontal impacts (at a given level of preintervention welfare). On average, benefits may well be positive but she argues that it is essential to understand who the losers are if one is to understand distributional impacts, and differences in impacts at given levels of living (as is attempted in chapter 6). Those with greater initial land, education, wealth or influence will be better able to take advantage of the changes (chapter 2). The distribution of current income and future income earning opportunities may widen. In short, Van de Walle(2009) emphasises that it is important to adopt a sampling design and collect data that allows one to explore whether and how impacts vary across groups. Van de Walle(2009) in her extensive account of the major generic issues in assessing the impacts of rural roads finds that little is known in the literature about the time it takes for the full welfare impacts of an improved road to emerge, how quickly intermediate impacts emerge and how short term impacts on these outcomes may differ from longer term impacts. In contrast to interventions with relatively rapid impacts (such as transfer programmes), the welfare impacts of rural road infrastructure are generally expected to take some time to appear. This creates problems for an impact evaluation. Van de Walle(2009) further warns that an evaluation that doesn‘t allow sufficient time for the linkages to play themselves out could vastly under- or over-estimate the net impacts on living standards. However, she argues that the more time one allows for impacts to emerge, the more the evaluation needs to contend with potential attrition in the data, confounding exogenous shocks and spillover effects such as related or unrelated other investment programmes in the control or treatment areas (chapter 7) that make it harder to assess the impacts of the original road project. 9.2.2. Evaluation Methodology There has been an increasing interest in undertaking detailed evaluations and assessments of the impact of rural roads investments on growth and poverty reduction mainly as a result of the strong political interest in solid documentation of results and what works and what doesn‘t (cf. the Paris Declaration on aid effectiveness and value for money). However, such assessments of projects funded by foreign donors‘ and/or from the host Government‘s own budget have significant methodological challenges in terms of attribution of impacts 342 from different interventions which may either have overlapping objectives and modes of implementation, or have different timelines and at best partial baselines. Randomized control group trials (RCTs) are often promoted as the only scientifically rigorous evaluation method. However, van de Walle(2009) raise the question: What are the alternatives to RCTs? She surveys the problems and discusses some practical implementation issues related specifically to conducting an impact evaluation of a rural roads project that is assigned to some geographic areas but not to others. She observes that very few of the many aid-financed rural road projects in developing countries have been subject to evaluations. According to Estache(2010) the reason is simple: they are simply hard to do using (quasi-) randomized evaluation techniques. Estache(2010) emphasises that the most challenging characteristic of road projects in terms of the techniques approximating random trials is that they have no natural comparison group. It is indeed hard to find two similar regions in all the relevant characteristics. In addition, he also suggests that evaluators also have a hard time addressing all relevant spillover effects as well as time dimensions. This is why it is still common to see assessments of the impact of rural roads interventions conducted through general equilibrium modeling. Instead, the impact evaluation analysis discussed and applied in this thesis is largely atheoretical and reduced form. However, as argued by van de Walle(2009) it often remains unclear how exactly roads have their impacts. To some extent, the careful choice of „outcome‟ variables can help identify the channels of impact, by focusing on intermediate outcomes. However, there are clear limits to how far this is possible in practice. Alternative approaches to ours identify impacts on the basis of economic assumptions about how the world works. These approaches rely on a structural model of behaviour where clear economic assumptions propel the specification. Van de Walle(2009) proposes that this helps make up for important missing data. The downside is that the assumptions on which the model is built may not be evidently plausible or empirically testable. The upside is the gain in terms of what we can learn, as long as the assumptions are valid. Structural methods can use either partial or general equilibrium models. There are a number of examples of partial equilibrium work estimating the impacts of reduced transport costs through road provision (chapter 2). A good example is a paper by Fan et al.,(2004), 343 which develops and adapts a simultanesous equations model to estimate the effects of government expenditures on agricultural production and on rural poverty through different channels. The model consists of four equations, which give the formal structure of the system. Ideally, they should also include a set of equations to model the relationship between government expenditures and improved public capital such as roads and education as Fan et al.,(2000 and 2002) have done in their studies in Asia. However, Fan et al.,(2004) due to data limitation in Uganda used a different approach. They first estimate growth and poverty impacts of physical infrastructure, health and education. They then use estimates of the unit cost of public expenditure to obtain benefit-cost ratios of various types of government spending. Public capital not only affects poverty through agricultural productivity but also through wages and employment. The unit of analysis in the Fan et al.(2004) study is a combination of national, regional and district level. Due to lack of systematic secondary data at the district level, they generated a panel dataset at the district level for 1992, 1995 and 1999 by directly aggregating survey data at household and community levels in these years (chapters 4-5). They use a double-log funcational forms for all equations in the system. They have 90 observations (3 years and 30 districts). The system is estimated using the full information maximum likelihood technique. Their results show that most government investments, such as agricultural services, rural infrastructure, rural education, and health, have contributed to agricultural productivity growth and reduced rural poverty. However, variations in their marginal effects on production and poverty reduction were large, among different types of spending and across regions. They find that government spending on rural roads had substantial marginal impact on rural poverty reduction mainly through improved agricultural productivity. Van de Walle(2009) draws some broader lessons from the structural model studies. First, some road impacts may be difficult to model and estimate in structural form. Second, road infrastructure may have unexpected impacts not factored into structural models. This suggests that classic reduced form approaches such as ex-post impact evaluations used in this thesis can help provide insights to structural models, and vice versa. The two approaches should be seen as complements according to van de Walle(2009). While there have been a number of recent attempts at assessing rural road impacts using impact evaluation methods (chapter 2), Van de Walle(2009) highlights that research has 344 newly underlined the enormous difficulties inherent in estimating the magnitude of the effects attributable to infrastructure. Problems arise due to the endogeneity of much infrastructural development and the many other factors that are at work (Jalan and Ravallion, 1998). Van de Walle(2009) warns that not allowing for initial conditions ─ both time-invariant and time-varying ─ and the way in which infrastructure is allocated to specific regions will tend to bias impact estimates, often downwards in poor areas. Some additional common criticisms of past impact evaluations of roads are that the results were not likely to be robust to unobserved factors influencing both programme placement, subsequent growth, and outcomes; and that they did not follow projects long enough to capture full impacts. Van de Walle(2009) proposes that doing better requires a combination of better evaluation design, methods and data. The first logical step of our analysis leads us to study first at the regional level (in chapters 4-6) and then at the below-district level (in chapters 78) the alleged successes of the EPFRP. In chapter 4 we carried out a study to measure the effects of the EPFRP on agricultural production and productivity. We implemented respectively matching estimators for average treatment effects; differences-to-differences estimators as well as censored regression models to control for potential selectivity biases arising from the systematic location of roads. However, there is insufficient model structure to generate insights about how rural roads impact on input use, because the focus is mainly on the resultant cotton yield effect. In chapters 5 we estimated the impact of rural roads improvements on p.a.e. expenditures using a range of methods, including a pseudo-panel approach. In chapter 7 we similarly estimated the impact of rural roads improvements on p.a.e. expenditures e.g. by using a Tobin approach. Our sample design does not have a sufficiently large sample size of households as to guarantee a minimum statistical representativeness at the sub-district level. Instead we followed the approach outlined in Deaton(1997), which recognizes that structural modeling is unlikely to give convincing and clean answers to our key policy question. Evaluating the medium-to long-term changes of the EPFRP also relies on more qualitative methods of data collection and analysis. Chapter 7 ensures that qualitative methods used have a strong conceptual basis, which is important in order that robust and credible results emerge. 345 In chapter 8 we estimate the impact of rural roads improvements on firm relocation decisions by using our own 2005 cross-sectional survey, which only captured firms existing in the area at that time. 9.3. The Empirical Findings Despite a general consensus on the importance of rural roads for development and living standards, van de Walle(2009) finds that there is surprisingly little hard evidence on the size and nature of their benefits, or their distributional impacts. Indeed, there have been relatively few rigorous and credible impact evaluations of rural roads. In recent years a number of studies have assessed the impacts of rural roads rigorously using impact evaluation methods that expressly deal with selection (chapter 2). These studies using various techniques show mixed results, some finding substantial impacts and others more muted impacts while examining disparate outcome variables. Thus, Van de Walle(2009) argues that it remains difficult to draw definitive conclusions concerning the impacts of rural roads. Despite an exhaustive applied micro-econometric analysis using reduced-form approaches, our Thesis unfortunately only manages to extract a relative few number of quantitative results about the impact of roads in our study region compared to those listed by van de Walle(2009) and what for example a structural model approach could have achieved. Determinants of economic growth and to a lesser degree, poverty reduction have been explored in a large cross-country literature. It is quite clear that these macroeconometric growth studies reviewed in chapter 2 are inconclusive and unable to provide details about the transmission mechanisms – the nexus – between access to hard and soft infrastructure, rural growth, and poverty. On the other hand many of the microeconometric studies show a positive correlation between transport infrastructure investment and production, productivity as well as poverty alleviation, although some doubt the benefits to the poor (table 9.1). This is why in order to overcome the limitations of the aggregate cross-country approach and to gain additional insights into these behavioural relationships we exclusively have resorted to a micro-level approach. Cross-sections of existing survey households were used for chapters 4-6 and novel survey data was collected through questionnaire based interviews for chapters 7-8. 346 In the short-to medium term, the reaction on the demand side is confined to travel variables such as travel volumes or choice of routes. In the longer term, this reaction must also be manifested in location decisions of the agribusinesses surveyed.310 In chapter 8 we find that it was mainly the ‗small-scale private companies‘ engaged in ‗agricultural marketing‘ which had made a positive relocation decision associated with the EθFRθ. However, this approach is not adequate for measuring the structural changes brought about by the EPFRP. So, in chapter 4 we find at the district level that the improved accessibility from the EPFRP led to changes in land allocation and in yields to the Eastern θrovince‘s most important cash crop – cotton (Table 9.1). In chapter 5 we find that, although the mean cotton sales share of household income has more than doubled despite the fact that the mean distance to the input market remained unchanged from 1998 to 2004, the estimation results are only small and statistically insignificant in terms of gains to mean consumption (Table 9.1). In chapter 6 we account for the fact that the distributional effects of rural road investments have not been addressed extensively in the rural road literature (Khandker et al., 2006). We find that the rural road rehabilitation investments are pro-poor. However, the very unequal expenditure distribution in 1998 had not improved much in 2004 (Table 9.1). In Chapter 7 through a qualitative analysis we find that between 1996 and 2005 access to external markets through the rehabilitated feeder road network, is a critical determinant of the rural households‘ ability to increase their income and improve their chances of breaking out of the poverty trap in the medium to long run. It is, however, difficult quantitatively to identify this impact. Moreover, the statistical significance of the association is not robust to different choices of price deflators and specifications tests for the Tobit Model (Table 9.1). In sum the thesis‘ contribution to the literature comes from its attempt to provide a better understanding of the different identified transmission mechanisms. These findings presented in table 9.1 lead to useful policy insights for improving the effectiveness of government investments designed to promote growth and reduce poverty through increased Since most of the land in Eastern Province is customary land the reaction can‘t be measured in changes in land and property prices. 310 347 agricultural trade in SSA in general and in Zambia in particular. The policy implications are discussed in the next two sections 9.4and 9.5. Table 9.1: Summary of Empirical Findings Chapter 2 Analytical Level Macro & Micro 4 Meso (regional) 5 Meso (regional) 6 Meso (regional) 7 Micro (sub-district) 8 Micro (sub-district) Method Literature Review Average treatment effects; Differencestodifferences estimators Parametric and Semiparametric Regression Models Poverty & Inequality Estimations Qualitative Quantitative (Tobit Models) Qualitative; Quantitative (MultiNominal Logit) Firm Survey Main Result (Impact Assessment) Macro: Inconclusive, ranging from no effect to rates of return in excess of 100% per annum. Micro: Positive correlation between transport infrastructure investment and production, productivity as well as poverty alleviation, although some doubt about the benefits to the poor. Improved accessibility from the EPFRP led to changes in land allocation and in yields. The mean cotton sales share of household income has more than doubled, but gains to mean consumption both small and insignificant. Rural road rehabilitation investments are pro-poor, but the very unequal expenditure distribution in 1998 had not improved much in 2004. Increase income and improve chances of breaking out of the poverty trap. Statistical significance of the association is not robust. It was mainly the ‗small-scale private companies‘ engaged in ‗agricultural marketing‘ which had made a positive relocation decision Dataset CrossCountry; Time Period 1980-2010 Household Panel Pooled Repeated crosssection postharvest surveys Successive cross sections (Pooled) 1997-2002 Successive cross sections 1998 & 2004 Successive cross sections (Pooled) Agribusiness (trasnport) crosssectional survey 1996 & 2005 1998 & 2004 1996 & 2005 Source: Author. Our entire analysis leads us to the most important policy question: To what extent, given the effectiveness, efficiency, and sustainability of this pilot project – the EPFRP – and the ensuing transport policy recommendations, have the best practices and the lessons learnt been enhanced to the national level and enshrined in Zambia‟s Transport Policy? We will take a look at this policy issue in what follows. 348 9.4. Changes in policy on transport infrastructure investment in Zambia It turns out that infrastructure development, most particularly in the rural areas, is one of the major needs for Zambia‘s development and is upheld in both the Fifth National Development Plan, about to be completed, and the Sixth National Development Plan, soon to be launched, as well as in the National Vision 2030 (OECD et al., 2010). In line with the above objective, the government recognised the major challenges to private-sector development in 2004 when it prepared its first private-sector-development reform programme (PSDRP-I)(OECD, 2010). Nevertheless, despite the importance of the rural road network in the national road system, and given the fact that it accounts for more than half of the transport network, it still only gets a marginal part of the national budget allocated to road construction, rehabilitation and maintenance (Escobal and Ponce, 2003)? Moreover, by taking a closer look at how employment generation is taken into account in the decision-making process on allocation and use of resources under the Zambian Public Investment Programme this leads us to shed light on the employment implications of the employment intensive transport infrastructure projects in Zambia? To answer these questions, we have to understand the political and institutional environment according to Howe(1984) and the history of the LBT approach in Zambia. From 1987 to 1994, a labour-based feeder roads project was implemented in Northern Province, funded by the Government of Norway, and with ILO technical assistance. The project was implemented through the Ministry of Local Government and Housing (MoLGH). Another project funded by the Government of Finland was implemented under the Provincial Roads Engineer in Lusaka from 1991 to 1993. This project was concerned with the improvement and maintenance of provincial and district roads in Lusaka Province (both earth and gravel roads). A national workshop formulated a policy on the use of LBT in the Zambian road sector, and provided an important input to the Road Maintenance Initiative (RMI) seminar held in Lusaka in 1993. Based on the experience of these projects and the agreement on the national policy, the Government decided to expand its involvement with LBT development. Since its inception in the early 1960s the Roads Training School (RTS) is the main training provider for road works in Zambia. The training component of the NORAD supported and ILO managed Northern Province project was, in mid-1994, transferred to the 349 RTS in Lusaka to form part of a broad Road Sector Programme (RSP-I) of the Ministry of Works and Supply funded by NORAD. The RTS started to train private small-scale routine maintenance contractors in 1995.311 The Roads Department of the Ministry of Works and Supply (MoWS) was responsible for constructing, operating and maintaining trunk and main roads (T and M) and about 60% of district roads (D and RD), whereas, technically, D and RD roads and the unclassified roads were under the jurisdiction of the District Councils. Roads were financed by the roads fund with resources derived from fuel taxes, levies on heavy trucks and other tariffs. The fuel tax and tariffs were adjusted regularly to ensure that income is sufficient to meet maintenance requirements. However, the government had been borrowing heavily from the Roads Fund, which had reduced resources and limited the National Road Board's (NRB) ability to significantly increase the amount of road works carried out in the country. Also, there were no fewer than seven different government agencies that dealt with roads. This created problems in terms of coordination and financing of roads, as different players were funding roads in isolation. Some of these roads could not be put on sustainable maintenance because their existence was not known to the Funding Agency. This fragmentation was perpetuated by not having a Transport Policy in place (National Road Fund Agency, 2010a, Schulz and Bentall, 1998). The EPFRP clearly illustrated that local capacity can be created in District Councils and in the local private construction industry for improving and preserving rural road infrastructure using LBT approaches. In time the project has become an important showcase and despite the lack of interest at the mid-term evaluation, eventually contributed greatly towards acceptance of the LBT approach in Zambia and beyond its borders. Several other projects have now adopted the approach.312 The challenge that remained at the completion of the project in 2002 and still at the time of our field visits in 2004 and 2005 was to ensure that the trained contractors continue to have 311 Following the passing of the National Council for Construction Bill, 2003, which establishes the National Council for Construction, the Roads Training School has been transferred to NCC. 312 Several other development agencies have adopted the LBT approach for the implementation of their programmes in Zambia, including: Smallholder Enterprise and Marketing Programme (SHEMP) - which is financed by the International Food and Agricultural Development Agency (IFAD); The Danida supported project on upgrading Great West (M9) Lusaka-Mongu Road; Projects under the Zambia Social Investment Fund (ZAMSIF); The Highly Indebted Poor Countries (HIPC) – Emergency Drought Recovery Programme etc. 350 access to contract work in a competitive environment (ILO, 2002b, 2002a). Eastern Province didn‘t suffer from the syndrome that once the project was completed then the whole thing collapses, because before the project round-up the MoLGH decided to prepare the maintenance programme for the feeder roads for the whole province, which provided an opportunity to included those roads in the overall maintenance programme from 2002 until 2005 (Simon Tembo, 2005). The EPFRP has shown that decentralisation of road administration has potential for improving the delivery of rural transport infrastructure services. But the evidence from another study by Robinson and Stiedl(2001) suggests that it is proving difficult to realize fully the expected benefits of targeting and involving local communities in rural road projects or decentralizing transportation. Drawing on the literature and on field surveys in Nepal, Uganda, and Zambia they find a number of constraints to successful devolution, including:    the lack of local government powers to exercise political influence; insufficient financial resources; and lack of management capability. Riverson et al.,(1991) and Robinson and Stiedl(2001) conclude that increased participation of the poor in the planning, financing, and implementation process is important. In May 2000, the Government of the Republic of Zambia (GRZ) promulgated a Transport Policy and this gave birth to the creation of three Agencies through Acts of Parliament. The Road Traffic Act No. 11 of 2002 created the Road Transport and Safety Agency (RTSA); the Public Roads Act No. 12 of 2002 created the Road Development Agency (RDA), which is now responsible for all roads in the country. However, section 73 of the Public Roads Act allows the Agency to delegate authority to other institutions. Hence, through this re-centralisation the District Councils now only have delegated authority from the RDA. The financial arrangement for the maintenance of the feeder roads was being taken care of by the road fund levy. Until 2005 the management was still under the control of the District Councils preparing the maintenance programmes, which then were submitted to the local government‘s Department of Infrastructure and Support Services (DISS) under the MoLGH. Then in turn sent to the Roads Department, which consolidated the national budget (Simon Tembo, 2005). The National Road Fund Act No. 13 of 2002 created the National Road Fund Agency which is responsible for administering and managing all financial resources in the road sector. 351 It was therefore decided to overhaul the road sector agencies and make them semiautonomous institutions (National Road Fund Agency, 2010a). According to the Policy Guidelines for the Road Fund, the Road Fund shall be disbursed for road maintenance only in the following proportions: 20% for Urban Roads; 40% for Feeder Roads; 40% for Trunk and Main Roads, but not for road rehabilitation, road reconstruction or new road construction (National Road Fund Agency, 2010a). Phase I of the 15-year Road Sector Investment Programme (ROADSIP) started in March 1998 and ended in 2003. The second phase started in 2004 and will end in 2013. The main objectives of ROADSIP-I were e.g. to: (a) Bring the road core network to maintainable condition; (b) Improve road conditions; (c) Build capacity of road authorities; (d) Create employment for poverty alleviation; (e) Improve rural transport services; (f) Manage community roads etc. Originally it was the intention to implement ROADSIP-II within the same economic, institutional and legal framework as outlined in ROADSIP-I. The objectives have been improved after taking into account lessons learnt from ROADSIP-I and the EPFRP. ROADSIP-II addresses: Poverty in rural areas and gender imbalance through the use of LBT methods and packaging of contracts, maximum involvement of road users, transparency and accountability in tenders and needs based management and budgets (National Road Fund Agency, 2010b). Unfortunately, the thesis has been unable to show how the national complementary policies changes, particularly those pertaining to the transport sector, indirectly have affected the impact of the rural transport infrastructure investments, nor been successful in teasing out their impact from that of the rural roads improvements in Zambia‘s Eastern θrovince. The main reason why this is not possible is due to the fact that the comparable cross-sectional district-level household survey datasets are only available for two points in time. Moreover, the availability of annual PHS data can‘t be fully optimized because we don‘t have sufficient structure to the phasing of the EPFRP across the five districts (see chapter 3). The policy and road impacts are too highly covariate to disentangle with so few degrees of freedom. 352 9.5. Policy Implications The results of the thesis have implications for future rural road project investments in general and in Zambia in particular. Despite their popularity, very few aid-financed rural road projects in developing countries have been the subject of rigorous impact evaluations. Knowledge about their impacts and the heterogeneity in those impacts continues to be limited (van de Walle, 2009). Hence, the purpose of the thesis is that the wider development community donors and the Government of Zambia and the private sector will benefit from the results of testing various approaches to retrospective evaluations of rural roads interventions and the identification of methods that offer both rigour and validity. However, can the feeder roads investment programme in the study region be justified on the basis of the empirical findings from the thesis? Although little is known about the magnitude of the benefits from rural roads, Van de Walle(2009) underlines that plenty is known about their costs. These are relatively expensive investments (chapter 3). An aid donor faced with a limited budget needs to know more about how benefits compare to costs in order to make informed decisions between a road and e.g. an education intervention. As a first step, Van de Walle(2009) recommends that much more needs to be known about the impacts of roads. According to Van de Walle(2009) more information based on careful, rigorous impact evaluations will help improve road project appraisals, design, and selection methodologies. The thesis has several limitations. Among the most critical are some data constraints. The Government should put serious effort into organized, coordinated, and systematic data collection for the long run. Without such data, it is difficult for the government to monitor and evaluate the impact of various investments and to set future investment priorities to achieve stated objectives (Fan et al., 2004). A number of studies have shown that low quality feeder roads raise more poor people out of poverty per currency invested than high quality trunk roads, making them a win-win strategy for growth and poverty alleviation (Fan, 2004, Dercon and Hoddinott, 2005). Hence, despite the increasing prioritisation of rural road maintenance as enshrined in ζRFA‘s policy guidelines, the GoRZ should consider giving greater priority to low quality and rural feeder roads in its future road sector investment strategy. In 2002 a project, officially called ‗Governanceμ Enhanced δocal Governance for θoverty Reduction,‘ was formulated as part of the Second ωountry ωooperation Framework, 353 2002-2006. It was meant as a follow -up of the two UNDP/UNCDF funded projects implemented in the Eastern Province under the first Country Assistance Framework, i.e. the DDP and the FRP (chapter 3), with the objective to nationally up-scale the lessons learnt from both pilot programmes. A document was signed on 21/11/2002, just days after the Government approved the Decentralisation Policy on 18/11/2002 (UNDP, 2002). The survival of project activities beyond the life of the EPFRP depends on the effective implementation of the government's decentralization policy. Project sustainability is dependent on two critical conditions. The first is the proper funding of the eight District Councils to reorganize and hire competent professional people to run their activities. The second is the availability of funds to continue maintenance and rehabilitation tasks at reasonable levels of activity, commensurate to network needs (Schulz and Bentall, 1998). A risk factor, which was not addressed in the EPFRP design was continuous training in the use of LBT methods. Others should have been given the opportunity to start on their own e.g. by assisting local contractors get access to finances, in order to fill out the gap of the 25% of the LBT trained contractors who subsequently had died (Simon Tembo, 2005). Moreover, political commitment towards the utilisation of LBT is crucial for the continuation of this choice of construction technology. There had been cases in Western Province where the local members of parliament had asked contractors working on feeder roads running through their constituency to demobilize, because they think that the maintenance of the feeder roads will be done better using equipment based methods. Whereas the situation was very different in Eastern Province characterised by strong support to LBT. Even the Chiefs, such as Paramount Chief Mpezeni, began to insist that the feeder roads be maintained using LBT (Simon Tembo, 2005). The GoRZ / UNDP country programme for 2007-2010, which drew on the lessons from GoRZ / UNDP cooperation during 2002-2006, was explicitly based on the FNDP 2006-2010 (GoRZ and UNDP, 2007).313 According to the FNDP the emphasis continues to be on maintaining the existing infrastructure (GoRZ, 2005).The success of the EPFRP has been tremendous and it has had enormous influence on the design of other LBT projects being implemented in Zambia such as SHEMP and a DANIDA programme running in Western 313 The FNDP plan 2006-2011 combines the essential elements of the two preceding instruments: the first PRSP 2002-2004 and the Transitional National Development Plan TNDP, 2002-2005. 354 Province, as well as outside of Zambia, which were designed using experiences from the EPFRP for example: Training of contractors; procurement of equipments for contractors; and especially adopting the labour-based construction methods; the management system using district councils; etc. Nevertheless, there should have been a greater effort to lobby the government for enshrining a clear statement within its transport or decentralization policy that LBT should be used for road maintenance on feeder roads and community roads, which still hadn‘t happened by 2005. Unfortunately some of the momentum was lost immediately upon the completion of the EPFRP, because the ILO was suppose to have assisted the MoLGH with a roll-out plan to expand by applying the lessons learnt from the EPFRP to other provinces building on the political good will at the time. Unfortunately, something prevented this from happening (Simon Tembo, 2005). 355 9.6. Future Research Household surveys are necessary to capture the full treatment effects of road development, and to avoid the displacement of smaller-scale rural transport infrastructure investment by other more large-scale projects that politicians perceive as more profitable in terms of votes, it is still important to keep documenting the best way possible the benefits that this kind of public investment brings about on the welfare of the population it serves. According to Grosh and Glewwe(2000b) the four most common kinds of household survey analysis, which can be used to keep track of who benefits from and who loses from government expenditures, are: Simple descriptive statistics on living standards;314 monitoring poverty and living standards over time;315 describing the incidence and coverage of e.g. PWPs; and measuring the impacts of e.g. PWP on household behaviour and welfare. Champions of rural roads have rightly observed that benefits are likely to be broad. One response among donor agencies has been to devise special selection and appraisal criteria for rural roads that simply assume important social benefits, despite a general lack of rigorous empirical evidence. These are used as justification for abandoning economic analysis when, as is the case in many rural areas of developing countries, traffic levels are too low for conventional consumer surplus measures to make sense (van de Walle, 2002). Van de Walle(2009) suggests that impact evaluations can help provide better information on the non-pecuniary benefits of rural roads. The value of household surveys would be greatly enhanced if they were followed up by the routine use of panel surveys using sub-samples, in order to track performance (Karlsson and Berkeley, 2005) for policy analysis. Further work to identify the channels and mechanisms through which the different variables operate would be of great interest. It would also provide an opportunity to address many methodological and substantive 314 One of the first calls on survey data is to calculate descriptive statistics that describe the population rather than the particular sample that is available for analysis, and provide a starting point for the analytic and econometric analysis (Deaton, 1997). 315 When data are used for this purpose they must be comparable over time; for this to be the case, the data must be gathered using the same methods each time the survey is implemented. 356 issues associated with the collection, interpretation and analysis of panel data (Deininger and Okidi, 2003). There is also a need to go beyond the impact study to examine the aggregate dimensions of public expenditure and to build political institutions for the development of infrastructure. Thus, the supply-side of infrastructure development should be an important part of future research (Ahmed and Hossain, 1990). Impact evaluation is now high on the development agenda. In fact, as mentioned by Estache(2010), since the mid-2000s, the interest in analytically robust evaluations of the impact of projects, programmes or policies has exploded among development academics and field workers. The World ψank‘s Development IMpact Evaluation (DIME) initiative in 2010 has completed 175 impact studies in selected areas such as: Local development, rural roads, rural electrification, etc.316 A new international agency, 3IE, or the International Initiative for Impact Evaluation, has been created with the sole purpose of supporting impact evaluations. And the Network of Networks for Impact Evaluation or NONIE in cooperation with the Evaluation Cooperation Group of the multilateral development institutions and the United Nations Evaluation Group, has made considerable progress in expanding awareness of impact evaluation. The increased importance of this type of evaluation is linked to the focus on outcomes, as embodied in the MDGs, and the need to demonstrate to donor nations the impact of the development projects they help finance (Thomas, 2008). Of the over 200 rigorous studies already catalogued, more than half are in social protection, many for cash transfers. Just 2 percent are for agriculture, showing that there are clear gaps in coverage. A worthwhile goal would be that impact evaluation, together with its powerful complements of CBA, good M&E, and the judicious use of indicators, become an integral part of the way we work (Thomas, 2008). 316 This initiative promotes the embedding of (prospective) impact evaluations in project designs. 357 Based on a survey of evaluation, research and practice of the last 5-10 years Estache(2010) concludes that not all infrastructure interventions are suitable for impact evaluations based on experiments or quasi-experiments in order to increase accountability for intervention selection, implementation and sustainability. Moreover, various evaluations have provided many insights on why not all apparently comparable interventions have sometimes generated dissimilar impacts across locations. Differences in institutions, in legal or social incentives and norms, in access to and sources of financial resources, in technological preferences and choices or in initials conditions can all explain quite convincingly differences in impact according to Estache(2010). Some of these differences can be internalized in the design of interventions and this may be the most important lesson to be learned from the fast growing number of experiences. Evaluations will according to Estache(2010) only survive if all development stakeholders accept to include them in the costs of the intervention (i.e. loan preparation and monitoring costs) as a matter of routine. This is a small price to generate a public good and, maybe more importantly to some actors, a small price to increase accountability for aid effectiveness.  Thus, future research should be devoted to:  Further improving and developing appraisal techniques for rural road projects.  How do we scale-up the lessons learnt.  Discuss how impact studies can best influence policy.  How to improve the incentives for doing impact evaluations need to be improved. How to realize the potential linkage between impact evaluation and the results agenda. Of considerable interest is to better understand which contingent factors influence welfare impacts. One objective concerns better understanding what complementary public policies or investments would enhance the returns to rural road investments. Future work needs to strive to understand the conditions and factors that influence mean outcomes and their distribution, including interaction effects with other development initiatives. There are obviously important implications for data collection ─ to assure that 358 one collects data on factors likely to be dimensions of heterogeneous or explanatory factors (van de Walle, 2009). Finally, an important objective of the analysis of ex-post impacts should be to draw implications for the ex-ante evaluation of rural road investments. Similarly, evaluations can help draw useful lessons for the choice of road project monitoring indicators (Van de Walle, 2009). Moreover, a complementary general-equilibrium analysis as shown in Fan et al.,(2004) is also needed to analyse how government investment in rural areas affects not only the agricultural sector and rural areas, but also other sectors and district centres/villages. Ignoring these impacts severely underestimates the overall impact of public investment on poverty. 359 References 360 List of References ABADIE, A., DRUKKER, D., HERR, J. L. & IMBENS, G. W. 2004. Implementing matching estimators for average treatment effects in Stata. Stata Journal, 4, 290-311. ACEMOGLU, D., JOHNSON, S. & ROBINSON, J. 2001. The Colonial Origins of Comparative Development: An Empirical Investigation. American Economic Review, 91, 1369-1401. ADES, A. & GLAESER, E. 1999. Evidence on Growth, Increasing Returns, and the Extent of the Market. The Quarterly Journal of Economics, 114, 1025-1045. AFDB 2004. Regional assistance strategy paper for Southern Africa for the period 2004–2008. . AGHION, P., CAROLI, E. & GARCIA-PENALOSA, C. 1999. Inequality and Economic Growth: The Perspective of the New Growth Theories. Journal of Economic Literature, 37, 1615-1660. AHMED, R. & DONOVAN, C. 1990. Issues of infrastructural development: a synthesis of the literature, Volume 1, Washington, D.C., Digitized by Google. AHMED, R. & HOSSAIN, M. 1990. Developmental impact of rural infrastructure in Bangladesh By Raisuddin Ahmed, Mahabub Hossain, Washtington, D.C., IPFRI. AKDER, A. H. 2002. National Accounts - Agriculture. Linking agricultural statistics to other data sources for analysing rural indicators of social well being and equity. 8th IWG.AGRI Seminar Perspectives for Agriculture and Rural Indicators and Sustainability. Château de la Muette, Paris: OECD. Akee, R. 2006. The Babeldaob Road: The Impact of Road Construction on Rural Labor Force Outcomes in the Republic of Palau. IZA Discussion Papers 2452, Institute for the Study of Labor (IZA). AKYUZ, Y. & GORE, C. 1994. The investment-profits nexus in East Asian industrialization. UNCTAD Discussion Papers, 91. ALCALÁ, F. & CICCONE, A. 2003. Trade, Extent of the Market, and Economic Growth 1960-1996. UPF mimeo. ALESINA, A., DEVLEESCHAUWER, A., EASTERLY, W., KURLAT, S. & WACZIARG, R. 2003. Fractionalization. Journal of economic growth, 8, 155. ALESINA, A. & RODRIK, D. 1994. Distributive politics and economic growth. Quarterly Journal of Economic, 109, 465-490. ALESINA, A., SPOLAORE, E. & WACZIARG, R. 2000. Economic integration and political disintegration. American Economic Review, 90, 1276-1296 ALI-NEJADFARD, F. 2000. Rural travel and transport and economic development: Problems and prospects - examples from Malawi and Zimbabwe. In: KAUMBOTHO, P., PEARSON, R. A. & SIMALENGA, T. E. (eds.) Empowering Farmers with Animal Traction into the 21st Century. Angrist, J.D. and J.-S. Pischke. 2009. Mostly harmless econometricsμ an Empiricist‘s Companion, Princeton University Press. ANGRIST, J., IMBENS, G. & RUBIN, D. 1996. Identification of Causal Effects Using Instrumental Variables,. Journal of the American Statistical Association,, XCI, 444-455. Antle, J. M, 1983. Infrastructure and Aggregate Agricultural Productivity: International Evidence, Economic Development and Cultural Change, vol. 31(3), pages 609-19, April. ARAAR, A. & DUCLOS, J.-Y. 2007a. DASP version 1.4. DASP: Distributive Analysis Stata Package. User Manual.: Université Laval. PEP, CIRPEE and World Bank. ARAAR, A. & DUCLOS, J.-Y. 2007b. Poverty and inequality components: a micro framework. CIRPEE Working Paper, 07, 44. Archondo-ωallao, R., β009, ―RηζET Road ζetwork Evaluation Tools, User Guide, Version β.0.‖ SSATP Working Paper No. 89-A, World Bank. ASCHAUER, D. A. 1989. Is public expenditure productive. Journal of Monetary Economics, 23, 77-. ASIST, 2003a. ASIST, 2003b. ATKINSON, A. B. 1987. On Measurement of Poverty. Econometrica, 55, 244-263. Badiani, R., & K P C Rao & Pramila Krishnan & Stefan Dercon, 2007. Changes in Living Standards in Villages in India 1975-2004: Revisiting the ICRISAT Village Level Studies. Working Paper. BALAT, J. F., BRAMBILLA, I. & PORTO, G. G. 2004. An Analysis of the WTO Development Round on Poverty in Rural and Urban Zambia. BALAT, J. F. & PORTO, G. G. 2005a. The WTO Doha Round, Cotton Sector Dynamics and Poverty Trends in Zambia. Policy Research Working Paper Series, 3697. 361 BALAT, J. F. & PORTO, G. G. 2005b. Globalization and complementary policies: poverty impacts in rural Zambia. Working Paper 11175, 42. BALDWIN, R. R. E., MARTIN, P. & I. P. OTTAVIANO, G. I. P. 2001. Global Income Divergence, Trade, and Industrialization: The Geography of Growth Take-Offs. Journal of economic growth, 6, 5. Balisacan, A., E. Pernia amd A Asra. 2002. Revisting growth and poverty reduction in Indonesia: What do Sub-National Data show?, ERD Working Paper Series No25, Asian Development Bank. Baltagi, B.H. 2010. Pooling Time-Series of Cross-Section Data in Solutions Manual for Econometrics. SpringerLink. BALTAGI, B. H. 2001. Econometric Analysis of Panel Data, Chichester, John Wiley and Sons, Chichester, 1995, (2nd Edition, 2001). Badiani, R. & K P C Rao & Pramila Krishnan & Stefan Dercon, 2007. Changes in Living Standards in Villages in India 1975-2004: Revisiting the ICRISAT Village Level Studies. BANERJEE, A. & DUFLO, E. 2000. A Reassessment of the relationship between inequality and growth: A comment. BANERJEE, A. & DUFLO, E. 2003. Inequality and Growth: What Can the Data Say? Journal of economic growth, 8, 267. BANERJEE, A., DUFLO, E., and QIAN, N. 2009. On the Road: Access to Transportation Infrastructure and Economic. Growth in China. (Preliminary Version) March, 2009. BANISTER, D. & BERECHMAN, J. 2000. Transport Investment and Economic Development, Routledge (UK). BANJO, G. 2005. Promoting Intermediate Means of Transport: Emerging Experience from Zambia. Transport Forum. Washington DC: World Bank. BANJO, G. A. 1999. Management and Financial Framework for Rural Transport Infrastructure: A Tool for Poverty Reduction. In: TRICORNE, E. D. (ed.) The Food Chain in Sub-Saharan Africa. Centre for Applied Studies in International Negotiations. Geneva. BAR-SHIRA, Z., FINKELSHTAIN, I. & SIMHON, A. 2003. Cross-Country Productivity Comparisons: The ?Revealed Superiority? Approach. Journal of economic growth, 8, 301. BARRO, R. J. 1990. Government spending in a simple endogenous growth model. Journal of Political Economy 98, 103-125. BARRO, R. J. 1991. Economic Growth in a Cross Section of Countries. The Quarterly Journal of Economics, 106, 407-43. BARRO, R. J. & SALA-I-MARTIN, X. 1992. Public finance in models of economic growth. Review of Economic Studies 59, 645-661. BARRO, R. J. & SALA-I-MARTIN, X. 1995. Economic Growth, New York, McGraw-Hill, Inc. BARRO, R. R. J. 2000. Inequality and Growth in a Panel of Countries. Journal of economic growth, 5, 5. BARWELL, I. 1996. Transport and the village: Findings from African village-level travel and transport surveys and related studies, Washington, D.C., Discussion Paper, no. 344: Africa Region Series. BASU, K. 1997. Analytical Development Economics. The Less Developed Economy Revisited., Cambridge, Massachusetts and London, England, The MIT Press. BAUM, C. F. 2006. An Introduction to Modern Econometrics Using Stata, Stata Press. BAXTER, M. & KING, R. G. 1993. Fiscal Policy in General Equilibrium. American Economic Review, 83, 315-34. BECKER, S. & ICHINO, A. 2002. Estimation of average treatment effects based on Propensity Scores. Stata Journal, 2, 358-377. BELLEMARE, M. F. 2009. The Productivity Impacts of de Jure and de Facto Land Rights. Duke University Working Paper. BEN-DAVID, D. 1993. Equalizing Exchange: Trade Liberalization and Income Convergence. The Quarterly Journal of Economics, 108, 653-79. BENHABIB, J. J. & SPIEGEL, M. M. 2000. The Role of Financial Development in Growth and Investment. Journal of economic growth, 5, 341. BHAGWATI, J. & SRINIVASAN, T. N. 2002. Trade and Poverty in the Poor Countries. American Economic Review, 92, 180-183 BIGSTEN, A. & SHIMELES, A. 2003. Prospects for Pro-Poor Growth in Africa. 362 Binswanger, H. P. & Khandker, Shahidur R. & Rosenzweig, Mark R., 1993. How infrastructure and financial institutions affect agricultural output and investment in India, Journal of Development Economics, Elsevier, vol. 41(2), pages 337-366, August. BLUNDELL, R. & COSTA DIAS, M. C. 2005. Alternative Approaches to Evaluation in Empirical Microeconomics. Institute for Fiscal Studies, 76. BLUNDELL, R. & MACURDY, T. 1999. Labor supply: A review of alternative approaches. In: ASHENFELTER, O. & CARD, D. (eds.) In: Handbook of Labor Economics. Elsevier BOCKSTETTE, V., CHANDA, A. & PUTTERMAN, L. 2002. States and Markets: The Advantage of an Early Start. Journal of economic growth, 7, 347. BOOTH, D., HOLLAND, J., HENTSCHEL, J., LANJOUW, P. & HERBERT, A. 1998. Participation and Combined Methods in African Poverty Assessment: Renewing the Agenda. Report commissioned by the UK Department for International Development for the Working Group on Poverty and Social Policy, Special Program of Assistance for Africa. Bourguignon, F. and L. Pereira da Silva, ed. (2003), The Impact of Economic Policies on Poverty and Income Distribution: Evaluation Techniques and Tools, Oxford University Press, New York BRAMBILLA, I. & PORTO, G. G. 2005. Farm Productivity and Market Structure. Evidence From Cotton Reforms in Zambia. Yale University Economic Growth Center Discussion Paper, 919, 42. BRAMBILLA, I. & PORTO, G. G. 2007. Market Structure, Outgrower Contracts and Farm Output. Evidence From Cotton Reforms in Zambia. BRAUTIGAM, D. 2009. The Dragon‟s Gift. The Real Story of China in Africa, Oxford University Press. BROCK, W. A. & DURLAUF, S. N. 2000. Growth Economics and Reality. NBER Working Papers, 8041. BROUWER, A. E., MARIOTTI, I. & VAN OMMEREN, J. N. 2002. The firm relocation decision: a logit model. Paper presented at the 42nd ERSA conference, Dortmund, Germany. BURKI, S. J., DAVIES, D. G., HOOK, R. H. & THOMAS, J. W. 1976. Public Works Programmes in Developing Countries: A comparative analysis. 78. BUYS, P., DEICHMANN, U. & WHEELER, D. 2006. Road Network Upgrading and Overland Trade Expansion in Sub-Saharan Africa World Bank Policy Research Working Paper No. 4097. Buyst, E., & Stefan Dercon & Bjorn Van Campenhout, 2000. Road expansion and market integration in the Austrian low countries during the second half of the 18th century. Center for Economic Studies Discussion papers ces0010, Katholieke Universiteit Leuven. Calvo, C., & Stefan Dercon, 2007. Chronic Poverty and All That: The Measurement of Poverty over Time, CSAE Working Paper Series 2007-04, University of Oxford. CAMERON, A. C. & TRIVEDI, P. K. 2005. Supplement to microeconometrics: Methods and applications, New York, Cambridge University Press. CAMERON, A. C. & TRIVEDI, P. K. 2009. Microeonometrics Using Stata, Stata Press. CANNING, D. 1998. A Database of World Stocks of Infrastructure, 1950-95. World Bank Economic Review, 12, 549-47. CANNING, D. & PEDRONI, P. 1999. Infrastructure and Long Run Economic Growth. 50. CARAPETIS, S., BEENHAKKER, H. L. & HOWE, J. D. F. 1984. The Supply and Quality of Rural Transport Services in Developing Countries. A Comparative Review. World Bank Staff Working Papers, 50. CARRUTHERS, I. & KYDD, J. 1997. The Development and Direction of Agricultural Development Economics: Requiem or Resurrection? Journal of agricultural economics, 48, 223-38. CASLEY, D. J. & KUMAR, K. 1987. Project monitoring and evaluation in agriculture., Baltimore and London, The Johns Hopkins University Press. CASSOU, S. P. & LANSING, K. J. 1999. Fiscal policy and productivity growth in the OECD. Canadian Economics Association, 32, 1215-1226. CENTER FOR GLOBAL DEVELOPMENT (CGD) 2006. When will we ever learn? Improving lives through Impact Evaluation. Evaluation Gap Working Group. CHEN, S., MU, R. & RAVALLION, M. 2006. 'Are There Lasting Impacts of a Poor-Area Development Project?' World Bank Policy Research Working Paper 4084, 48. CHEN, S. & RAVALLION, M. 2008. The Developing World Is Poorer Than We Thought, But No Less Successful in the Fight against Poverty. Policy Research Working Paper 4703. CHEN, X., ENDER, P., MITCHELL, M. & WELLS, C. 2010. Stata Web Books: Logistic Regression with Stata. UCLA. 363 CHIPIKA, S. 2005. A Study to Assess the Evidence Base and Methodologies for Poverty Reduction Impact Assessment in EIIP/ASIST- Africa Programme. ILO/EIIP ASIST Programme Harare. December 2005CHIWELE, D. K., MUYATWA-SIPULA, P. & KALINDA, H. 1996. Private Sector Response to Agricultural Marketing Liberalisation in Zambia. A Case Study of Eastern Province Maize Markets. In: OLUKOSHI, A. (ed.). Uppsala: Nordiska Afrikainstitutet. CHIWELE, D. K., MUYATWA-SIPULA, P. & KALINDA, H. 1998. Private Sector Response to Agricultural Marketing Liberalisation in Zambia. A Case Study of Eastern Province Maize Markets. In: OLUKOSHI, A. (ed.) Research report. Uppsala: Nordiska Afrikainstitutet. CHUNG, K. 2000. Qualitative Data Collection Techniques. In: GROSH, M. & GLEWWE, P. (eds.). Washington D.C.: World Bank. CICCONE, A. & PAPAIOANNOU, E. 2005. Human Capital, the Structure of Production, and Growth. UPF mimeo, 52. Cirera, X., & Channing Arndt, 2008. Measuring the impact of road rehabilitation on spatial market efficiency in maize markets in Mozambique. Agricultural Economics, International Association of Agricultural Economists, vol. 39(1), pages 17-28. CLAESSENS, S. & LAEVEN, L. 2002. Financial Development, Property Rights, and Growth World Bank Policy Research Working Paper, 2924. CLAY, E. J. 1986. Rural Public Works and Food-for-Work: A Survey. World Development, 14, 12371252. Clifton, J. 1998. Preparation of Enhanced Training Programme for District Staff Forming Key - Staff of Contract Management Units. Consultancy Report. July. Clifton, J., Simon Tembo and Arjen During. 2001. Lessons Learned and Possible Scenarios for Future Support. Ministry of Local Government and Housing Feeder Roads Project Eastern Province Project no. ZAM/93/009 and ZAM/93/CO2. ILO, UNDP and UNCDF. April 2001. COHEN, D. & SOTO, M. 2001. Growth and Human Capital: Good Data, Good Results OECD Development Centre Technical Paper, 179. COLLIER, P. 2007. The Bottom Billion, Oxford University Press. COLLIER, P. 2009. Wars, Guns, and Votes. Democracy in Dangerous Places, Harper. COLLIER, P. & GUNNING, J. W. 1999. Why Has Africa Grown Slowly. Journal of Economic Perspectives, , 13, 3-22. COMMISSION FOR AFRICA 2005. Our Common Interest. Report of the Commission for Africa. Action for a Strong and Prosperous Africa. CORNIA, G. A. (ed.) 2004. Inequality, Growth and Poverty in an Era of Liberalization and Globalization., Oxford: Oxford University Press. COUDOUEL, A., HENTSCHEL, J. & WODON, Q. 2002a. Chapter 1: Poverty Measurement and Analysis. In: BANK, W. (ed.) PRSP Sourcebook. Washington D.C. COUDOUEL, A. J., HENTSCHEL, J. & WODON, Q. 2002b. Poverty Measurement and Analysis. In: WORLDBANK (ed.) PRSP Sourcebook. Washington, DC. CSO 1994. National Census of Agriculture 1990/92. Census Report (Part I). Lusaka: Central Statistical Office. CSO 1996a. Living Conditions Monitoring Survey I 1996. Data user guide. Document on the collection, entry and editing of the LCMS data.: CENTRAL STATISTICAL OFFICE. CSO 1996b. Living Conditions Monitoring Survey - I 1996. . CSO 1996c. Sampling for the living conditions monitoring survey. In: SOCIAL STATISTICS BRANCH OF THE CSO, C. T. L. C. M. U. L. (ed.). CSO 1996e. Living Conditions Monitoring Survey-I 1996: Survey Design Logistics. In: OFFICE, C. S. (ed.). CSO 1998a. Living Conditions in Zambia (1998). Preliminary Report. In: UNIT, L. C. M. (ed.). Lusaka: Central Statistical Office. CSO 2001. 2000 Census of Population and Housing. Preliminary Report. In: CENTRAL STATISTICAL OFFICE, R. O. Z. (ed.) Preliminary report for the 2000 Census of Population and Housing Population count ed. CSO 2003. Zambia 2000 Census of Population and Housing. Agriculture Analytical Report. In: OFFICE, C. S. (ed.). Lusaka: Republic of Zambia Central Statistical Office. CSO. 2004b. RE: CSO Press Release: CPI - December 2004. 364 CSO 2005. Living Conditions Monitoring Survey report 2004. In: BRANCH, L. C. M. (ed.). Central Statistical Office. CURRY, S. & WEISS, J. 1993. Project Analysis in Developing Countries, MacMillan. DAGDEVIREN, H., VAN DER HOEVEN, R. & WEEKS, J. 2001. Redistribution Matters: Growth for Poverty Reduction. EMPLOYMENT PAPER. Dargay, J.M. 2002. Determinants of car ownership in rural and urban areas: A pseudo-panel analysis. Transportation Research Part E 38 (2002) 351-366. DE BOK, M. 2004. Explaining the location decision of moving firms using their mobility profile and the accessibility of locations. European Regional Science Association in its series ERSA conference papers with number ersa04p338. DE LA FUENTE, A. & DOMENECH, R. 2001. Schooling data, technological diffusion and the neoclassical Model. American Economic Review, 91, 323-327. Deaton, A. 2009. Instruments of Development: Randomization in the Tropics, and the Search for the Elusive Keys to Economic Development,‖ ζψER Working θaper #14θ90. DEATON, A. 1997. The analysis of household surveys: a microeconometric approach to development policy, Washington, D.C., The Johns Hopkins University Press. Deaton, A. 1985, Panel data from time series of cross-sections, Journal of Econometrics 30(1-2), 109–26. DEGNBOL MARTINUSSEN, J. 1994. Samfund, Stat og Marked. En Kritisk gennemgang af teorier om udvikling i den 3. verden, Mellemfolkeligt Samvirke. DEHEJIA, R. H. & WAHBA, S. 2002. Propensity score-matching methods for Non-experimental causal studies. The Review of Economics and Statistics, 84, 151-161. DEICHMANN, U., FAY, M., KOO, J. & LALL, S. V. 2002. Economic structure, productivity, and infrastructure quality in southern Mexico. DEININGER, K. & OKIDI, J. 2003. Growth and Poverty Reduction in Uganda, 1999-2000: Panel Data Evidence. SSRN eLibrary. DEININGER, K. & OLINTO, P. 2000. Why liberalization alone has not improved agricultural productivity in Zambia : The role of asset ownership and working capital constraints, Volume 1. Policy Research Working Paper, 2302, 28. DENT, C. M. 2008. The Asian Development Bank and Developmental Regionalism in East Asia. Third World Quarterly, 29, 767-786. DERCON, S., GILLIGAN, D. O., HODDINOTT, J. & WOLDEHANNA, T. 2007. The impact of roads and agricultural extension on consumption growth and poverty in fifteen Ethiopian villages. CSAE WPS/2007-01. DERCON, S. & HODDINOTT, J. 2005. Livelihoods, growth, and links to market towns in 15 Ethiopian villages FCND Discussion Paper 194. DERCON, S. & KRISHNAN, P. 1998. Changes in Poverty in Rural Ethiopia 1989-1995: Measurement, Robustness Tests and Decomposition. 45. DESDOIGTS, A. A. 1999. Patterns of Economic Development and the Formation of Clubs. Journal of economic growth, 4, 305. DEVEREUX, S. 2002. FROM WORKFARE TO FAIR WORK. The Contribution of Public Works and other Labour–Based Infrastructure Programmes to Poverty Alleviation. Issues in Employment and Poverty 45. DINKA, T. & KENNES, W. 2007. Africa's Regional Integration Arrangements: History and Challenges. ECDPM Discussion Paper 74. DOLLAR, D. 1992. Outward-Oriented Developing Economies Really Do Grow More Rapidly: Evidence from 95 LDCs, 1976-1985. Economic Development and Cultural Change, 40, 523-44. DOLLAR, D. & KRAAY, A. 2001. Trade, Growth, and Poverty. Policy Research Working Paper Series, 2615, 46. DONER, R. F. & SCHNEIDER, B. 2000. The New Institutional Economics, Business Associations and Development. 25. DOROSH, P., HYOUNG-GUN WANG, LIANG YOU, AND EMILY SCHMIDT. 2008. Crop Production and Road Connectivity in Sub-Saharan Africa: A Spatial Analysis. DORWARD, A., FAN, S., KYDD, J., LOFGREN, H., MORRISON, J., POULTON, C., RAO, N., SMITH, L., TCHALE, H., THORAT, S., UREY, I. & WOBST, P. 2004a. Institutions and Policies for Propoor Agricultural Growth. Development Policy Review, 22, 611-622. 365 DORWARD, A., FAN, S., KYDD, J., LOFGREN, H., MORRISON, J., POULTON, C., RAO, N., SMITH, L., TCHALE, S., THORAT, S., UREY, I. & WOBST, P. 2004b. Rethinking Agricultural Policies for Pro-Poor Growth. Natural Resource Perspectives, 94. DORWARD, A., KYDD, J., MORRISON, J. & IAN UREY, I. 2002. A Policy Agenda for Pro-Poor Agricultural Growth 71. DORWARD, A., KYDD, J., MORRISON, J. & UREY, I. 2004c. A Policy Agenda for Pro-Poor Agricultural Growth. World Development, 32, 73–89. DORWARD, A., KYDD, J. & POULTON, C. (eds.) 1998. Smallholder Cash Crop Production under Market Liberalisation. A New Institutional Economics Perspective: Cab International. DTZ PIEDA CONSULTING 2004. Economic Effects of Road Infrastructure Improvements: Stage 3 Report. Planning, Economic & Development Consultants. DUFδη, ESTHER AζD θAζDE, RηHIζI ―Dams‖, Quarterly Journal of Economics, 1ββ(β)μθ01-646. Duflo, E., R. Glennerster, and ε. Kremer, (β008), ―Using Randomization in Development Economics Researchμ A Toolkit,‖ in Schultz, T.θ. and J. Strauss eds, Handbook of Development Economics, 38953962. EASTERLY, W. 2001. The Lost Decades: Developing Countries' Stagnation in Spite of Policy Reform 1980?1998. Journal of economic growth, 6, 135. EASTERLY, W. & LEVINE, R. 1997. Africa's Growth Tragedy: Policies and Ethnic Divisions. Quarterly Journal of Economics, 112, 1203 - 1250 EASTERLY, W. & LEVINE, R. 2002. It's Not Factor Accumulation: Stylized Facts and Growth Models. Working Papers Central Bank of Chile, 164, 64. EASTERLY, W. & REBELO, S. 1993. Fiscal policy and economic growth. Journal of Monetary Economics, 32, 417-58. EASTERN PROVINCE CHAMBER OF COMMERCE AND INDUSTRY. INVESTMENT PROFILE [Online]. Chipata. Available: http://www.easternchamber.com/Investment/socio-economics.htm [Accessed 9th of June 2005]. EDMONDS, G. & DE VEEN, J. 1992. A labour-based approach to roads and rural transport in developing countries. International Labour Review, 131, 95-110. EDMONDS, G. A. & HOWE, J. (eds.) 1980. Roads and Resources: An Intermediate Technology Publication. EDWARDS, S. 1998. Openness, Productivity and Growth: What Do We Really Know? Economic Journal, 108, 383-98. Escobal, J., 2005. The Role of Public Infrastructure in Market Development in Rural Peru. MPRA Paper 727, University Library of Munich, Escobal, G. and M. Torero. 2005. Measuring the impact of asset complementarity: the Case of Rural Peru, Cuadernos de Economia. 42 (May), 137-164.ESCOBAL, J. & PONCE, C. 2003. The benefits of rural roads. Enhancing income opportunities for the rural poor. Documentos de Trabajo dt40b, Grupo de Análisis para el Desarrollo (GRADE), 56. ESTACHE, A. 2010. A Survey of Impact Evaluation of Infrastructure Projects, Programs and Policies. ECORE Discussion Paper No. 2010/16. Estache, A., V. Foster and Q. Wodon. 2002. Accounting for Poverty in Infrastructure Reform – Learning from δatin America‘s Experience, World ψank Institute θublications, Studies in Development Series. FAFCHAMPS, M. & HILL, R. V. 2004. Selling at the farm-gate or travelling to market. CSAE WPS/200430. . FALKINGER, J. & GROSSMANN, V. 2005. Institutions and Development: The Interaction Between Trade Regime and Political System. Journal of economic growth, 10, 231. FAN, S., CHAN-KANG, C. 2004. Road development, economic growth, and poverty reduction in China. IFPRI DSGD Discussion Paper, 12, 86. FAN, S. & HAZELL, P. 2000. Should developing countries invest more in less-favored areas? An empirical analysis of rural India. Economic and Political Weekly 35 (April 22): 1455-1464. FAN, S., HAZELL, P. & THORAT, S. 2000a. Government spending, agricultural growth, and poverty in rural India. American Journal of Agricultural Economics 82 (4): 1038-1051. FAN, S., ZHAN, X. & RAO, N. 2004. Public expenditure, growth and poverty reduction in rural Uganda. DSCD Discussion Paper No.4, 71. 366 FAN, S., ZHANG, L. & ZHANG, X. 2002. Growth, inequality, and poverty in rural China: the role of public investments. IFPRI Research reports, 125, 81. FAO 2009. Statistical On-line Crop Database. FAYE, M. L., MCARTHUR, J. W., SACHS, J. D. & SNOW, T. 2004. The Challenges Facing Landlocked Developing Countries. Journal of Human Development, Vol. 5,. FEARON, J. J. D. 2003. Ethnic and Cultural Diversity by Country. Journal of economic growth, 8, 195. FISMAN, R. & LOVE, I. 2003. Trade Credit, Financial Intermediary Development, and Industry Growth. Journal of Finance. FISMAN, R. & LOVE, I. 2004. Financial Development and Growth in the Short and Long Run. NBER Working Paper, 10236, 28. FOSTER, J., GREER, J. & THORBECKE, E. 1984. A Class of Decomposable Poverty Measures. Econometrica, 52, 761-765. Foster, V., and Briceño-Garmendia, C. (eds.). β010. Africa‘s Infrastructure. A Time for Transformation. A co-publication of the Agence Française de Développement and the World Bank. FRANKEL, J. & ROMER, D. 1999. Does trade cause growth? American Economic Review, 89, 379-399. FRANKENBERG, E. 2000. Community and Price Data. In: GROSH, M. & GLEWWE, P. (eds.) Designing Household Survey Questionnaires for Developing Countries. Lessons from 15 years of the Living Standards Measurement Study. Washington D.C.: World Bank. FREEMAN, R. B. & LINDAUER, D. L. 1999. Why Not Africa? . NBER Working Paper No. W6942 FUJITA, M., KRUGMAN, P. & VENABLES, A. J. 1999. The Spatial Economy. Cities, Regions, and International Trade., Cambridge, Massachusetts, London, England, The MIT Press. FUTAGAMI, K., MORITA, Y. & SHIBATA, A. 1993. Dynamic Analysis of an Endogenous Growth Model with Public Capital. Scandinavian Journal of Economics, 95, 607-25. GALLUP, J. L. & SACHS, J. D. 2000. Agriculture, Climate, and Technology: Why are the Tropics Falling Behind? American Journal of Agricultural Economics, 82, 731-737. GALLUP, J. L. & SACHS, J. D. 2001. The economic burden of malaria. American Journal of Tropical Medicine and Hygiene, 64, 85-96. GALLUP, J. L., SACHS, J. D. & MELLINGER, A. 1998. Geography and Economic Development. NBER Working Paper No. 6849. Gannon, C. and Liu, Z. 2000. Transport: Infrastructure and Services. Poverty Reduction Strategy Paper – A Source Book. Washington, D.C.: Southern Africa Department, The World Bank. Gannon, C. and Zhi Liu. 1997. Poverty and Transport. TWU discussion papers, TWU-30, World Bank, Washington, DC. GAUDE, J., GUICHAOUA, A., MARTENS, B. & MILLER, S. 1987. Rural Development and LabourIntensive Schemes: Impact Studies of Some Pilot Programmes. International Labour Review, 126, 423-446. GAUDE, J. & MILLER, S. 1992. Preface. International Labour Review, 131, 1. GAVIRIA, J., BINDISH, V. & LELE, U. 1989. The rural road question and Nigeria's agricultural development Managing Agricultural Development in Africa (MADIA) discussion paper, 40. Gibson, J. and S. Rozelle. 2003. Poverty and access to roads in Papua New Guinea, Economic development and cultural change, 52, 159-185. GILLIS, M., PERKINS, D. H., ROEMER, M. & SNODGRAS, D. R. 1992. Economics of Development, New York London, W.W. Norton & Company. GLEWWE, P. & JACOBY, H. 2000. Recommendations for Collecting Panel Data. In: GROSH, M. & GLEWWE, P. (eds.) Designing Household Survey Questionnaires for Developing Countries. Lessons from 15 years of the Living Standards Measurement Study. Washington, D.C.: World Bank. GOLDSTEIN, M. & UDRY, C. 2008. The Profits of Power: Land Rights and Agricultural Investment in Ghana. Journal of Political Economy,, 116, 981-1022. GOLLIN, D. & ROGERSON, R. 2010. Agriculture, Roads, and Economic Development in Uganda. NBER Working Papers, 15863, 64. GOMANEE, K., GIRMA, S. & MORRISSEY, O. 2005. Aid and growth in Sub-Saharan Africa: Accounting for transmission mechanisms. Journal of International Development, 17, 1055–1075. GORZ 2005. The Fifth National Development plan (FNDP) In: PLANNING, M. O. F. A. N. (ed.). Lusaka. 367 GORZ 2006a. Fifth national development plan 2006-2010. Broadbased wealth and job creation through citizenry participation and technological advancement. In: PLANNING., M. O. F. A. N. (ed.). Lusaka. GORZ & UNDP 2007. Country Programme Action Plan for 2007-2010 GOVEREH, J., JAYNE, T. S., TSCHIRLEY, D., DONOVAN, C., NIJHOFF, J. J., WEBER, M. & ZULU, B. 2000. Improving Smallholder and Agribusiness ηpportunities in Zambia‘s ωotton Sectorμ Key Challenges and Options. FSRP Working Paper No. 1. GRAMLICH, E. M. 1994. Infrastructure Investment: A review. Journal of Economic Literature, 32, 11761196. GREENE, W. H. 2000. Econometric Analysis, Prentice Hall. GROOTAERT, C. 1995. Structural Change and Poverty in Africa: A Decomposition Analysis of Côte d‘Ivoire. Journal of Development Economics, 47, 375-401. GROSH, M. & GLEWWE, P. 2000b. Making Decisions on the Overall Design of the Survey. In: GROSH, M. & GLEWWE, P. (eds.) Designing Household Survey Questionnaires for Developing Countries. Lessons from 15 years of the Living Standards Measurement Study. Washington: The World Bank. GRZ 2002b. Zambia Poverty Reduction Strategy Paper, 2002-2004. In: PLANNING, M. O. F. A. N. (ed.). Gwilliam, K., Foster, V., Archondo-Callao, R., Briceño-Garmendia, C., Nogales, A. and Sethi, K. 2008. The Burden of Maintenance: Roads in Sub-Saharan Africa. AICD, Background Paper, World Bank, Washington, D.C. G20. 2010. G20 Seoul Summit. Annex II. Multi-year Action Plan on Development. HAANYIKA, C. M. 2008. Rural electrification in Zambia: A policy and institutional analysis. Energy Policy, 36, 1044-1058. HAGGBLADE, S. & TEMBO, G. 2003. Development, diffusion and impact of conservation farming in Zambia. 79. HALL, R. & JONES, C. 1999. Why do some countries produce more output per worker than others? Quarterly Journal of Economics, 114, 83-116. HANUSHEK, E. & KIMKO, D. 2000. Schooling, labor-force quality, and the growth of nations. American Economic Review, 90, 1184-1208. HARDIN, J. & HILBE, J. 2007. Generalized Linear Models and Extensions. HÄRDLE, W., MÜLLER, M., SPERLICH, S. & WERWATZ, A. 2004. Nonparametric and Semiparametric Models: An Introduction, Berlin, Heidelberg, New York, Springer. Hammer, B. 2007. The FE-Estimator in Pseudo Panels. HARRISON, A. 1996. Openness and growth: A time-series, cross-country analysis for developing countries. Journal of Development Economics, 48, 419-447. HAUSMANN, R., PRITCHETT, L. & RODRIK, D. 2005. Growth Accelerations. Journal of economic growth, 10, 303. HAZELL, P. B. R. & HOJJATI, B. 1999. Farm/Non-Farm Growth Linkages in Zambia. Journal of African Economics, 4, 406-435. HE, X., FUNG, W. K. & ZHU, Z. Y. 2005. Robust estimation in generalized partial linear models for clustered data. J. Amer. Statist. Assoc., 100. HECKMAN, J., ICHIMURA, H. & TODD, P. 1997b. Matching as an Econometric Evaluation Estimator: Evidence from Evaluating a Job Training Programme. Review of Economic Studies, 64, 605-54. HERTEL, T. W. & WINTERS, L. A. (eds.) 2006. Poverty and the. WTO: Impacts of the Doha Development Agenda, New York: Palgrave. MacMillan. HIDEHIKO, I. 2005. Semiparametric Data Analysis. UCL. HILL, C. B. 2004. Trade Policy Reform and Liberalization during the 1990s. In: MCPHERSON, C. B. H. A. M. F. (ed.) Promoting and Sustaining Economic Reform in Zambia. Cambridge, Massachusetts and London, England.: Harvard University Press. HILL, C. B. & MCPHERSON, M. F. (eds.) 2004. Promoting and Sustaining Economic Reform in Zambia, Cambridge, Massachusetts, and London, England: Harvard University Press. HIRWAY, I. & TERHAL, P. 1995. Towards Employment Guarantee in India. Indian and International Experiences in Rural Public Works Programmes. HOLTZ-EAKIN, D. 1994. Public Sector Capital and the Productivity Puzzle. Review of Economics and Statistics, 76, 12-21. 368 HOROWITZ, J. L. & LEE, S. 2002. Semiparametric methods in applied econometrics: do the models fit the data? Statistical Modelling, 2, 3–22. HOWE, J. 1984. Rural Roads and Poverty Alleviation in Botswana. In: HOWE, J. & RICHARDS, P. (eds.) Rural Roads and Poverty Alleviation. HOWE, J. 2001. Assessing the impact of transport infrastructure on poverty reduction. HOWE, J. & RICHARDS, P. (eds.) 1984. Rural Roads and Poverty Alleviation., London: Intermediate Technology Publications Ltd. HSIEH, C. T. 2002. What Explains the Industrial Revolution in Asia: Evidence from Factor Markets. AER HULTEN, C. R. & SCHWAB, R. M. 1991. Public capital formation and the growth of regional manufacturing industries. National Tax Journal, 44, 121-135. HUNT, D. 1989. Economic Theories of Development: An Analysis of Competing Paradigms, New York, London, Toronto, Sydney, Tokyo, Singapore, Harvester Wheatsheaf. IAS 1996. Agricultural Sector Performance Analysis. In: MINISTRY OF AGRICULTURE, F. A. F. (ed.). Lusaka: Institute for African Studies. ICHINO, A., MEALLI, F. & NANNICINI, T. 2007. From Temporary Help Jobs to Permanent Employment: What Can We Learn from Matching Estimators and their Sensitivity?‖. Journal of Applied Econometrics. ILO 1981. Basic Needs in an economy under pressure. ILO 2002a. Joint Final Evaluation, Rehabilitation and Maintenance of Feeder Roads. ILO 2002b. Employment Intensive Investment Programme. Zambia. . ILO 2004. Why agriculture still matters. In: ILO (ed.) Employment, Productivity and Poverty Reduction. World Employment Report 2004-05. Geneva. IMBENS, G. & ANGRIST, J. 1994. Identification and estimation of local average treatment effects. Econometrica, 62, 467-76. IMF 2006. Direction of Trade Statistics, CD-ROM, September. IMF 2005. ZAMBIA. Report on the Observance of Standards and Codes (ROSC)—Data Module. Prepared by the IMF Statistics Department. INSTITUTE FOR TRANSPORT STUDIES UNIVERSITY OF LEEDS & LTD., I. T. T. 2003. Toolkit for the economic evaluation of world bank transport projects. Final report. In: WORLD BANK (ed.). Washington D.C.: University of Leeds. ISLAM, N. 1995. Growth Empirics: A panel data approach. Quarterly Journal of Economics, 110, 11271170. Ito, S. N. Kobayashi and Y. Wada. 2008. Leearning to evaluate the impact of aid. Nonie Working Paper, # 6, January. JACOBY, H. G. 1999. Access to Markets and the Benefits of Rural Roads. The World Bank, Policy Research Working Paper Series, 2028. JACOBY, H. G. 2002. Access to Markets and the Benefits of Rural Roads. The Economic Journal, 110, 713-737. JACOBY, H. G. & MINTEN, B. 2009. On Measuring the Benefits of Lower Transport Costs. Journal of Development Economics, 89, 28-38. JALAN, J. & RAVALLION, M. 1997. Spatial Poverty Traps. In: GROUP, D. R. (ed.). Washington D.C.: World Bank. Jalan, J. & Ravallion, M. 1998. Are there dynamic gains from a poor-area development program? Journal of Public Economics 67 (1998) 65–85. JALAN, J. & RAVALLION, M. 2002. Geographic Poverty Traps? A Micro Model of Consumption Growth in Rural China. Journal of Applied Econometrics, Vol. 17, No. 4, 2002, pp. 329-346. Jalan, J. and Ravallion, M. 2003. Estimating the Benefit Incidence of an Antipoverty Program by Propensity-Score Matching. Journal of Business & Economic Statistics 21(1): 19-30.. Jensen, T. H. 2009 .General equilibrium impact evaluation of road sector investment projects in Ghana. Paper prepared for presentation at the Twelfth GTAP conference in Santiago (10-12 June 2009) and the ECOMOD 2009 conference in Ottawa (24-26 June2009).JHA, D. & HOJJATI, B. 1993. Fertiliser use on Smallholder Farms in Eastern Province, Zambia. Washington D.C.: International Food Policy Research Institute. JOHNSTON, J. & DINARDO, J. 1997a. Econometric Methods, McGraw-Hill International Editions. KABWE, S. 2009. The Evolution of the Cotton and Textile Industry in Zambia. A presentation at ICAC Research Associate Program, Washington DC, 10 April, 2009. 369 KABWE, S. & TSCHIRLEY, D. 2007. An Effective Public-θrivate ωoordination in Zambia‘s ωotton Sector: Deliberation on the Cotton Act. Presented at the Agricultural Consultative Forum. May. KALDOR, N. Year. Capital Accumulation and Economic Growth. In: LUTZ, F. A. & HAGUE, D. C., eds. International Economics Association, 1963 London. London: Macmillan. K & Associates Professional Services and Daima Associates Limited (Kapsel and Daima). 2004. Comparative study on the impact of labour-based and equipment-based methods in road works in Tanzania. Final report. The United Republic of Tanzania Ministry of works. Kalinda, T. 2001. Lessons Learned from the Implementation of the District Development Planning and Implementation Project. Ministry of Local Government and Housing District Development Planning and Implementation Project. Eastern Province Project no. ZAM / 93 / 005 and ZAM / 93 / CO1. ILO, UNDP & UNCDF. KARLSSON, J. & BERKELEY, H. (eds.) 2005. Handbook on Rural Household, Livelihood and WellBeing, Wye: UNECE. KEDDEMAN, W. 1998. Of NETS and ASSETS. Effects and impacts of employment-intensive programmes - A review of ILO experience. KELLER, W. 2002. Trade and the Transmission of Technology. Journal of economic growth, 7, 5. KENYA ROADS BOARD. 2004. Labour-based technology [Online]. Available: http://www.krb.go.ke/labourbased.php http://www.krb.go.ke/map.php [Accessed 8th of November 2005 2005]. KHANDKER, S. R., BAKHT, Z. & KOOLWAL, G. B. 2006. The Poverty Impact of Rural Roads: Evidence from Bangladesh1. World Bank Policy Research Working Paper, 3875. KINGOMBE, C. K. M. 2010b. The Modus Operandi of the Investment-Trade Nexus. Conclusions and Policy Implications. Imperial College London (ICL) Working Paper. KINGOMBE, C. K. M. 2010a. Theories of Economic Growth and Empirics of Growth: An Exposition and Assessment of the Macro-level Theoretical and Empirical foundations of the Dynamic Models of Production and Consumption Growth at the Micro-level.‖ Iωδ Working θaper. KINGOMBE, C. K. M. 2009e. θoverty over Time in Zambia‘s Eastern θrovince.‖ Iωδ Working Paper. KINGOMBE, C. K. M. 2009d. The Rural Transport Infrastructure and Marketing Linkage within the Context of the Sub-Regional Zambia-Malawi-εozambique Growth Triangle.‖ Iωδ Working paper. KINGOMBE, C. K. M. 2009c. Analysis of Rural Household Surveys in ωhipata and δundazi District.‖ Iωδ Working paper. KINGOMBE, C. K. M. 2009b. District level estimation of the Differences in Outcomes linked to the θroduction of the εain ωash ωrops of Rural Roads Improvements in Eastern θrovince.‖ Iωδ working paper. KINGOMBE, C. K. M. 2009a. Regional Analysis of Eastern Province Feeder Road Project. District level estimation of the Consumption / Poverty Alleviation Effects of Rural Roads Improvements in Eastern Province of Zambia." ICL Working paper. Kilkenny, M., 1998. Transport Costs and Rural Development. Staff General Research Papers 1185, Iowa State University, Department of Economics. KÖGEL, T. T. & PRSKAWETZ, A. 2001. Agricultural Productivity Growth and Escape from the Malthusian Trap. Journal of economic growth, 6, 337. Koop, G., Dale J. Poirier, Justin L. Tobias. 2007. Bayesian Econometric Methods (Econometric Exercises). Cambridge University Press (January 15, 2007). KRUEGER, A. O. & BERG, A. 2003. Trade, Growth, and Poverty: A Selective Survey. IMF Working Papers 03/30. KRUGMAN, P. 1991a. Geography and Trade. KRUGMAN, P. 1991b. Increasing returns and economic geography. Journal of Political Economy, 99. KRUGMAN, P. 1998. The role of geography in development. Annual World Bank Conference on Development Economics. Washington D.C.: World Bank. KYDD, J. & DORWARD, A. 2001. The Washington Consensus on Poor Country Agriculture: Analysis, Prescription and Institutional Gaps. In: ASHLEY, C. & MAXWELL, S. (eds.) Rethinking Rural Development. KYDD, J. & DORWARD, A. 2004. Implications of Market and Coordination Failures for Rural Development in Least Developed Countries. Journal of international development, 16, 951-972. 370 KYDD, J., DORWARD, A., MORRISON, J. & CADISCH, G. 2004. Agricultural Development and Propoor Economic Growth in Sub-Saharan Africa: Potential and Policy. Oxford Development Studies, , 32. KYDD, J., DORWARD, A. & POULTON, C. 2002. Institutional dimensions of trade liberalisation and poverty OECD Global Forum on Agriculture, meeting on Agricultural Trade Reform, Adjustment and Poverty. OECD, Paris: ADU Working Paper 02/05. Laszlo, M. (ed.). 1996. Econometrics of panel data. A handbook of the theory with applications, 2nd revised Ed. 1996. LATHAM, M. C. 1965. Human Nutrition in Tropical Africa. Rome: FAO. LEBO, J. & SCHELLING, D. 2005. Design and Appraisal of Rural Transport Infrastructure: Ensuring Basic Access for Rural Communities. World Bank Technical Paper, 106. LEUVEN, E. & SIANESI, B. 2003. PSMATCH2: Stata module to perform full Mahalanobis and propensity score matching, common support graphing, and covariate imbalance testing. . LEVCHENKO, A. A. 2004. Institutional Quality and International Trade. IMF Working Paper Series, WP/04/231, 47. LEVINE, R. & RENELT, D. 1992. A Sensitivity Analysis of Cross Country Growth Regressions. American Economic Review, 82, 942-963. δEVIζSηHζ, J. β00γ. The World ψank‘s θoverty Reduction Strategy Paper Approach: Good Marketing or Good Policy? . G-24 Discussion Paper Series 21. LIMAO, N. & VENABLES, A. J. 1999. Infrastructure, geographical disadvantage, and transport costs. Policy Research Working Paper Series, 2257. LIPTON, M. 1996. Successes in anti-poverty. Issues in Development Discussion Paper. ILO. LOKSHIN, M. 2006. Semi-parametric difference-based estimation of partial linear regression models. LOKSHIN, M. & YEMTSOV, R. 2005. Has Rural Infrastructure Rehabilitation in Georgia Helped the Poor? World Bank Economic Review, 19, 311-333. LUBKER, M. 2002. Assessing the impact of past distributional shifts on global poverty levels. Employment paper, 39. Lucas, R.E. Jr. 1988. On the Mechanics of Economic Development. Journal of Monetary Economics 22 (1988) 3-42. North-Holland. LUSAKA TIMES. 2010. Zambia: Chipata-Mchinji railway commissioned. 27th of August 2010. MANKIW, G. N., ROMER, D. & WEIL, D. N. 1992. A Contribution to the Empirics of Economic Growth. The Quarterly Journal of Economics, 107, 407-437 MARTIMORT, D. D. & VERDIER, T. 2000. The Internal Organization of the Firm, Transaction Costs, and Macroeconomic Growth. Journal of economic growth, 5, 315. Mashiri, M. and Mahapa, S. 2002. Social Exclusion and Rural Transport: A Road Improvement Project. In Fernando, P. and Porter, G. (eds) Balancing the Load: Women, Gender and Transport. London: Zed Books. MATTHEWS, A. 2003. Regional Integration and Food Security in Developing Countries. Mccann and Sheppard. 2003. Regional Studies, 1360-0591, Volume 37, Issue 6, 2003, Pages 649 – 663. McCord, A and Slater, R. 2009. Overview of Public Works Programme in Sub-Saharan Africa. Report produced by ODI for the World Bank Social Protection Network. McCord, A. and Wilkinson, C. 2009. PSM and PWP Incidence; Evidence from Two Case Studies in South Africa, SALDRU Working Paper. Cape Town: Southern Africa Labour and Development Research Unit, University of Cape Town. MCCULLOCH, N., BAULCH, B. & CHEREL-ROBSON, M. 2000. Globalisation, poverty and inequality in Zambia during the 1990s. OECD Conference on Poverty and Income Inequality in Developing Countries: A Policy Dialogue on the Effects of Globalisation. OECD Paris: IDS. MCCULLOCH, N., BAULCH, B. & CHEREL-ROBSON, M. 2001. Poverty, Inequality and Growth in Zambia during the 1990s. Discussion Paper No. 2001/123, 47. MCFADDEN, D. 1974. Conditional logit analysis of qualitative choice behaviour. In: P.C. ZAREMBKA (ed.) Frontiers in Econometrics. New York and London: Academic Press. MCFADDEN, D. 1978. Modelling the choice of residential location. In: AL., A. K. E. (ed.) Spatial interaction theory and planning models. Amsterdam: North Holland. McKenzie, D.J., 2004. Asymptotic theory for heterogeneous dynamic pseudo-panels. Journal of Econometrics 120, 235–262. 371 MCPHERSON, M. F. 2004. The Role of Agriculture and Mining in Sustaining Growth and Development in Zambia. In: MCPHERSON, C. B. H. A. M. F. (ed.) Promoting and Sustaining Economic Reform in Zambia. Cambridge, Massachusetts and London, England.: Harvard University Press. MEGILL, D. J. 2000. Review of Sample Design for Zambia Post Harvest Survey (1997/98) and the Recommendations for Improving the Sampling Strategy and Estimation Procedures. FSRP Working Paper No. 2. Mengesha, T. and Osei-Bonsu, K.D. 2007. Rapid Assessment of Poverty Impacts (RAPI): Results of Baseline and Follow-up Surveys in Ethiopia. Paper Presented at the 12th Annual ILO Regional Seminar on Labour Intensive Practices, International Conference Centre, Durban, October. Mengesha, T. 2007. The Baseline and Follow-up Surveys of the Rapid Assessment of Poverty Impacts (RAPI) as Part of the Emergency Recovery Programme. Tigray Rual Road. Final Assessment Report Volume I. ILO Sub-Regional Office Addis Ababa. METZGER, M. 2008. Regional cooperation and integration in Sub-Saharan Africa. UNCTAD Discussion Paper, 189, 42. MILIMO, J. T. 1995. An Analysis of Qualitative Information on Agriculture from beneficiary assessments, participatory poverty assessment and other studies which used qualitative research methods. MILIMO, J. T. 1997. Participatory Poverty Monitoring III. Lusaka: A Report by the Participatory Assessment Group. MILIMO, J. T., SHILITO, T. & BROCK, K. 2002. The poor of Zambia speak. Who would ever listen to the poor? . MINISTRY OF FINANCE AND NATIONAL PLANNING 2004. ZAMBIA First PRSP Implementation Progress Report January 2002 – June 2003. Lusaka: MINISTRY OF FINANCE AND NATIONAL PLANNING. MINTEN, B. & KYLE, S. 1999. The effect of distance and road quality on food collection, marketing margins, and traders' wages: evidence from the former Zaire. Journal of Development Economics, 60, 467-495. MITCHENER, K. J. & MCLEAN, I. W. 2003. The Productivity of US States since 1880. Journal of economic growth, 8, 73. Moffitt, R. 1993, Identification and estimation of dynamic models with a time series of repeated crosssections, Journal of Econometrics 59(1), 99–124. MORRISON, C. J. & SCHWARTZ, A. E. 1996. State Infrastructure and Productive Performance. American Economic Review, 86, 1095-1111. MOURMOURAS, I. A. & LEE, J. E. 1999. Government spending on infrastructure in an endogenous growth model with finite horizons. Journal of Economics and Business, 51, 395-407. MOYO, D. 2009. Dead Aid. Why Aid Is Not Working and How There Is a Better Way for Africa. MOYO, G. I. & MOYO, G. 1996. Introducing a Living Condition Monitoring Survey: The Case of Zambia. In: GULLØY, E. & WOLD, B. K. (eds.) Statistics for Development. Policy and Democracy. Successful Experience and Lessons Learned through 10 years of statistical and institutional development assistance and cooperation by Statistics Norway. Statistics Norway Development Co-Operation. MU, R. & VAN DE WALLE, D. 2007. Rural roads and poor area development in Vietnam. Policy Research Working Paper Series, 4340. MUCHANGA, A. 2001. Statement delivered by deputy permanent secretary ministry of foreign affairs at the twelfth session of the high level committee on the review of technical cooperation among developing countries. MUZIRA, T. & MANJE, L. 2006. Annual Report 2006: Business Information Services for developing rural commmodity and services markets. In: ILO (ed.). Lusaka. NADIRI, M. I. & MAMUNEAS, T. P. 1994. Infrastructure and Public R&D Investments, and the Growth of Factor Productivity in US Manufacturing Industries. NBER Working Papers, 4845. NANNICINI, T. 2007b. A Simulation-Based Sensitivity Analysis for Matching Estimators. Stata Journal, 17. NATIONAL ROAD FUND AGENCY (NRFA). 2010a. National Road Fund Agency: About Us [Online]. Lusaka. Available: http://www.nrfa.org.zm/about_us.htm http://www.nrfa.org.zm/statistical_information.htm [Accessed 4th of June 2010]. 372 NATIONAL ROAD FUND AGENCY (NRFA). 2010b. Overview of ROADSIP II 2004 - 2013 [Online]. Lusaka. Available: http://www.nrfa.org.zm/roadsip.htm [Accessed 4th of June 2010]. Naudé, W. and Matthee, M. 2007. The Significance of Transport Costs in Africa. UNU Policy Brief 05/2007. Nogales, A. 2008. Lifecycle Costs of Roads Under Alternative Maintenance Scenarios. AICD, Working Paper, World Bank, Washington, D.C. NORTH, D. C. 1990. Institutions, Institutional Change and Economic Performance. NSEMUKILA, B. 2001. Poverty and food security indicators in Zambia: Analysis of household survey data. Strengthening Food and Agricultural Statistics in Africa in Support of Food Security and Poverty Reduction Policies and Programmes. Pretoria, South Africa.: FAO. NUNN, N. 2005. Relationship-Specificity, Incomplete Contracts and the Pattern of Trade. η‘ζeill, θ., Shinok θark, Hudayberdi Ahmedov. β010. η0 Years of IDA Assistance to Transport Period: FY 1961-2010. O'ROURKE, K. H. & WILLIAMSON, J. G. 2005. From Malthus to Ohlin: Trade, Industrialisation and Distribution Since 1500. Journal of economic growth, 10, 5. OECD 2008. Turning African Agriculture Into a Business: A Reader. In: DEVELOPMENT CENTRE (ed.). Paris: OECD. OECD, AFDB & UNECA 2009. African Economic Outlook 2009: Innovation in Information Technologies and Communications. In: CENTRE, O. D. (ed.). Paris. OECD, AFDB & UNECA 2010. African Economic Outlook 2010: Public resource mobilization and aid in Africa. In: CENTRE, O. D. (ed.). Paris. OMAMO, S. W. 1998. Farm-to market transaction costs and specialization in small-scale agriculture: explorations with a non-separable household model. Journal of Development Studies, 35, 152-163. OYEJIDE, A. T. 2000. Policies for Regional Integration in Africa. AfDB Economic Research Papers, 62. Paris21. National strategy for the development of statistics knowledge base in Zambia. Accessed at http://www.paris21.org/pages/knowledge-base/national-papers/index.asp?lettre=Z PEDRONI, P. & CANNING, D. 2008. Infrastructure, Long-Run Economic Growth And Causality Tests For Cointegrated Panels. Manchester School, University of Manchester, , 76, 504-527. PEDERSEN, P.O. 2000. The changing structure of transport under trade liberalisation and globalization and its impact on african development. CDR Working Paper No.00.1 PEGUY, P.-Y.1998. The central theme of development policies. PELLENBARG, P. H., WISSEN, L. J. G. V. & DIJK, J. V. 2002. Firm relocation: state of the art and research prospects. Research Report 02D31, University of Groningen, Research Institute SOM (Systems, Organisations and Management). PEROTTI, R. 1992. Income Distribution, Politics, and Growth. American Economic Review, 82, 311-16. PERSSON, T. & TABELLINI, G. 1994. Is inequality harmful for growth? American Economic Review, 84, 600-621. PINKOVSKIY, M. & SALA-I-MARTIN, X. 2010. African Poverty is Falling...Much Faster than You Think! [Online]. [Accessed]. PLATTEAU, J.-P. 1996. Physical Infrastructure as a Constraint on Agricultural Growth: The case of SubSaharan Africa. Oxford Development Studies,, 24, 189-219. PLETCHER, J. 2000. The Politics of δiberalizing Zambia‘s εaize εarkets. World Development Vol. β8, No. 1, pp. 129-142. . PORTER, G. 2002. Living in a Walking World: Rural Mobility and Social Equity Issues in Sub-Saharan Africa. World Development, 30, 285-300. POULTON, C., KYDD, J. & ANDREW DORWARD, A. 2004. Overcoming Market Constraints to ProPoor Agricultural Growth in Sub Saharan Africa In: LONDON, I. C. (ed.) November 2004 ed.: Paper prepared for the Africa Commission RADELET, S. 2001. Regional Integration and Cooperation in Sub-Saharan Africa: Are Formal Trade Agreements the Right Strategy? Journal of African Finance and Economic Development, 4-1. Randall A., 2006. The Babeldaob Road: The Impact of Road Construction on Rural Labor Force Outcomes in the Republic of Palau, IZA Discussion Papers 2452, Institute for the Study of Labor (IZA). RAGHURAM, R. & ZINGALES, L. 1998. Financial Dependence and Growth. American Economic Review, 88, 559-586. RAVALLION, M. 1991. Reaching the Rural Poor through Public Employment. Arguments, Evidence and Lessons from South Asia. The World Bank Research Observer, 6, 153-175. 373 RAVALLION, M. 1992. Poverty Comparison - A Guide to Concepts and Methods. World Bank - Living Standards Measurement, 88, 123. RAVALLION, M. 2004. Pessimistic on poverty? Economist.com. From The Economist print edition ed. Ravallion, Martin, 2008. "Evaluating Anti-Poverty Programs," Handbook of Development Economics, Elsevier. RAVALLION, M. 2010. Is African poverty falling? . Martin Ravallion's blog. [Online]. Available: http://blogs.worldbank.org/africacan/is-african-poverty-falling [Accessed March 2010]. RAVALLION, M. & BIDANI, B. 1994a. How robust is a poverty line? World Bank Economic Review, 8, 75-102. RAVALLION, M. & BIDANI, B. 1994b. How Robust Is a Poverty Profile? . World Bank Economic Review, 8, 75-102. RAVALLION, M. & JALAN, J. 1996. Growth divergence due to spatial externalities. Economics Letters, 53, 227-232. REYNOLDS, L. G. 1985. Economic Growth in the Third World, 1985-1980, New Haven and London, Yale University Press. RIVERSON, J., GAVIRIA, J. & THRISCUTT, S. 1991. Rural Roads in Sub-Saharan Africa: Lessons from World Bank Experience. Technical Paper, 141. ROBINSON, P. M. 1988. Root-N-Consistent Semiparametric Regression. Econometrica, 56, 931-954. ROBINSON, R. & STIEDL, D. 2001. Decentralization of Road Administration: Case Studies in Africa and Asia. Public Administration and Development, 21, 53-64. RODRIGUEZ, F. & RODRIK, D. 2001. Trade Policy & Economy Growth: a Skeptic's Guide to the CrossNational Evidence. NBER Macroeconomics Annual 2000, 15, 261-338. RODRIK, D. 1995. Trade Strategy, Investment and Exports: Another Look at East Asia. CEPR Discussion Papers, 1305. RODRIK, D., SUBRAMANIAN, A. & TREBBI, F. 2004. Institutions Rule: The Primacy of Institutions Over Geography and Integration in Economic Development. Journal of economic growth, 9, 131. ROEMER, M. & STERN, J. 1975. The Appraisal of Development Projects: A Practical Guide to Project Analysis with Case Studies and Solutions, New York, Praeger Publishers. ROMER, P. M. 1986. Increasing Returns and Long-Run Growth. Journal of Political Economy, 94, 10021037. ROMER, P. M. 1990. Endogenous Technological Change. Journal of Political Economy, 98, 71-102. ROSENBAUM, P. R. & RUBIN, D. B. 1983. The central role of the propensity score in observational studies for causal effects Biometrika, 70, 41-55. ROSENZWEIG, M. R. & WOLPIN, K. I. 2000. Natural .Natural Experiments in Economics. Journal of Economic Literature, 38(4), 827-874. Ruijs, A., & Schweigman, Caspar & Lutz, Clemens, 2004. The impact of transport- and transaction-cost reductions on food markets in developing countries: evidence for tempered expectations for Burkina Faso. Agricultural Economics, Blackwell, vol. 31(2-3), pages 219-228, December. RUSIKE, J., HOWARD, J. & MAREDIA, M. 1997. Seed sector evolution in Zambia and Zimbabwe: has farmer access improved following economic reforms? . Policy Syntheses, 31. RWAMPORORO, R. K., TOURNÉE, J. M. & CHITEMBO, A. 2002. Joint Final Evaluation. District Development Planning and Implementation in Eastern Province (ZAM/93/CO1) and Rehabilitation and Maintenance of Feeder Roads (ZAM/93/CO2 & ZAM/93/009). In: UNCDF (ed.). SAASA, O. S. Year. Rural Development Projects and Popular Participation in Zambia. Some Methodological Issues. In: SAASA, O. S., ed. Institute for African Studies (IAS) Seminar Proceedings No.1., 1990. SACHS, J. 2009. Globalization in the Era of Environmental Crisis. UNCTAD/PRESS/PR/2009/053. 14th Prebisch Lecture. SACHS, J., MCARTHUR, J. W., SCHMIDT-TRAUB, G., KRUK, M., BAHADUR, C., FAYE, M. & εωωηRD, G. β004. Ending Africa‘s θoverty Trap. ψrookings θapers on Economic Activity, 1:2004 SACHS, J. D. & WARNER, A. M. 1995. Natural Resources and Economic Growth. HIID Development Discussion Paper, 517a. SADC 1992. Declaration and Treaty of SADC. 374 SADC 2001. Consolidated Text of the Treaty of the Southern African Development Community, as amended. . In: SADC (ed.). SADC. 2005. Transport [Online]. Available: http://www.sadcreview.com/directorate_reports/frreports.htm [Accessed]. SALA-I-MARTIN, X. 1997. I Just Ran Two Million Regressions. American Economic Review, 87, 178183. SALA-I-εARTIζ, X. β004. The World Distribution of Incomeμ Falling θoverty and… ωonvergence, Period. 64. SCHULZ, R. & BENTALL, P. 1998. Rehabilitation and Maintenance of Feeder Roads (Eastern Province). Project Evaluation Summaries – 1998. Prepared by the Policy, Planning and Evaluation Unit (PPEU). Project Number: ZAM/93/C02. SIANESI, B. 2001. Implementing Propensity Score Matching Estimators with STATA. University College London and Institute for Fiscal Studies. . SRINIVASAN, T. N. & BHAGWATI, J. 1999. Outward-orientation and development: are revisionists right? for Professor Anne Krueger festchrift. STERN, E. 2005. Evaluation Research Methods, SAGE Publications Ltd. STIROH, K. J. 2000. Investment and Productivity Growth - A Survey from the Neoclassical and New Growth Perspectives. Gouvernement du Canada - Industry Canada & Gouvernement du Canada Industry Canada., 24, 62. SUNDRUM, R. M. 1992. Income distribution in less developed countries. TABELLINI, G. 2005. Culture and institutions: economic development in the regions of Europe. TATOM, J. 1991. Public Capital and Private Sector Performance. St. Louis Federal Reserve Bank Review, 3-15. Taylor, G., McCord, A. and van Seventer, D. 2006. Limpopo Province Labour Intensive Rural Roads Maintenance Programme (Gundo Lashu) Cost Comparison Report. Draft 6. United Kingdom: Department for International Development, Unpublished. March. Taylor, G., Zewdu, G.A., Feleke, B., Ahmed, F. 2008. Comparative Study on the Impact of Labour-Based and Equipment-Based Methods in Roadworks in Ethiopia. Ethiopian Ministry of Labour & Social Affairs, Ethiopian Roads Authority, SIDA and ILO. Te Velde, D.W. 2005. Globalisation and education. What do the trade, investment and migration literatures tell us? ODI. August. TE VELDE, D. W. & MEYN, M. 2008. Regional Integration in African, Caribbean and Pacific countries A review of the literature. Tembo, S. and F.Blokhuis. 2004. Manual for supervision of labour based road rehabilitation works. Prepared for ILO ASIST. TEMPLE, J. 1999. The New Growth Evidence. Journal of Economic Literature, XXXVII, 112-156. THOMAS, V. 2008. Impact evaluation: Its status and its future. 7th meeting of the DAC Network on Development Evaluation. Independent Evaluation Group (IEG) of the World Bank. THURLOW, J. & WOBST, P. 2004b. The road to pro-poor growth in Zambia: past lessons and future challenges. 123. THURLOW, J. & WOBST, P. 2005. Operationalizing Pro-Poor Growth. The Department for International Development. TINBERGEN 1957. The Appraisal of Road Construction: Two Calculation Schemes. The Review of Economics and Statistics, 39, 241-249. TRESSEL, T. 2003. Dual Financial Systems and Inequalities in Economic Development. Journal of economic growth, 8, 223. TSCHIRLEY, D. & KABWE, S. 2007. Cotton in Zambia: 2007 Assessment of its Organization, Performance, Current Policy Initiatives, and Challenges for the Future. Working Paper No. 26. . TSCHIRLEY, D. & KABWE, S. 2007a. Cotton in Zambia: 2007 Assessment of its Organization, Performance, Current Policy Initiatives, and Challenges for the Future. FSRP Working paper 26. TSCHIRLEY, D. & ZULU, B. 2003. Zambian Cotton in a Regional Context: Performance under Liberalization and Future Challenges. FSRP Working paper, 7. TURNOVSKY, S. J. & FISHER, W. H. 1995. The composition of government expenditure and its consequences for macroeconomic performance. Journal of Economic Dynamics and Control, 19, 747-786. 375 Ulimwengu, John & Funes, Jose & Headey, Derek & You, Liangzhi, 2009. Paving the way for development?: The impact of transport infrastructure on agricultural production and poverty reduction in the Democratic Republic of Congo, IFPRI discussion papers 944. UNCTAD 1999b. UNCTAD's contribution to the implementation of the United Nations new agenda for the development of Africa in the 1990s: African transport infrastructure, trade and competitiveness. Trade and Development Board, Forty-sixth session. Geneva: UNCTAD. UNCTAD 2002a. The Least Developed Country Report 2002. Escaping the Poverty Trap. UNCTAD. UNCTAD 2004. The Least Developed Countries Report 2004. Linking International Trade with Poverty Reduction. United Nations. UNCTAD 2006. Investment Policy Review Zambia. Geneva: United Nations. UNCTAD 2009a. Economic Development in Africa. Report 2009. Strengthening Regional economic integration for Africa's development. In: DIVISION, A. (ed.). Geneva: United Nations. UNDP 2002. PART I : Evaluation Report Governance: Enhanced Local Governance for Poverty Reduction Project Number: ZAM/02/002/01/99/A. UNDP 2008. What will Zambia look like when the MDGs are achieved? UNECA 2000. Pre-feasibility Study of the Transport Component of the Zambia-Malawi-Mozambique Growth Triangle Initiative of the United Nations Development Programme. UNECA 2001a. Report of the proceedings of a roundtable on partnership among the public sector, private sector and civil society. Lusaka, Zambia. UNECA 2001b. Partnership Building for Subregional Cooperation and Economic Integration in Africa: The Case of Zambia-Malawi-Mozambique Growth Triangle. SRDC-SA Development Bulletin Issue, 9. UNECA. Year. Aide Memoire. In: Workshop on Growth Strategies for Poverty Reduction in Southern Africa, 15-17 December, 2003 2003 Lusaka. United Nations Economic Commission for Africa, Southern Africa Office. UNECA 2005. The Statistical Estimation of Poverty Duration and Transitions in Zambia. 38. VAMVAKIDIS, A. A. 2002. How Robust is the Growth-Openness Connection? Historical Evidence. Journal of economic growth, 7, 57. Van de Walle, D. 2009. Impact evaluation of rural road projects. Journal of Development Effectiveness, Vol. 1, No.1, March, 15-36. van de Walle, D. 2003. Are Returns to Investment Lower for the Poor? Human and Physical Capital Interactions in Rural Vietnam. Review of Development Economics, vol. 7(4), pages 636-653, November. VAN DE WALLE, D. 2002. Choosing Rural Road Investments to Help Reduce Poverty. World Development, 30, 575-589. Van de Walle, D., & C., Dorothyjean, 2005. Do donors get what they paid for? micro evidence on the fungibility of development project aid. Policy Research Working Paper Series 3542, The World Bank. Van de Walle, D. & Mu, Ren, 2007. Fungibility and the flypaper effect of project aid: Micro-evidence for Vietnam. Journal of Development Economics, Elsevier, vol. 84(2), pages 667-685, November. VAN DER HOEVEN, R. 2008. Income Inequality Revisited: Can one bring sense back into economic policy? Inaugural Address as Professor of Employment and Development Economics. Haague. VAN DIJK, J. & PELLENBARG, P. H. 2000. Firm relocation decisions in The Netherlands: An ordered logit approach. Papers in Regional Science, , 79, 191-219. VAN LEEUWEN, B. 2004. Literature on the relation between human capital and economic growth: definitions and problems. VENABLES, A. J. & LIMAO, N. 2002. Geographical disadvantage: a Heckscher-Ohlin-von Thunen model of international specialisation. Journal of International Economics, 58, 239-263. Verbeek, M., and F. Vella. Estimating Dynamic Models from Repeated Cross-Sections. Journal of Economics 127 (2005) 83-102. Elsevier. Verbeek, M.: 1996, Pseudo Panel Data. Chapter 11 in: Matyas, L. and Sevestre, P.(eds.): The Econometrics of Panel Data. A Handbook of the Theory with Applications. Verbeek, M. and Nijman, T.: 1993, Minimum mse estimation of a regression model with fixed effects from a series of cross-sections, Journal of Econometrics 59(1-2), 125–136. Verbeek, M. and Nijman, T.: 1992a, Can cohort data be treated as genuine panel data?, Empirical Economics 17(1), 9–23. 376 Verbeek, M. & Nijman, T., 1992b. Incomplete Panels and Selection Bias: A Survey. Papers 9207, Tilburg Center for Economic Research VIJVERBERG, W., VIJVERBERG C. & GAMBLE, J. 1997. Public Capital and Private Productivity. Review of Economics and Statistics, 267-278. VISSER, M. & HARTZENBERG, T. 2004. Trade Liberalization and regional integration in SADC: Policy synergies assessed in an industrial organization framework. Working Paper No. 3 Trade Law Centre for Southern Africa (TRALAC), Stellenbosch. VOITCHOVSKY, S. 2005. Does the Profile of Income Inequality Matter for Economic Growth? Journal of economic growth, 10, 273. VON BRAUN, J. 1995. Employment for Poverty Reduction and Food Security. IFPRI Occasional Papers. IFPRI. VON BRAUN, J., TEKLU, T. & WEBB, P. 1992. Labour-intensive public works for food security in Africa: Past experience and future potential. International Labour Review, 131, 19-33. WACZIARG, R. 1998. Measuring the dynamic gains from trade. Policy Research Working Paper Series 2001. Warr, P. 2005. "Roads and Poverty in Rural Laos," Departmental Working Papers 2005-04, Australian National University, Economics RSPAS. WEIL, D. N. 2010. Endemic Diseases and African Economic Growth: Challenges and Policy Responses. Journal of African Economies Wiggins, S. and θroctor, S. β001. ‗How Special Are Rural Areas? The Economic Implications of Location for Rural Development‘, Development θolicy Review, 19 (4), 4β7–436 Irz, X., δin, δ., Thirtle, ω. and Wiggins, S. β001. ‗Agricultural Growth and θoverty Alleviation‘, Development Policy Review, 19 (4), 449–466 Wiggins, S. 2000. ‗Interpreting changes from the 1970s to the 1990s in African agriculture through village studies‘, World Development, β8 (4), θγ1-662 WILSON, F. 2004. Towards a Political Economy of Roads: Experiences from Peru. Development and Change, 35, 525-546. WINTERS, L. A. 2000a. Trade Liberalisation and Poverty. Poverty Research Unit at Sussex, University of Sussex, Discussion Paper 7. WINTERS, L. A. 2000b. Trade, Trade Policy and Poverty: What are the Links? Centre for Economic Policy Research Paper No.2382. WINTERS, L. A. 2002a. Trade Liberalisation and Poverty: What are the Links? . World Economy, 25, 1339-1367. WINTERS, L. A., MCCULLOCH, N. & MCKAY, A. 2004. Trade Liberalization and Poverty: The Evidence So Far. Journal of Economic Literature, American Economic Association, 42, 72-115. WOOLDRIDGE, J. 2002. Econometric Analysis of Cross-Section and Panel Data,, Cambridge, Mass., MIT Press. WORLD BANK 1994. World Development Report 1994: Infrastructure for Development In: BANK, W. (ed.) World Development Report. Washington, D.C.: World Bank. WORLD BANK. 2005a. The Africa Action Plan. Planning for the "Decade of Africa" [Online]. Available: http://web.worldbank.org/WBSITE/EXTERNAL/COUNTRIES/AFRICAEXT/0,,contentMDK:20 687937~menuPK:258649~pagePK:146736~piPK:226340~theSitePK:258644,00.html [Accessed 29th of October 2005]. WηRδD ψAζK β00ηb. ηED and impact evaluation―a discussion note. ηperations Evaluation Department, World Bank. WORLD BANK. 2005c. World Development Report 2006: Equity and Development [Online]. Available: http://web.worldbank.org/WBSITE/EXTERNAL/EXTDEC/EXTRESEARCH/EXTWDRS/EXT WDR2006/0,,menuPK:477658~pagePK:64167702~piPK:64167676~theSitePK:477642,00.html [Accessed]. WORLD BANK 2005d. Zambia: Poverty and Vulnerability Assessment. In: REGION, A. (ed.). World Bank. WORLD BANK 2007a. World Development Report, 2008. Agriculture for Development. Washington, DC.: The World Bank,. WTO 1996. Trade Policy Review. Zambia. In: DIVISION, T. P. R. (ed.). Geneva: WTO. WTO 2002. Trade policy review: Zambia 2002. In: DIVISION, T. P. R. (ed.). Geneva: WTO. WTO 2010. AGRICULTURE: The Cotton Sub-Committee. WTO website accessed September 2010. 377 YATCHEW, A. 1998a. Nonparametric regression techniques in economics. Journal of Economic Literature, 36, 669-721. YATCHEW, A. 2005. Semiparametric Regression for the Applied Econometrician, Cambridge University Press. YOUNG, A. 1994. The Tyranny of Numbers: Confronting the Statistical Realities of the East Asian Growth Experience. YOUNG, A. 1998. Alternative Estimates of Productivity Growth in the NICs: A Comment on the Findings of Chang-Tai Hsieh. NBER Working Paper, 6657, 63. ZAMBIA FOOD SECURITY RESEARCH PROJECT 2000. Improving smallholder and agri-business opportunities in Zambia‘s cotton sectorμ Key challenges and options. δusaka, Zambia. ZAMBIA VULNERABILITY ASSESSMENT COMMITTEE 2004. Zambia Livelihood Map Rezoning and Baseline Profiling. Lusaka. ZHOUA, J., ZHUB, Z. & FUNGC, W. K. 2008. Robust testing with generalized partial linear models for longitudinal data. Journal of Statistical Planning and Inference, 138, 1871-1883. ZULU, B. & TSCHIRLEY, D. 2002. An Overview of the Cotton Sub-Sector in Zambia. 378 Annexes 379 Annex: Chapter 1 Figure A1: A proposed Scheme for the Evaluation of Economic Growth Benefits from Transportation Investment Infrastructure Investment Investment multiplier Travel Effects: Network accessibility Welfare gaings Primary Benefits: Travel time and costs Traffic volume Activity spatial redistribution Externalities Allocative Externalities Pecuniary Environmental Transport network economies Relative prices And land rent Economic Growth Source: Banister & Berechman, 2000, p.173. 380 Labour Market Agglomeration: Firm‘s cost reduction Spatial and organizational changes Figure A2: Flow chart of the relationships among road investment, input and output markets, and household and intra-household outcomes Source: Khandker et al., 2006, p.34. 381 Annex: Chapter 2 Table A1: Successful growth experiences: Pace and quality of economic growth, 1999-2007 No. Country Successful Achievement Annual real GDP growth 1. Mozambique Achieving shared growth in post-stabilization 7.7% GDP per Capita at constant 2000 Prices (USD) 348 2. Tanzania Successful reformer: Transformation to an open market economy 6.5% 403 3. Uganda A decade of strong growth, but limited economic transformation 5.8% 273 4. Burkina Faso Beginning to diversify its highly cotton-dependent economy 5.6% 268 5. Mauritius Adapting to a changing world 4.1% 4649 6. Botswana Sustained economic progress through prudent macroeconomic management, institutional development, and good governance 5.8% 4439 Table A2: Institutions and governance, 1999-2007 No. Country 7. 8. 9. Successful Achievement Annual GDP per Capita at real GDP constant 2000 growth Prices (USD) 8.9% 237 Sierre Rebuilding local governments in post-conflict Leone Liberia Promoting fiscal accountability and transparency in post-conflict economies: Governance and Economic Management Assistance Program (GEMAP) South Political transformation and sound macroeconomic management: Africa laying the foundation for sustained growth and development 10. Nigeria Fiscal decentralization: Lagos State‘s progress in strengthening public finance management and service delivery 5.3% 130 4.0% 3657 4.9% 446 Table A3a: Competitiveness and export dynamism, 1999-2007 No. Country Successful Achievement Annual real GDP per Capita GDP at constant 2000 growth Prices (USD) Agri-business 11. Mali 12. Ghana Exporting fruits: Mangoes 4.9% 322 Bananas and pineapples 5.1% 306 5.6% 264 5.8% 273 3.8% 456 4.6% 385 3.7% 642 Cocoa sector is showing impressive growth as supply has responded to policy reforms, which allow a larger pass through of world cocoa prices to producers 13. Rwanda 14. Uganda 15. Kenya 16. Zambia Reforms have transformed coffee sector and boosted exports Scaling up quantity and quality: success of the fishprocessing industry Success in increasing its global market share of cut flowers Path to developing the cotton sector 17. Cameroon Path to developing the cotton sector Sources: Authors Adapted from World Bank “Africa Can ... End Poverty” website and ηEωD & AfDψ‘s African Economic Outlook Report 2008. 382 Table A3b: Competitiveness and export dynamism, 1999-2007 No. Country 18. Ethiopia 19. Lesotho 20 REC(*) 21. REC(*) Successful Achievement Annual real GDP growth GDP per Capita at constant 2000 Prices (USD) 5.8% 141 3.2% 550 Manufacturing Improvements in product quality, marketing and management are helping to re-establish and grow the footwear industry. Responding to trade opportunitiesμ The case of δesotho‘s apparel industry Leveraging regional markets to build Africa's domestic manufacturing sector Indigenous African financial services companies: Venturing into the regional market Notes: (*) Regional Economic Communities. Table A3c: Competitiveness and export dynamism, 1999-2007 No. Country Successful Achievement Annual real GDP per Capita GDP at constant 2000 growth Prices (USD) Tourism 21. Rwanda 22. Cape Verde Leveraging gorilla tourism for development 5.6% 264 Tourism has played a key role in the country‘s successful graduation to middle income status 7.0% 1602 Table A3d: Competitiveness and export dynamism, 1999-2007 No. Country Successful Achievement Annual real GDP per Capita GDP at constant 2000 growth Prices (USD) Services 22. Senegal 23. Uganda An early reformer: Privatization of its telecommunications operator Sonatel Uganda-Rural development fund: Uganda is benefiting from having one of the most liberal ICT markets in the region 4.2% 497 5.8% 273 Table A4: Agriculture and rural development, 1999-2007 No. Country Successful Achievement 24. Malawi Raising yields: The case for fertilizer subsidies Annual real GDP growth 3.1% GDP per Capita at constant 2000 Prices (USD) 158 Table A5: Microfinance providing access & financial products for underserved populations, 1999-2007 No. Country Successful Achievement 25. Kenya 26. Ethiopia 27. Burkina Faso Reaching rural markets: Equity Building Society of Kenya Kenyan Finance Women's Trust—bringing financial services to women Reaching rural markets: Ethiopia's Amhara Credit and Savings Institution Providing financial services to rural populations through financial cooperatives - the case of RCPB Annual real GDP growth 3.8% GDP per Capita at constant 2000 Prices (USD) 456 5.8% 141 5.6% 268 Sourcesμ Authors Adapted from World ψank ―Africa ωan ... End θoverty‖ website and ηEωD & AfDψ‘s African Economic Outlook Report 2008. 383 Table A6: Infrastructure - improving efficiency and leveraging the private sector Information communication technology, 1999-2007 No. Country 28. Africa Successful Achievement ωonnecting a continentμ Africa‘s mobile success story Annual GDP per Capita real GDP at constant 2000 growth Prices (USD) 4.7% 848 Transport 29. Africa Road funding institutional schemes implemented across African countries are showing some steady improvements in road quality associated with more management oriented financing. 4.7% 848 30. Nigeria δagos ωity‘s ψus Rapid Transit system—the first of its kind in subSaharan Africa, and the first example of a comprehensive and integrated approach to improving public transport—is providing clean, safe, and reliable public transport 4.9% 446 31. Mali Expanding rural electrification through an adaptive and multilayered approach that mobilizes local private sector operators and community organizations in the delivery of energy services 4.9% 322 Connecting rural populations to water supply: Making progress on the water MDG; Progress in expanding well/borehole coverage for rural populations 5.8% 141 5.8% 273 34. Madagascar The Rural Water Supply and Sanitation Pilot Project in Madagascar brings ―eau pour tous‖ and revives a neglected sector 3.8% 246 5.6% 264 1.2% 166 4.6% 385 4.3% 2,246 5.8% 4,439 2.3% 93 2.9% 400 3.9% 167 Access to safe water 32. Ethiopia 33. Uganda Improving health and education outcomes 35. 36. 37. 38. 40. Rwanda Scaling up performance incentives to improve health delivery Therapeutic coverage increased from 1% in 2003 to almost 71% in 2007, aided by a 40-fold growth in the number of antiretroviral treatment sites. Eritrea Controlling malaria, saving lives: Has dramatically lowered the incidence of malaria—malaria deaths were 80% lower in β00θ compared to 2001. Zambia The Zambia Malaria Booster Project has contributed to reducing malaria cases and deaths—malaria cases declined by γ1% and malaria deaths by 37% between 2006-08. Namibia Antiretroviral therapy coverage has risen from under 1% in 2003 to 88% in 2007. Botswana The coverage rate was around 80%. 41. DRC 42. Guinea 43. Niger The treatment adherence rates in some conflict-affected areas are comparable with those reported in non-conflict settings. Raising primary school completion rates: Guinea has more than doubled primary school completion rates since 2000. Niger has more than doubled primary school completion rates since 2000. Sources: Authors Adapted from World Bank ―Africa ωan ... End θoverty‖ website and ηEωD & AfDψ‘s African Economic Outlook Report 2008. 384 Table A7: Impact Assessment in EIIP/ASIST: Summary of Selected Cases COUNTRY Type of works (1) / Examined sectors (2) Botswana Public works and road development Ghana Public works and road development Author/Date of publication ILO and Impact rating of evidence vis a vis poverty and reference employment – the gaps identified Republic of Botswana and Public Roads Administration Directorate of Socio-economic Impact Study Labour-based Public Roads, Report No. IC 007: Road Maintenance Demonstration Project October 2002; Ref. BWA 110332 Useful but long term socio-economic impacts not Socio-Economic ImpactStudiesof Non-Gazetted demonstrated with the data analysed. Methodology Roads in Botswana: Objectives and applied in study not focused enough to link adequately ILO, Geneva, 1984 Methodology EII to poverty Title of study Study of the Social and Economic Impact of Feeder Roads Improvement - Technical Proposal Department of Planning, University Insufficient analysis and material available overall, of Science and Technology, Kumasi, without sufficient depth of analysis of poverty and the Ghana/1987 relationship between EII and poverty impacts Socio-economic impact study - Baseline Report Ghanexim/1989 Socio-economic impact study - Baseline Report Ghanexim/1989 Socio-economic impact study - Immediate Impact Report Ghanexim/1989 Kenya Public works and road development Socio-Economic Impact Study of feeder roads improvements using LBT - Final Report Ghanexim/1990 The reports are valuable for the purposes for which they were commissioned but their usefulness in terms of providing evidence of relationship between EII and poverty is not demonstrated adequately Assessment of the socio-economic impacts of the Kenya Rural Access Roads Programme Ministry of Transport and Communication, Nairobi/1984 Fair but limited focus on poverty issues and long term socio-economic effects Baseline Survey for the Kenya Minor Roads Programme - Phase 1: Survey design and methodology Fair but not sufficient basis for carrying out impact The Management Centre and studies on poverty and long-term socio-economic Economic Consultants, Nairobi, 1988 impacts Fair but not adequate analysis of long-term socioeconomic impacts and links between EII and poverty Baseline Survey - Phase II: Survey Results, Vol The Management Centre and I and II Economic Consultants, Nairobi, 1988 unclear Socio-Economic Impact Evaluation of Kenya The Management Centre and Fair but not linked well to previous baseline studies Minor Roads Programme SIDA funded districts Economic Consultants, Nairobi, 1991 because of deficient baseline methodology Socio-Economic Impact Evaluation of Kenya The Management Centre, 1989, Minor Roads Programme SIDA funded districts 1990/91 Fair but not linked well to previous baseline studies because of deficient baseline methodology Comprehensive but analysis of poverty and marginalized groups insuffficiently done A good account of the Impacts of Feeder Roads Programmes and its macro links No systematic data collection to eanble impact studies to link EII to poverty - information and analysis gap exists Mozambique Public works and road development Impact Study of the Minor Roads Programme in Nyanza Province, Kenya, Phase 1. Baseline COWI/DANIDA, 1988 Study UNDP in association with SIDA, ILO Lessons Learned from Feeder Road Programme and ANE Namibia Public works and road development A baseline Socio-Economic Survey, Onaanda Community-Based Road construction Project Western Owambo Region Andrew Botelle, Namibian Institute for Social and Economic Research (NISER), 1991, 1992 Impact Assessment of LBT Application in the Provision of Infrastructural Services, Vol. 1-4 Ministry of Works, Transport and Communication, 2000 Managing the Sustainable Growth and Development of Dar es Salaam, Hanna Nassif Settlement, Baseline Data United Nations Centre for Human Settlements (HABITAT), 1995 Baseline Study and Mid-term Impact Assessment, Hanna Nassif Community Based Infrastructure Upgrading Phase II University College of Lands and Comprehensive but analysis of poverty and socioArchitectural Studies (UCLAS), ILO, economic impacts in marginalized groups not 1999 demonstrated well Tanzania South Africa Infrastructure Upgrading Public works and road development Good Analysis of Immediate Effects but not long-term impacts Comprehensive but more analysis on poverty needed. Also a failure to link initial baseline studies to subsequent impact assessment due to methodological deficiencies Independent Evaluation Hanna Hassif Community based Settlement Upgrading Phase II Kinondori District, Dar-es-Salaam IT Transport/UCLAS/ILO, 2001 Gary Taylor, Angela Bester and Peter Delius, 2003 Output to Purpose Review, Gundo Lashu Comprehensive but analysis of poverty and socioeconomic impacts in marginalized groups not demonstrated well Comprehensive but links between EII and poverty not clear The Economy-wide Impacts of the Labour Anna McCord and Dirk Ernst van Intensification of Infrastructure in South Africa Seventer, 2004 Very well argued case which shows that link between EIIP and poverty is weak unless well targeted The Expanded PWP and Chronic Poverty in John Howe, Maikel Lieuw-Kie-Song, Good critique of McCord putting EIIP in context being South Africa: Inadequate Response or Unique Dr Sean Phillips and Gary Taylor, one among many strategies for employment creation and poverty reduction Opportunity 2005 Source: Stephen Chipika, 2005:44-50. 385 Table A8: Summary of a Sample Key Impact Evaluation Studies Author Mu, R., and van de Walle, D., 2007 Country covered WB financed Rural road rehabilitation project implemented in rural Vietnam between 1997 and 2001. Data The "Survey of Impacts of Rural Roads in Vietnam" consists of a panel of 200 communes and 3000 HHs. The survey design implicitly takes the commune as the project‘s zone of influence. Micro-level survey and paneldata evidence of about 1,200 households spanning 19922000 Method Double difference and matching methods Major Findings Significant average impacts on the development of local markets Deininger, K., Okidi, J., 2003 Uganda They proceed in three stages. First, they estimate determinants of economic growth at the household level. Second, they expand this to a consideration of poverty reduction. Third, they perform simulations. Bangladesh Household-level panel data Jacoby, H.G., 2000 Nepal Nepal Living Standard Surveys Use a household fixedeffects technique to estimate the returns to road investment in terms of its impact on household per capita consumption. A method for nonparametrically estimating the benefits from road projects at the household level Escobal, J. Ponce, C., 2003 Peru Using information from rural households living in some of the poorest districts of Peru The propensity score matching methodology is used, after adapting it to the specific characteristics of the data used. Lokshin and Yemtsow, 2005 Georgia (Rural): Infrastructure rehabilitation projects between 1998 & 2001. 15 Ethiopian Villages, 1994-2004 Community-level panel data from a regular household survey augmented with a special community module Propensity score--matched difference-in-difference comparisons Making use of new the longitudinal household survey data that were not used in earlier Dercon papers. An instrumental variables model using Generalized Methods of Moments and controlling for household fixed effects 15 Ethiopian Villages Data are taken from the Ethiopia Rural Household Survey (ERHS), a unique longitudinal household data set covering households in 15 areas of rural Ethiopia. Data collection started in 1989. The survey was expanded in 1994 to yield a sample of 1,477 households. An additional round was conducted in late 1994, with further rounds in 1995, 1997, 1999, and 2004. Estimate a series of probit regressions. Fixed effect IV regression Access to key public goods such as infrastructure, and the avoidance of civil strife has been a critical determinant of households‘ ability to increase their income and reduce the risk of falling into poverty. Road investments are pro-poor, meaning the gains are proportionately higher for the poor than for the non-poor. Large benefits from extending roads into remote rural areas, much of these gains going to poorer households. But rural road construction is not the magic bullet for poverty alleviation. Rehabilitated road accessibility can be related to changes in income sources, as the rehabilitated road enhances nonagricultural income opportunities, especially from wageemployment sources. Plausible results regarding the size of welfare gains from a particular project at the village level. Access to all-weather roads: Reduces poverty by 6.9 pct points and increases consumption growth by 16.3 percent. These results are robust. An increase of 10 km in the distance from the rural village to the closest market town has a dramatic effect on the likelihood that the household purchases inputs, controlling for the effect of other factors. Increases in road quality have strong positive growth effects Khandker, S.R., Bakht, Z., Koolwal, G.B., 2006 Dercon, S., Gilliagan, D.O., Hoddinott, J., Woldehanna, T., 2008 Dercon and Hoddinott, 2005 Source: Author. 386 Annex: Chapter 3 387 Table A3.1: Rainfall (12-months moving avg.) (mm.), Agricultural season 1994/95 – 2004/2005 Year 1994/1995 1995/1996 1996/1997 1997/1998 1998/1999 1999/2000 2000/2001 2001/2002 2002/2003 2003/2004 2004/2005 Long-term Mean Eastern 528,55 805,71 813,72 719,02 759,38 678,55 914,78 700,40 835,42 781,11 788,61 Chadiza (301) (i) Chama (302) (iii) Chipata (303) (i) Katete (304) Lundazi (305) Mambwe (306) (iii) Nyimba (307) (ii) Petauke (308) (ii) Long-term Mean 610,83 494,08 610,83 214,17 693,58 494,08 555,42 555,42 756,84 868,50 945,33 708,50 901,42 584,83 1165,17 771,92 871,33 915,92 1007,92 756,84 850,15 688,58 708,00 760,00 574,00 695,25 562,58 564,67 685,50 746,00 750,17 657,17 868,50 945,33 708,50 901,42 584,83 1165,17 771,92 871,33 915,92 1007,92 850,15 936,83 826,00 874,75 791,08 711,83 1163,50 846,17 877,47 878,01 895,40 819,56 721,67 609,08 696,58 543,42 563,75 761,58 607,67 815,42 658,37 681,36 668,41 688,58 708,00 760,00 574,00 695,25 562,58 564,67 685,50 746,00 750,17 657,17 836,50 884,00 621,92 894,83 796,33 968,83 738,08 938,42 694,33 608,00 776,06 836,50 884,00 621,92 894,83 796,33 968,83 738,08 938,42 694,33 608,00 776,06 756,84 756,84 756,84 756,84 756,84 756,84 756,84 756,84 756,84 756,84 Notes: We assume the following coverage for the five weather stations: Chipata covers Chipata and Chadiza districts; Lundazi covers Lundazi district; Petauke covers Petauke and Nyimba districts; Msekere covers Katete district; and Mfuwe covers Chama and Mambwe districts. Lundazi and Katete (Msekere) was closed in respectively 2002/20032004/2005 and 2003/2004 - 2004/2005 due to lack of manpower. Hence, the figures in italic have been extrapolated as 5 year moving averages for these same years. Source: Author's calculations based on Zambia Meteorological Service data. Table A3.2a: Rehabilitation and Maintenance of Feeder Roads Eastern Province (EPFRP) Source: Rwampororo et al., 2002:xiii. Table A3.2b: Financial Performance Source: Rwampororo et al., 2002:xiii. 388 Chart A1: The Eastern Province Feeder Road Project Source: Rwampororo et al., 2002:7. Table A3.3: Equipment list procured for rehabilitation contractors (all in USD) Source: Clifton et al., 2001:15. 389 Table A3.4: Summary Fact Sheet of Completed Rehabilitation Contracts as at end of Q4-98 Contract No. Name of Contractor Construction time Extend of Number Road Number Length Km laterite of w idth of (km) Months constructed surf acing culverts (m) culverts per month (10cm) per km Major structures Worker days generated Female Total Male w orker days Worker days % of total Worker days input per km Construction Costs Total ZK Zk per km Total USD Community impact USD per km Total ZK in w ages % of total construction cost Compl eted Contra cts: FTC-1 STC-1/7 STC-2/7 STC-3/7 STC-4/7 STC-5/7 STC-6/7 STC-7/7 KAT/98/01 All contractors Justed Development Mtondo Construction Wheeltrax Contractors Libean Contractors Kaw aye Chataya Camber Contractors Rapid Construction Camber Contractors 20.4 8.0 7.4 6.1 13.0 7.9 6.0 7.5 4.6 4.5 9.0 8.5 6.0 11.0 5.0 4.5 7.5 4.0 4.5 0.9 0.9 1.0 1.2 1.6 1.3 1.0 1.2 80.9 60.0 1.3 5.9 12.1 7.9 4.6 3.8 2.6 5.8 5.0 7.0 5.0 4.0 3.0 3.5 5.0 1.2 1.7 1.6 1.2 1.3 0.7 1.2 Tota l ongoi ng contra cts: 42.7 32.5 Tota l a l l compl eted work s: 123.6 Tota l compl eted contra cts: 5.5 5.5 5.5 5.5 5.5 5.5 4.5 4.5 4.5 100% 93% 100% 100% 52% 100% 100% 96% 100% 53 22 21 8 8 11 8 18 12 2.6 2.8 2.8 1.3 0.6 1.4 1.3 2.4 2.6 161 2.0 29 14 26 9 9 25 8 4.9 1.2 3.3 2.0 2.4 9.7 1.4 1.3 120 1.3 281 1 drif t of 25m none 1 drif t of 30m none 1 drif t of 55m none none none none 42,157 19,983 15,843 11,336 13,818 10,467 9,353 10,129 6,232 10,291 2,042 2,472 121 1,602 267 1,982 3,950 1,972 20% 9% 13% 1% 10% 2% 17% 28% 24% 52,448 22,025 18,315 11,457 15,420 10,734 11,334 14,079 8,204 2,571 2,753 2,483 1,878 1,186 1,359 1,889 1,877 1,783 259,473,040 106,569,737 120,761,756 79,980,730 110,662,077 94,201,798 60,830,961 91,152,249 71,856,904 12,719,267 13,321,217 16,374,475 13,111,595 8,512,467 11,924,278 10,138,494 12,153,633 15,621,066 200,675 65,780 80,170 53,064 69,348 62,995 41,683 60,305 38,471 9,837 8,223 10,871 8,699 6,799 7,974 6,947 7,832 8,363 105,000,000 50,500,557 48,260,727 27,782,500 34,201,509 24,797,155 24,240,300 29,079,800 23,554,707 40% 47% 40% 35% 31% 26% 40% 32% 33% 139,318 24,699 15% 164,016 2,027 995,489,252 12,305,182 672,491 8,313 367,417,255 37% 11,960 9,971 14,692 7,279 5,524 7,182 7,490 1,507 681 4,095 777 1,271 1,260 1,934 11% 6% 22% 10% 19% 15% 21% 13,467 10,652 18,787 8,056 6,795 8,442 9,424 2,270 880 2,392 1,751 1,788 3,259 1,632 148,750,082 115,464,736 160,954,238 69,837,837 50,831,069 106,118,317 80,816,211 25,075,874 9,542,540 20,490,673 15,182,138 13,376,597 40,972,323 13,994,149 75,375 58,578 81,482 34,138 24,971 52,656 35,574 12,707 4,841 10,373 7,421 6,571 20,331 6,160 34,836,150 29,546,157 47,538,150 21,221,859 23,497,930 22,334,657 28,332,200 23% 26% 30% 30% 46% 21% 35% 2.8 64,098 11,525 15% 75,623 1,773 732,772,490 17,180,261 362,774 8,505 207,307,103 28% 2.3 203,416 36,224 15% 239,639 1,940 1,728,261,742 13,988,132 ####### 8,379 574,724,358 33% Ongoi ng Contra cts: CHI/98/01 LUN/98/01 LUN/98/02 CHA/98/01 KAT/98/03 KAT/98/02 PET/98/01 PET/99/01 CHI/99/01 Kaw aye Chataya Wheeltrax Contractors Mtondo Construction Libean Contractors Camber Contractors KC/Justed Rapid Construction Rapid Construction Kaw aye Chataya 5.5 4.5 5.5 5.5 4.5 5.5 4.5 100% 100% 100% 100% 100% 100% 100% bridge repair none none none none none none Table A3.5: Summary Fact Sheet of Completed Rehabilitation Contracts as at end of Q1-99 Contract No. Name of Contractor Construction time Extend of Number Road Number Length Km laterite of w idth of Months constructed (km) surf acing culverts (m) culverts per month (10cm) per km Major structures Worker days generated Female Total Male w orker days Worker days % of total Worker days input per km Construction Costs Total ZK Zk per km Total USD Community impact USD per km Total ZK in w ages % of total construction cost Compl eted Contra cts: FTC-1 STC-1/7 STC-2/7 STC-3/7 STC-4/7 STC-5/7 STC-6/7 STC-7/7 KAT/98/01 All contractors Justed Development Mtondo Construction Wheeltrax Contractors Libean Contractors Kaw aye Chataya Camber Contractors Rapid Construction Camber Contractors 20.4 8.0 7.4 6.1 13.0 7.9 6.0 7.5 4.6 4.5 9.0 8.5 6.0 11.0 5.0 4.5 7.5 4.0 4.5 0.9 0.9 1.0 1.2 1.6 1.3 1.0 1.2 80.9 60.0 1.3 10.0 15.7 12.4 10.8 8.5 7.1 10.0 1.3 1.0 8.0 10.0 8.0 7.0 6.0 6.5 8.0 1.0 1.0 1.3 1.6 1.6 1.5 1.4 1.1 1.3 1.3 1.0 Tota l ongoi ng contra cts: 76.7 55.5 Tota l a l l compl eted work s: 157.6 Tota l compl eted contra cts: 5.5 5.5 5.5 5.5 5.5 5.5 4.5 4.5 4.5 100% 93% 100% 100% 52% 100% 100% 96% 100% 53 22 21 8 8 11 8 18 12 2.6 2.8 2.8 1.3 0.6 1.4 1.3 2.4 2.6 161 2.0 41 26 30 28 30 44 30 2 0 4.1 1.7 2.4 2.6 3.6 6.2 3.0 1.5 0.0 1.4 231 1.4 392 1 drif t of 25m none 1 drif t of 30m none 1 drif t of 55m none none none none 42,157 19,983 15,843 11,336 13,818 10,467 9,353 10,129 6,232 10,291 2,042 2,472 121 1,602 267 1,982 3,950 1,972 20% 9% 13% 1% 10% 2% 17% 28% 24% 52,448 22,025 18,315 11,457 15,420 10,734 11,334 14,079 8,204 2,571 2,753 2,483 1,878 1,186 1,359 1,889 1,877 1,783 259,473,040 106,569,737 120,761,756 79,980,730 110,662,077 94,201,798 60,830,961 91,152,249 71,856,904 12,719,267 13,321,217 16,374,475 13,111,595 8,512,467 11,924,278 10,138,494 12,153,633 15,621,066 200,675 65,780 80,170 53,064 69,348 62,995 41,683 60,305 38,471 9,837 8,223 10,871 8,699 6,799 7,974 6,947 7,832 8,363 105,000,000 50,500,557 48,260,727 27,782,500 34,201,509 24,797,155 24,240,300 29,079,800 23,554,707 40% 47% 40% 35% 31% 26% 40% 32% 33% 139,318 24,699 15% 164,016 2,027 995,489,252 12,305,182 672,491 8,313 367,417,255 37% 17,800 16,399 22,686 19,402 11,173 12,894 13,795 2,869 1,775 2,224 887 6,240 2,127 2,292 1,822 3,421 1,092 194 11% 5% 22% 10% 17% 12% 20% 28% 10% 20,024 17,286 28,926 21,529 13,465 14,716 17,216 3,961 1,969 2,002 1,105 2,331 1,993 1,593 2,084 1,722 3,047 1,969 227,901,086 156,563,496 234,382,815 171,755,079 146,894,666 170,047,763 130,838,301 24,556,501 8,953,323 22,790,109 10,004,057 18,886,609 15,903,248 17,383,984 24,086,085 13,083,830 18,889,616 8,953,323 111,058 77,093 114,364 80,289 68,791 81,456 62,313 11,694 4,263 11,106 4,926 9,215 7,434 8,141 11,538 6,231 8,995 4,263 48,147,094 49,437,505 75,121,550 43,187,687 43,763,225 40,343,337 50,654,200 10,878,600 5,426,033 21% 32% 32% 25% 30% 24% 39% 44% 61% 3.0 118,793 20,299 15% 139,092 1,814 1,271,893,030 16,589,188 611,321 7,973 366,959,231 29% 2.5 258,111 44,998 15% 303,108 1,924 2,267,382,282 14,389,683 1,283,812 8,148 734,376,486 32% Ongoi ng Contra cts: CHI/98/01 LUN/98/01 LUN/98/02 CHA/98/01 KAT/98/03 KAT/98/02 PET/98/01 PET/99/01 CHI/99/01 Kaw aye Chataya Wheeltrax Contractors Mtondo Construction Libean Contractors Camber Contractors KC/Justed Rapid Construction Rapid Construction Kaw aye Chataya 5.5 4.5 5.5 5.5 4.5 5.5 4.5 4.5 5.5 100% 100% 100% 100% 100% 100% 100% 100% 100% bridge repair none none none none none none none none 390 Table A3.6: Summary Fact Sheet of Completed Rehabilitation Contracts as at end of Q2-99 Contract No. Name of Contractor Construction time Road Length Km w idth (km) Months constructed (m) per month Extend of Number Number laterite of of surf acing culverts culverts (10cm) per km Major structures Worker days generated Female Total Male w orker days Worker days % of total Construction Costs Worker days input per km Total ZK Zk per km Total USD Community impact USD per km Total ZK in w ages % of total construction cost Compl eted Contra cts: FTC-1 STC-1/7 STC-2/7 STC-3/7 STC-4/7 STC-5/7 STC-6/7 STC-7/7 KAT/98/01 CHI/98/01 ERM/2/2 LUN/98/01 All contractors Justed Development Mtondo Construction Wheeltrax Contractors Libean Contractors Kaw aye Chataya Camber Contractors Rapid Construction Camber Contractors Kaw aye Chataya Libean Contractors Wheeltrax Contractors 20.4 8.0 7.4 6.1 13.2 7.9 6.0 7.7 4.6 10.0 1.8 15.3 4.5 9.0 8.5 6.0 11.0 5.0 4.5 7.5 4.0 8.0 2.5 11.0 4.5 0.9 0.9 1.0 1.2 1.6 1.3 1.0 1.2 1.3 0.7 1.4 108.3 81.5 1.3 2.4 17.7 16.3 16.0 10.1 0.2 9.2 6.8 4.3 8.7 3.0 11.0 10.0 9.0 9.5 1.0 11.0 4.0 3.0 4.0 0.8 1.6 1.6 1.8 1.1 0.2 0.8 1.7 1.4 2.2 Tota l ongoi ng contra cts: 91.8 65.5 Tota l a l l compl eted work s: 200.1 Tota l compl eted contra cts: 5.5 5.5 5.5 5.5 5.5 5.5 4.5 4.5 4.5 5.5 4.5 4.5 100% 93% 100% 100% 52% 100% 100% 96% 100% 100% 100% 100% 53 22 21 8 8 11 8 18 12 41 8 28 2.6 2.8 2.8 1.3 0.6 1.4 1.3 2.4 2.6 4.1 4.5 1.8 238 2.2 26 38 53 56 51 10 33 18 7 19 10.8 2.1 3.2 3.5 5.0 50.0 3.6 2.7 1.6 2.2 1.4 311 1.4 549 1 drif t of 25m none 1 drif t of 30m none 1 drif t of 55m none none none none bridge repair none none 42,157 19,983 15,843 11,336 13,818 10,467 9,353 10,129 6,232 19,036 2,896 16,399 10,291 2,042 2,472 121 1,602 267 1,982 3,950 1,972 2,335 447 887 20% 9% 13% 1% 10% 2% 17% 28% 24% 11% 13% 5% 52,448 22,025 18,315 11,457 15,420 10,734 11,334 14,079 8,204 21,371 3,343 17,286 2,571 2,753 2,483 1,878 1,186 1,359 1,889 1,877 1,783 2,137 1,892 1,134 259,473,040 106,569,737 120,761,756 79,980,730 110,662,077 94,201,798 60,830,961 91,152,249 71,856,904 227,901,086 34,696,714 156,563,496 12,719,267 13,321,217 16,374,475 13,111,595 8,512,467 11,924,278 10,138,494 12,153,633 15,621,066 22,790,109 19,635,945 10,266,459 200,675 65,780 80,170 53,064 69,348 62,995 41,683 60,305 38,471 111,058 16,057 77,093 9,837 8,223 10,871 8,699 6,799 7,974 6,947 7,832 8,363 11,106 9,087 5,055 105,000,000 50,500,557 48,260,727 27,782,500 34,201,509 24,797,155 24,240,300 29,079,800 23,554,707 48,147,094 8,533,664 49,437,505 40% 47% 40% 35% 31% 26% 40% 32% 33% 21% 25% 32% 177,649 28,368 14% 206,016 1,902 1,414,650,548 13,058,473 876,699 8,093 473,535,518 33% 6,177 31,729 28,263 18,452 19,315 1,329 14,164 10,240 6,524 12,700 2,075 7,937 3,043 3,460 2,739 136 3,441 2,818 749 1,069 25% 20% 10% 16% 12% 9% 20% 22% 10% 8% 8,252 39,666 31,306 21,912 22,054 1,465 17,605 13,058 7,273 13,769 3,438 2,235 1,918 1,370 2,179 7,325 1,919 1,932 1,682 1,579 69,074,182 337,510,402 264,204,869 246,722,317 235,232,878 31,998,460 136,232,439 115,147,196 66,147,360 135,780,865 28,780,909 19,018,956 16,189,024 15,420,145 23,244,356 159,992,300 14,848,222 17,033,609 15,294,187 15,567,630 31,938 161,491 124,312 114,706 111,928 13,616 64,882 53,526 29,701 61,495 13,308 9,100 7,617 7,169 11,060 68,080 7,072 7,918 6,867 7,051 22,914,800 108,227,750 68,546,656 74,081,032 65,987,564 4,297,950 51,631,920 41,160,162 20,596,750 5,426,033 33% 32% 26% 30% 28% 13% 38% 36% 31% 4% 3.4 148,893 27,467 16% 176,360 1,922 1,638,050,968 17,849,915 767,595 8,365 462,870,617 28% 2.7 326,542 55,835 15% 382,376 1,911 3,052,701,516 15,255,880 1,644,294 8,217 936,406,135 31% Ongoi ng Contra cts: ERM/2/1 LUN/98/02 CHA/98/01 KAT/98/03 KAT/98/02 KAT/99/01 PET/98/01 PET/99/01 LUN/99/01 CHI/99/01 Mtondo Construction Mtondo Construction Libean Contractors Camber Contractors KC/Justed KC/Justed Rapid Construction Rapid Construction Wheeltrax Contractors Kaw aye Chataya 5.5 5.5 5.5 4.5 5.5 5.5 4.5 4.5 5.5 5.5 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% vented drif t none none none none none none none none none Table A3.7: Summary Fact Sheet of Completed Rehabilitation Contracts as at end of Q3-99 Construction time Contract No. Road Km w idth constructed (m) per month Length (km) Months 20.4 8.0 7.4 6.1 13.2 7.9 6.0 7.7 4.6 10.0 1.8 15.3 3.0 20.1 18.2 17.9 10.1 10.0 4.5 9.0 8.5 6.0 11.0 5.0 4.5 7.5 4.0 8.0 2.5 11.0 5.5 14.0 10.0 12.0 9.5 11.0 4.5 0.9 0.9 1.0 1.2 1.6 1.3 1.0 1.2 1.3 0.7 1.4 0.5 1.4 1.8 1.5 1.1 0.9 187.6 143.5 1.3 5.1 10.2 9.7 16.3 4.0 7.0 6.0 7.0 1.3 1.5 1.6 2.3 Tota l ongoi ng contra cts: 41.4 24.0 Tota l a l l compl eted work s: 228.9 Name of Contractor Extend of Number Number laterite of of surf acing culverts culverts (10cm) per km Major structures Worker days generated Female Total Male w orker days Worker days % of total Worker days input per km Construction Costs Total ZK Zk per km Total USD Community impact USD per km Total ZK in w ages % of total construction cost Compl eted Contra cts: FTC-1 STC-1/7 STC-2/7 STC-3/7 STC-4/7 STC-5/7 STC-6/7 STC-7/7 KAT/98/01 CHI/98/01 ERM/2/2 LUN/98/01 ERM/2/1 LUN/98/02 CHA/98/01 KAT/98/03 KAT/98/02 PET/98/01 All contractors Justed Development Mtondo Construction Wheeltrax Contractors Libean Contractors Kaw aye Chataya Camber Contractors Rapid Construction Camber Contractors Kaw aye Chataya Libean Contractors Wheeltrax Contractors Mtondo Construction Mtondo Construction Libean Contractors Camber Contractors KC/Justed Rapid Construction Tota l compl eted contra cts: 5.5 5.5 5.5 5.5 5.5 5.5 4.5 4.5 4.5 5.5 4.5 4.5 5.5 5.5 5.5 4.5 5.5 4.5 100% 93% 100% 100% 52% 100% 100% 96% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 53 22 21 8 8 11 8 18 12 41 8 28 31 38 73 71 51 33 2.6 2.8 2.8 1.3 0.6 1.4 1.3 2.4 2.6 4.1 4.5 1.8 10.5 1.9 4.0 4.0 5.0 3.3 535 2.9 28 25 25 37 5.5 2.4 2.6 2.3 1.7 115 1.4 650 1 drif t of 25m none 1 drif t of 30m none 1 drif t of 55m none none none none bridge repair none none vented drif t none none none none none 42,157 22,682 16,837 11,336 13,818 10,467 9,353 10,129 6,232 22,632 2,931 16,399 9,191 37,736 35,422 23,234 19,315 14,164 10,291 2,042 2,472 121 1,602 267 1,982 3,950 1,972 2,335 447 887 3,017 9,233 3,694 4,595 2,739 3,441 20% 9% 13% 1% 10% 2% 17% 28% 24% 9% 13% 5% 25% 20% 9% 17% 12% 20% 2,571 2,753 2,483 1,878 1,186 1,359 1,889 1,877 1,783 2,497 1,912 1,134 4,117 2,341 2,149 1,556 2,179 1,761 259,473,040 106,569,737 120,761,756 79,980,729 110,662,077 94,201,798 60,893,961 91,152,249 71,856,904 227,901,086 34,696,714 172,260,550 99,916,982 390,660,498 367,925,935 315,081,093 235,652,878 136,232,439 12,719,267 13,321,217 16,374,475 13,111,595 8,512,467 11,924,278 10,138,494 12,153,633 15,621,066 22,790,109 19,635,945 11,295,774 33,698,813 19,474,601 20,215,711 17,612,135 23,285,858 13,623,244 200,675 65,780 80,170 53,064 69,348 62,995 41,723 60,305 38,471 111,058 16,057 84,568 44,799 183,653 167,599 143,233 112,103 64,882 9,837 8,223 10,871 8,699 6,799 7,974 6,947 7,832 8,363 11,106 9,087 5,545 15,109 9,155 9,209 8,006 11,077 6,488 105,000,000 50,500,557 48,260,727 27,782,500 34,201,509 24,797,155 24,240,300 29,079,800 23,554,707 48,147,094 8,533,664 49,437,505 34,233,500 128,950,050 89,349,487 89,130,507 65,987,564 51,631,920 40% 47% 40% 35% 31% 26% 40% 32% 33% 21% 25% 29% 34% 33% 24% 28% 28% 38% 15% 52,448 24,724 19,309 11,457 15,420 10,734 11,335 14,079 8,204 24,967 3,378 17,286 12,208 46,969 39,116 27,829 22,054 17,605 379,122 324,035 55,087 2,021 2,975,880,423 15,865,693 1,600,484 8,533 932,818,546 31% 8,838 17,391 15,403 24,888 735 4,215 1,843 2,630 8% 20% 11% 10% 9,573 21,606 17,246 27,518 1,877 2,116 1,773 1,686 151,388,845 209,032,395 172,512,682 282,159,934 29,684,087 20,473,300 17,739,093 17,287,093 63,566 92,801 74,963 122,660 12,464 9,089 7,708 7,515 28,567,318 69,657,962 53,077,750 41,440,247 19% 33% 31% 15% 2.8 66,520 9,423 12% 75,943 1,836 815,093,856 19,708,728 353,990 8,559 192,743,277 24% 2.8 390,555 64,510 14% 455,065 1,988 3,790,974,279 16,559,969 1,954,474 8,538 1,125,561,823 30% Ongoi ng Contra cts: KAT/99/01 PET/99/01 LUN/99/01 CHI/99/01 KC/Justed Rapid Construction Wheeltrax Contractors Kaw aye Chataya 5.5 4.5 5.5 5.5 100% 100% 100% 100% none none none none 391 Table A3.8: Summary Fact Sheet of Completed Rehabilitation Contracts as at end of Q4-99 Construction time Contract No. Name of Contractor Length completed (km) Extend of Road laterite Km w idth surf acing Months constructed (m) (10cm) per month Number Number of of culverts culverts completed per km Worker days generated Female Total Major structures Male w orker days Worker days % of total Worker days input per km Construction Costs Total ZK Zk per km Total USD Community impact USD per km Total ZK in w ages % of total construction cost Compl eted Contracts: FTC-1 STC-1/7 STC-2/7 STC-3/7 STC-4/7 STC-5/7 STC-6/7 STC-7/7 KAT/98/01 CHI/98/01 ERM/2/2 LUN/98/01 ERM/2/1 LUN/98/02 CHA/98/01 KAT/98/03 KAT/98/02 PET/98/01 KAT/99/01 PET/99/01 LUN/99/01 CHI/99/01 All contractors Justed Development Mtondo Construction Wheeltrax Contractors Libean Contractors Kaw aye Chataya Camber Contractors Rapid Construction Camber Contractors Kaw aye Chataya Libean Contractors Wheeltrax Contractors Mtondo Construction Mtondo Construction Libean Contractors Camber Contractors KC/Justed Rapid Construction KC/Justed Rapid Construction Wheeltrax Contractors Kaw aye Chataya 20.4 8.0 7.4 6.1 13.2 7.9 6.0 7.7 4.6 10.0 1.8 15.3 3.0 20.1 18.2 17.9 10.1 10.0 10.8 11.3 9.7 16.1 4.5 9.0 8.5 6.0 11.0 5.0 4.5 7.5 4.0 8.0 2.5 11.0 5.5 14.0 10.0 12.0 9.5 11.0 7.0 9.0 9.0 9.0 4.5 0.9 0.9 1.0 1.2 1.6 1.3 1.0 1.2 1.3 0.7 1.4 0.5 1.4 1.8 1.5 1.1 0.9 1.5 1.3 1.1 1.8 235.5 177.5 1.3 1.0 4.0 4.5 5.3 4.4 2.0 2.0 3.0 2.0 3.0 3.0 1.0 0.0 1.3 2.3 1.8 1.5 2.0 Total ongoi ng contracts: 21.2 14.0 Total al l compl eted works: 256.7 Total compl eted contracts: 5.5 5.5 5.5 5.5 5.5 5.5 4.5 4.5 4.5 5.5 4.5 4.5 5.5 5.5 5.5 4.5 5.5 4.5 5.5 4.5 5.5 5.5 100% 93% 100% 100% 52% 100% 100% 96% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 53 22 21 8 8 11 8 18 12 41 8 28 31 38 73 71 51 33 41 27 27 37 2.6 2.8 2.8 1.3 0.6 1.4 1.3 2.4 2.6 4.1 4.5 1.8 10.5 1.9 4.0 4.0 5.0 3.3 3.8 2.4 2.8 2.3 666.619 2.8 2 21 12 9 13 0 2.0 5.3 2.6 1.6 3.0 0.0 1.5 56.28571 1.3 723 1 drif t of 25m none 1 drif t of 30m none 1 drif t of 55m none none none none bridge repair none none vented drif t none none none none none none 2 drif ts 60m none 2 vented drif ts 42,157 22,682 16,837 11,336 13,818 10,467 9,353 10,129 6,232 22,632 2,931 16,399 9,191 37,736 35,422 23,234 19,315 14,164 16,449 19,093 19,140 30,476 10,291 2,042 2,472 121 1,602 267 1,982 3,950 1,972 2,335 447 887 3,017 9,233 3,694 4,595 2,739 3,441 2,007 4,658 2,747 3,750 20% 9% 13% 1% 10% 2% 17% 28% 24% 9% 13% 5% 25% 20% 9% 17% 12% 20% 11% 20% 13% 11% 2,571 2,753 2,483 1,878 1,186 1,359 1,889 1,877 1,783 2,497 1,912 1,134 4,117 2,341 2,149 1,556 2,179 1,761 1,709 2,101 2,256 2,126 259,473,040 106,569,737 120,761,756 79,980,729 110,662,077 94,201,798 60,893,961 91,152,249 71,856,904 227,901,086 34,696,714 172,260,550 99,916,982 390,660,498 367,925,935 315,081,093 235,652,878 136,232,439 265,457,692 239,394,093 257,667,166 339,726,693 12,719,267 13,321,217 16,374,475 13,111,595 8,512,467 11,924,278 10,138,494 12,153,633 15,621,066 22,790,109 19,635,945 11,295,774 33,698,813 19,474,601 20,215,711 17,612,135 23,285,858 13,623,244 24,579,416 21,175,740 26,563,625 21,101,037 200,675 65,780 80,170 53,064 69,348 62,995 41,723 60,305 38,471 111,058 16,057 84,568 44,799 183,653 167,599 143,233 112,103 64,882 111,351 105,579 108,799.1 146,859 9,837 8,223 10,871 8,699 6,799 7,974 6,947 7,832 8,363 11,106 9,087 5,545 15,109 9,155 9,209 8,006 11,077 6,488 10,310 9,339 11,216 9,122 105,000,000 50,500,557 48,260,727 27,782,500 34,201,509 24,797,155 24,240,300 29,079,800 23,554,707 48,147,094 8,533,664 49,437,505 34,233,500 128,950,050 89,349,487 89,130,507 65,987,564 51,631,920 54,326,577 77,652,862 69,061,410 63,629,230 40% 47% 40% 35% 31% 26% 40% 32% 33% 21% 25% 29% 34% 33% 24% 28% 28% 38% 20% 32% 27% 19% 14% 52,448 24,724 19,309 11,457 15,420 10,734 11,335 14,079 8,204 24,967 3,378 17,286 12,208 46,969 39,116 27,829 22,054 17,605 18,456 23,751 21,887 34,226 477,442 409,193 68,249 2,028 4,078,126,068 17,318,934 2,073,071 8,804 1,197,488,625 29% 2,659 8,478 7,539 8,883 9,936 867 287 937 1,510 1,475 1,361 195 10% 10% 17% 14% 12% 18% 2,946 9,415 9,049 10,358 11,297 1,062 2,946 2,354 2,004 1,936 2,593 531 23,510,888 110,887,265 88,870,849 131,539,774 124,520,567 46,296,846 23,510,888 27,721,816 19,677,523 24,591,052 28,583,603 23,148,423 9,675.26 46,252 36,826 54,846 51,372 19,371 9,675 11,563 8,154 10,253 11,792 9,686 8,335,380 27,751,204 31,877,000 29,156,141 36,939,300 4,048,746 35% 25% 36% 22% 30% 9% 2.7 38,362 5,765 13% 44,127 2,079 525,626,189 24,768,198 218,343 10,289 138,107,771 26% 2.8 447,555 74,014 14% 521,569 2,032 4,603,752,257 17,934,792 2,291,414 8,927 1,335,596,396 29% Ongoi ng Contracts: CHA/99/01 CHI/99/02 KAT/99/02 LUN/99/02 PET/99/02 PET/99/03 Wheeltrax Contractors Libean Contractors Rapid Construction Camber Construction Mtondo Construction Kaw aye Chataya 5.5 5.5 5.5 5.5 5.5 5.5 100% 100% 100% 100% 100% 100% 2 x vented drif ts 3 x vented drif ts bridge repair none none 1 x vented drif t 392 Table A3.9: Summary Fact Sheet of Road Rehabilitation Contracts as at end of Fourth Quarter 1999 District Chipata Road name Tamanda Loop FTC-1 Name of Contractor All contractors Total length of Length Number of contracted completed culverts road (km) completed (km) Petauke Total ZK Total USD Community impact USD per km 20.4 20.4 53 1 drift of 25m 52,448 259,473,040 200,675 9,837 Justed Development 8.0 8.0 22 none 24,724 106,569,737 65,780 8,223 Chiparamba road (B) STC-2/7 Mtondo Construction 7.4 7.4 21 1 drift of 30m 19,309 120,761,756 80,170 10,871 11,106 East. Dairy - Madzimoyo CHI/98/01 Kaw aye Chataya 10.0 10.0 41 bridge repair 24,967 227,901,086 111,058 East. Dairy - Madzimoyo CHI/99/01 Kaw aye Chataya 16.1 16.1 37 2 vented drifts 34,226 339,726,693 146,859 9,122 Chizongw e School road ERM/2/2 1.8 1.8 8 none 3,378 34,696,714 16,057 9,087 Lundazi - Mw ase Libean Contractors Total ZK in w ages 50,500,557 1. The completed roads have increased accessibility to the areas of agricultural production . There is evidence of increased 48,147,094 hectarage especially of cotton and 63,629,230 tobacco because buyers are now able to 8,533,664 access the production areas easily. 34,233,500 48,260,727 ERM/2/1 Mtondo Construction 3.0 3.0 31 vented drift 12,208 99,916,982 44,799 15,109 Libean Contractors 19.4 4.0 21 3 x vented drifts 9,415 110,887,265 46,252 11,563 27,751,204 86.0 70.6 180,675 1,299,933,271 711,650 10,079 386,055,976 6.1 6.1 11,457 79,980,729 53,064 8,699 27,782,500 STC-3/7 Wheeltrax Contractors 233.9 8 none Lundazi - Mw ase LUN/98/01 Wheeltrax Contractors 15.3 15.3 28 none 17,286 172,260,550 84,568 5,545 49,437,505 Mw ase - Lundazi LUN/98/02 Mtondo Construction 20.1 20.1 38 none 46,969 390,660,498 183,653 9,155 128,950,050 9.7 27 none 21,887 257,667,166 108,799 11,216 69,061,410 5.3 9 none 10,358 131,539,774 54,846 10,253 29,156,141 107,957 1,032,108,716 484,930 8,589 304,387,606 Lundazi - Mw ase LUN/99/01 Wheeltrax Contractors Mphamba - Chitungulu LUN/99/02 Camber Construction 9.7 Nsadzu - Naviluli - Mlolo STC-4/7 Libean Contractors Nsadzu - Naviluli - Mlolo STC-5/7 Kaw aye Chataya 56.5 109.6 13.2 13.2 8 1 drift of 55m 15,420 110,662,077 69,348 6,799 34,201,509 7.9 7.9 11 none 10,734 94,201,798 62,995 7,974 24,797,155 89,349,487 Vubw i - Zozw e CHA/98/01 Libean Contractors 18.2 18.2 73 none 39,116 367,925,935 167,599 9,209 Chadiza - Tafelansoni CHA/99/01 Wheeltrax Contractors 20.0 1.0 2 2 x vented drifts 2,946 23,510,888 9,675 9,675 8,335,380 59.3 40.3 68,216 596,300,698 309,617 7,675 156,683,531 94.0 T6 - Kalambana STC-6/7 Camber Contractors 6.0 6.0 8 none 11,335 60,893,961 41,723 6,947 T6 - Kalambana KAT/98/01 Camber Contractors 4.6 4.6 12 none 8,204 71,856,904 38,471 8,363 T4 - Chikhombe KAT/98/02 KC/Justed 10.1 10.1 51 none 22,054 235,652,878 112,103 11,077 T6 - Kalambana KAT/98/03 Camber Contractors 17.9 17.9 71 none 27,829 315,081,093 143,233 8,006 T4 - Chikhombe KAT/99/01 KC/Justed 10.8 10.8 41 none 18,456 265,457,692 111,351 10,310 Mbinga - T6 KAT/99/02 Rapid Construction 17.5 4.5 12 bridge repair 66.9 53.9 194.3 9,049 88,870,849 36,826 8,154 96,927 1,037,813,377 483,707 8,970 3. The incomes received by the communities through wages have also contributed to poverty alleviation . Most families w ere able to buy food during the famine months and also bought farming inputs like fertiliser and seed w hich increased their food security. A total of K1.3b has been paid as w ages already. 4. There has been a general improvement in the standard of living of the local communities. More stores are opened along the rehabilitated 89,130,507 sections of the roads and families w ere able to 54,326,577 take children to school. 31,877,000 65,987,564 289,116,655 T4 - Mumbi (Minga) STC-7/7 Rapid Construction 7.7 7.7 18 none 14,079 91,152,249 60,305 7,832 29,079,800 PET/98/01 Rapid Construction 10.0 10.0 33 none 17,605 136,232,439 64,882 6,488 51,631,920 Sichilima - Maw anda PET/99/01 Rapid Construction 11.3 11.3 27 2 drifts 60m 23,751 239,394,093 105,579 9,339 77,652,862 T4 - Chikalawa PET/99/02 Mtondo Construction 20.0 4.4 13 none 11,297 124,520,567 51,372 11,792 36,939,300 T4 - Chataika PET/99/03 Kawaye Chataya 0 1 x vented drift 20.0 2.0 1,062 46,296,846 19,371 9,686 4,048,746 69.0 35.4 91.2 67,794 637,596,194 301,510 8,526 199,352,628 332.3 256.7 722.9 521,569 4,603,752,257 2,291,414 8,927 1,335,596,396 * contracts in italics are on-going 393 2. The labour based technology w hich is used to rehabilitate the roads allow s a higher participation of labour in the w orks. The technology has contributed to the creation of employment . To date, a total of 521,000 w orkerdays have been generated through the Project activities. 24,240,300 23,554,707 T4 - Mumbi (Minga) Totals : Remarks 105,000,000 CHI/99/02 51.1 Katete Construction Costs STC-1/7 GER - Stadium road Chadiza Major structures Total w orkerdays generated (w ds) Chiparamba road (A) Nzamane - Kazimule Lundazi Contract number Table A3.10: Summary Fact Sheet of Road Rehabilitation Contracts as at end of First Quarter 2001 District Chipata Lundazi Chadiza Katete Petauke Mambwe Road name Tamanda Loop Contract number FTC-1 Name of Contractor All contractors Total length of Length Number of contracted completed culverts road (km) completed (km) Major structures Total w orkerdays generated (w ds) Construction Costs Total ZK Total USD Community impact USD per km Total ZK in w ages 20.4 20.4 53 1 drift of 25m 52,448 259,473,040 200,675 9,837 Chiparamba road (A) STC-1/7 Justed Development 8.0 8.0 22 none 24,724 106,569,737 65,780 8,223 50,500,557 Chiparamba road (B) STC-2/7 Mtondo Construction 7.4 7.4 21 1 drift of 30m 19,309 120,761,756 80,170 10,871 48,260,727 105,000,000 East. Dairy - Madzimoyo CHI/98/01 Kaw aye Chataya 10.0 10.0 41 bridge repair 24,967 227,901,086 111,058 11,106 48,147,094 East. Dairy - Madzimoyo CHI/99/01 Kaw aye Chataya 16.1 16.1 37 2 vented drifts 34,226 339,726,693 146,859 9,122 63,629,230 Chizongw e School road ERM/2/2 Libean Contractors 1.8 1.8 8 none 3,378 34,696,714 16,057 9,087 8,533,664 GER - Stadium road ERM/2/1 Mtondo Construction 3.0 3.0 31 vented drift 12,208 99,916,982 44,799 15,109 34,233,500 114,827,159 Nzamane - Kazimule CHI/99/02 Libean Contractors 19.4 19.6 64 3 x vented drifts 30,717 504,109,353 186,998 9,541 Chikando road CHI/00/01 Libean Contractors 5.0 3.0 15 none 1,907 172,256,671 67,202 22,401 6,431,000 91.0 89.2 203,884 1,865,412,030 919,598 10,309 479,562,931 53,064 8,699 27,782,500 291.9 Lundazi - Mw ase STC-3/7 Wheeltrax Contractors 6.1 6.1 8 none 11,457 79,980,729 Lundazi - Mw ase LUN/98/01 Wheeltrax Contractors 15.3 15.3 28 none 17,286 172,260,550 84,568 5,545 49,437,505 Mw ase - Lundazi LUN/98/02 Mtondo Construction 20.1 20.1 38 none 46,969 390,660,498 183,653 9,155 128,950,050 Lundazi - Mw ase LUN/99/01 Wheeltrax Contractors 9.7 9.7 27 none 21,887 257,667,166 108,799 11,216 69,061,410 Mphamba - Chitungulu LUN/99/02 Camber Construction 20 18.7 43 1 x vented drift 41,816 555,531,683 204,921 10,959 129,946,482 71.1 69.8 144.1 139,415 1,456,100,626 635,005 9,096 405,177,947 Nsadzu - Naviluli - Mlolo STC-4/7 Libean Contractors Nsadzu - Naviluli - Mlolo STC-5/7 Kaw aye Chataya 13.2 13.2 8 1 drift of 55m 15,420 110,662,077 69,348 6,799 34,201,509 7.9 7.9 11 none 10,734 94,201,798 62,995 7,974 24,797,155 Vubw i - Zozw e CHA/98/01 Libean Contractors 18.2 18.2 73 none 39,116 367,925,935 167,599 9,209 89,349,487 Chadiza - Tafelansoni CHA/99/01 Wheeltrax Contractors 20.0 19.6 37 2 x vented drifts 44,558 549,519,704 184,040 9,372 125,514,570 59.3 59.0 129.0 109,828 1,122,309,513 483,982 8,206 273,862,721 T6 - Kalambana STC-6/7 Camber Contractors 6.0 6.0 8 none 11,335 60,893,961 41,723 6,947 24,240,300 T6 - Kalambana KAT/98/01 Camber Contractors 4.6 4.6 12 none 8,204 71,856,904 38,471 8,363 23,554,707 T4 - Chikhombe KAT/98/02 KC/Justed 10.1 10.1 51 none 22,054 235,652,878 112,103 11,077 65,987,564 T6 - Kalambana KAT/98/03 Camber Contractors 17.9 17.9 71 none 27,829 315,081,093 143,233 8,006 89,130,507 T4 - Chikhombe KAT/99/01 KC/Justed 10.8 10.8 41 none 18,456 265,457,692 111,351 10,310 54,326,577 Mbinga - T6 KAT/99/02 Rapid Construction 17.5 17.5 51 bridge repair 34,794 436,951,831 160,327 9,162 156,839,309 Kavulamungu - T6 KAT/00/01 Rapid Construction 6.1 3.9 15 73.1 70.8 248.7 7,170 146,156,294 43,739 11,278 31,436,400 129,842 1,532,050,653 650,948 9,196 445,515,364 T4 - Mumbi (Minga) STC-7/7 Rapid Construction 7.7 7.7 18 none 14,079 91,152,249 60,305 7,832 29,079,800 T4 - Mumbi (Minga) PET/98/01 Rapid Construction 10.0 10.0 33 none 17,605 136,232,439 64,882 6,488 51,631,920 Sichilima - Maw anda PET/99/01 Rapid Construction 11.3 11.3 27 2 drifts 60m 23,751 239,394,093 105,579 9,339 77,652,862 T4 - Chikalaw a, section 1 PET/99/02 Mtondo Construction 20.0 20.0 57 none 49,222 502,740,802 188,686 9,434 157,445,254 T4 - Chataika PET/99/03 Kaw aye Chataya 21.7 21.7 70 1 x vented drift 36,195 614,928,046 225,567 10,395 129,248,843 T4 - Chikalawa, section 2 PET/00/01 Mtondo Construction 5.0 4.8 12 none 12,587 146,156,294 43,739 9,185 40,000,000 75.7 75.5 217 153,439 1,730,603,923 688,759 9,127 485,058,679 20.0 18.1 76 32,906 647,779,451 209,235 11,531 97,741,593 20.0 18.1 76 32,906.0 647,779,451.4 209,235.1 11,531.0 97,741,593.0 390.2 382.4 1,107 769,314 8,354,256,196 3,587,527 9,382 2,186,919,235 Mambwe - Msoro MAM/99/01 Kalumbu Contractors 1 x Vented drift * contracts in italics are on-going Totals : 394 Remarks 1. The completed roads have increased accessibility to the areas of agricultural production . 2. The labour based technology w hich is used to rehabilitate the roads allow s a higher participation of labour in the w orks. The technology has contributed to the creation of employment . To date, a total of 769,314 w orkerdays have been generated through the Project activities. 3. The incomes received by the communities through wages have also contributed directly or indirectly to improved living standards . Most families w ere able to buy farming inputs like fertiliser and seed w hich is expected to increase their food security this year. A total of K2.187 billion has been paid as w ages already. 15% has Table A3.11: Summary Fact Sheet of Road Rehabilitation Contracts as at end of Second Quarter 2001 District Chipata Road name Tamanda Loop Contract number FTC-1 Name of Contractor All contractors Total Length Number of length of culverts contracted completed (km) completed road (km) 20.4 Katete Petauke Mambw e 1 drift of 25m 52,448 Total ZK 259,473,040 Total USD 200,675 Community impact USD per km 9,837 Total ZK in w ages STC-1/7 Justed Development 8.0 8.0 22 none 24,724 106,569,737 65,780 8,223 STC-2/7 Mtondo Construction 7.4 7.4 21 1 drift of 30m 19,309 120,761,756 80,170 10,871 48,260,727 50,500,557 East. Dairy - Madzimoyo CHI/98/01 Kaw aye Chataya 10.0 10.0 41 bridge repair 24,967 227,901,086 111,058 11,106 48,147,094 East. Dairy - Madzimoyo CHI/99/01 Kaw aye Chataya 16.1 16.1 37 2 vented drifts 34,226 339,726,693 146,859 9,122 63,629,230 ERM/2/2 Libean Contractors 1.8 GER - Stadium road ERM/2/1 Mtondo Construction 3.0 Nzamane - Kazimule CHI/99/02 Libean Contractors 19.4 Chikando road CHI/00/01 Libean Contractors 5.0 1.8 16,057 9,087 3.0 31 vented drift 12,208 99,916,982 44,799 15,109 34,233,500 19.6 64 3 x vented drifts 30,717 504,109,353 186,998 9,541 114,827,159 5.0 21 none 91.2 8 none 3,378 298.1 34,696,714 8,533,664 6,593 225,227,260 67,202 13,464 23,332,624 208,570 1,918,382,619 919,598 10,084 496,464,555 27,782,500 Lundazi - Mw ase STC-3/7 Wheeltrax Contractors 6.1 6.1 8 none 11,457 79,980,729 53,064 8,699 Lundazi - Mw ase LUN/98/01 Wheeltrax Contractors 15.3 15.3 28 none 17,286 172,260,550 84,568 5,545 49,437,505 Mw ase - Lundazi LUN/98/02 Mtondo Construction 20.1 20.1 38 none 46,969 390,660,498 183,653 9,155 128,950,050 Lundazi - Mw ase LUN/99/01 Wheeltrax Contractors 9.7 9.7 27 none 21,887 257,667,166 108,799 11,216 69,061,410 Mphamba - Chitungulu LUN/99/02 Camber Construction 20 19.9 43 3 x vented drift 43,563 555,531,683 204,921 10,296 134,983,330 Chiginya - Phikamalaza LUN/01/01 Camber Construction 0 1xvented drift Nsadzu - Naviluli - Mlolo STC-4/7 Libean Contractors Nsadzu - Naviluli - Mlolo STC-5/7 Kaw aye Chataya Vubw i - Zozw e CHA/98/01 Libean Contractors Chadiza - Tafelansoni CHA/99/01 Wheeltrax Contractors T6 - Kalambana STC-6/7 Camber Contractors 8 2.0 79.1 73.0 13.2 13.2 7.9 7.9 18.2 144.1 10,090.79 5,045 4,746,000 1,489,803,876 645,096 8,835 414,960,795 1 drift of 55m 15,420 110,662,077 69,348 6,799 34,201,509 none 10,734 94,201,798 62,995 7,974 24,797,155 18.2 73 none 19.6 52 2 x vented drifts 59.0 6.0 33,703,250.00 8 20.0 6.0 1,378 142,540 11 59.3 39,116 144.4 8 none 367,925,935 167,599 9,209 89,349,487 48,680 549,519,704 184,040 9,372 142,581,570 113,950 1,122,309,513 483,982 8,206 290,929,721 11,335 6,947 24,240,300 T6 - Kalambana KAT/98/01 Camber Contractors 4.6 4.6 12 none 8,204 71,856,904 38,471 8,363 23,554,707 T4 - Chikhombe KAT/98/02 KC/Justed 10.1 10.1 51 none 22,054 235,652,878 112,103 11,077 65,987,564 T6 - Kalambana KAT/98/03 Camber Contractors 17.9 17.9 71 none 27,829 315,081,093 143,233 8,006 89,130,507 T4 - Chikhombe KAT/99/01 KC/Justed 10.8 10.8 41 none 18,456 265,457,692 111,351 10,310 54,326,577 Mbinga - T6 KAT/99/02 Rapid Construction 17.5 17.5 51 bridge repair 34,794 436,951,831 160,327 9,162 156,839,309 Kavulamungu - T6 KAT/00/01 Rapid Construction 6.1 6.0 21 11,370 172,256,671 53,488 8,858 49,682,500 73.1 72.9 134,042 1,558,151,030 660,697 9,057 463,761,464 254.3 60,893,961 41,723 T4 - Mumbi (Minga) STC-7/7 Rapid Construction 7.7 7.7 18 none 14,079 91,152,249 60,305 7,832 29,079,800 T4 - Mumbi (Minga) PET/98/01 Rapid Construction 10.0 10.0 33 none 17,605 136,232,439 64,882 6,488 51,631,920 Sichilima - Maw anda PET/99/01 Rapid Construction 11.3 11.3 27 2 drifts 60m 23,751 239,394,093 105,579 9,339 77,652,862 T4 - Chikalaw a, section 1 PET/99/02 Mtondo Construction 20.0 20.0 57 none 49,222 502,740,802 188,686 9,434 157,445,254 T4 - Chataika PET/99/03 Kaw aye Chataya 21.7 21.7 70 1 x vented drift 614,928,046 129,248,843 T4 - Chikalaw a, section 2 PET/00/01 Mtondo Construction 12 none Mambw e - Msoro MAM/99/01 Kalumbu Contractors 5.0 5.5 75.7 76.2 217 20.0 18.1 76 20.0 18.1 76 398.2 390.5 36,195 1 x Vented drift 225,567 10,395 12,587 146,156,294 43,739 7,979 40,000,000 153,439 1,730,603,923 688,759 9,040 485,058,679 34,329 647,779,451 209,235 11,531 101,771,705 34,329.0 647,779,451.4 209,235.1 11,531.0 101,771,705.0 786,870 8,467,030,412 3,607,367 9,239 2,252,946,919 * contracts in italics are on-going Totals : 1,135 395 Remarks 105,000,000 Chiparamba road (A) 91.0 Chadiza 53 Construction Costs Chiparamba road (B) Chizongw e School road Lundazi 20.4 Major structures Total w orkerdays generated (w ds) 1. The completed roads have increased accessibility to the areas of agricultural production . 2. The labour based technology w hich is used to rehabilitate the roads allow s a higher participation of labour in the w orks. The technology has contributed to the creation of employment . 3. The incomes received by the communities through wages have also contributed directly or indirectly to improved living standards Table A3.12: Summary Fact Sheet of Road Rehabilitation Contracts as at 6th December 2001 District Chipata Road name Tamanda Loop Chiparamba road (A) Chiparamba road (B) East. Dairy - Madzimoyo East. Dairy - Madzimoyo Road Number RD 118 RD 121 U 33 Contract number FTC-1 Name of Contractor All contractors Total length of Length Number of contracted completed culverts road (km) completed (km) Total ZK in w ages 20.4 20.4 53 1 drift of 25m 52,448 2,571 259,473,040 200,675 9,837 105,000,000 Justed Development 8.0 8.0 22 none 24,724 3,091 106,569,737 65,780 8,223 50,500,557 STC-2/7 Mtondo Construction 7.4 7.4 21 1 drift of 30m 19,309 2,618 120,761,756 80,170 10,871 48,260,727 CHI/98/01 Kaw aye Chataya 10.0 10.0 41 bridge repair 24,967 2,497 227,901,086 111,058 11,106 48,147,094 CHI/99/01 Kaw aye Chataya 16.1 16.1 37 2 vented drifts 34,226 2,126 339,726,693 146,859 9,122 63,629,230 3,378 1,912 34,696,714 16,057 9,087 8,533,664 1.8 1.8 8 none GER - Stadium road ERM/2/1 Mtondo Construction 3.0 3.0 31 vented drift 12,208 4,117 99,916,982 44,799 15,109 34,233,500 119,723,659 RD 595 CHI/99/02 Libean Contractors 19.6 19.4 64 3 x vented drifts 31,926 1,646 566,939,753 204,095 10,520 U01 CHI/01/01 Wheeltrax Contractors 5.0 0.5 31 none 2,571 5,142 23,298,930 6,289 12,577 23,332,624 RD 596 CHI/00/01 Libean Contractors 5.0 5.0 21 none 6,593 1,321 225,227,260 67,202 13,464 23,332,624 212,350 2,321 2,004,511,949 942,984 10,306 524,693,679 27,782,500 Lundazi - Mw ase Lundazi - Mw ase Mw ase - Lundazi RD 110 Lundazi - Mw ase Wheeltrax Contractors 6.1 6.1 8 none 11,457 1,878 79,980,729 53,064 8,699 Wheeltrax Contractors 15.3 15.3 28 none 17,286 1,134 172,260,550 84,568 5,545 49,437,505 LUN/98/02 Mtondo Construction 20.1 20.1 38 none 46,969 2,341 390,660,498 183,653 9,155 128,950,050 9.7 27 none 21,887 2,256 257,667,166 108,799 11,216 69,061,410 20.0 47 3 x vented drift 43,563 2,178 634,809,351 226,287 11,314 134,983,330 9 1xvented drift LUN/99/01 Wheeltrax Contractors LUN/99/02 Camber Construction Chiginya - Phikamalaza R 245 LUN/01/01 Camber Construction U3 329.1 STC-3/7 D 104 Nsadzu - Naviluli - Mlolo 91.5 LUN/98/01 Mphamba - Chitungulu Nsadzu - Naviluli - Mlolo STC-4/7 Libean Contractors STC-5/7 Kaw aye Chataya 9.7 20 8 6.0 79.1 77.1 13.2 13.2 8 7.9 7.9 11 6,697 1,116 88,006,817 24,968.48 4,161 23,950,700 147,859 1,918 1,623,385,110 681,339 8,836 434,165,495 1 drift of 55m 15,420 1,168 110,662,077 69,348 6,799 34,201,509 none 10,734 1,352 94,201,798 62,995 7,974 24,797,155 157.1 Vubw i - Zozw e RD 406 CHA/98/01 Libean Contractors 18.2 18.2 73 none 39,116 2,149 367,925,935 167,599 9,209 89,349,487 Chadiza - Tafelansoni RD 405 CHA/99/01 Wheeltrax Contractors 20.0 20.0 55 2 x vented drifts 53,735 2,687 549,519,704 184,040 9,202 156,431,770 146.7 T6 - Kalambana T6 - Kalambana RD 585 59.3 59.3 119,005 2,005 1,122,309,513 483,982 8,156 304,779,921 STC-6/7 Camber Contractors 6.0 6.0 8 none 11,335 1,889 60,893,961 41,723 6,947 24,240,300 KAT/98/01 Camber Contractors 4.6 4.6 12 none 8,204 1,783 71,856,904 38,471 8,363 23,554,707 T4 - Chikhombe U 23 KAT/98/02 KC/Justed 10.1 10.1 51 none 22,054 2,179 235,652,878 112,103 11,077 65,987,564 T6 - Kalambana RD 585 KAT/98/03 Camber Contractors 17.9 17.9 71 none 27,829 1,556 315,081,093 143,233 8,006 89,130,507 T4 - Chikhombe U 23 KAT/99/01 KC/Justed 10.8 10.8 41 none 18,456 1,709 265,457,692 111,351 10,310 54,326,577 Mbinga - T6 U 29 KAT/99/02 Rapid Construction 17.5 17.5 51 bridge repair 34,915 1,995 436,951,831 160,327 9,162 157,182,809 Unclassified KAT/00/01 Rapid Construction 6.1 6.0 21 73.1 72.9 Kavulamungu - T6 T4 - Mumbi (Minga) RD 415 Sichilima - Maw anda T4 - Chikalaw a, section 1 T4 - Chataika T4 - Chikalaw a, section 2 254.3 11,400 1,888 172,256,671 53,488 8,858 49,787,500 134,193 1,840 1,558,151,030 660,697 9,057 464,209,964 29,079,800 STC-7/7 Rapid Construction 7.7 7.7 18 none 14,079 1,828 91,152,249 60,305 7,832 PET/98/01 Rapid Construction 10.0 10.0 33 none 17,605 1,761 136,232,439 64,882 6,488 51,631,920 RD 135 PET/99/01 Rapid Construction 11.3 11.3 27 2 drifts 60m 23,751 2,101 239,394,093 105,579 9,339 77,652,862 T4 - Mumbi (Minga) Mambw e USD per km STC-1/7 96.2 Petauke Total USD Libean Contractors Chikando road Katete Total ZK Community impact ERM/2/2 Kapata - Chizongwe road Chadiza Workerdays per km (WDS/km) Chizongw e School road Nzamane - Kazimule Lundazi Major structures Construction Costs Total w orkerdays generated (w ds) R 13 PET/99/02 Mtondo Construction 20.0 20.0 57 none 49,222 2,461 502,740,802 188,686 9,434 157,445,254 RD 413 PET/99/03 Kaw aye Chataya 21.7 21.7 70 1 x vented drift 36,195 1,668 614,928,046 225,567 10,395 129,248,843 R 13 PET/00/01 Mtondo Construction 5.0 5.5 12 none 12,587 2,296 146,156,294 43,739 7,979 40,000,000 75.7 76.2 153,439 2,014 1,730,603,923 688,759 9,040 485,058,679 Mambw e - Msoro MAM/99/01 Kalumbu Contractors 217 20.0 19.9 76 20.0 19.9 76 403.4 397.0 1 x Vented drift 34,329 1,725 647,779,451 209,235 10,514 101,771,705 34,329.0 1,725.1 647,779,451.4 209,235.1 10,514.3 101,771,705.0 801,175 2,018 8,686,740,977 3,666,996 9,237 2,314,679,443 * contracts in italics are on-going Totals : 1,181 Source: EPFRP Management Unit. 396 Box A1. Steps in the evaluation of a rural roads project Step 1: Deciding whether to implement an impact evaluation Is there sufficient support and cooperation: • From the government? • From the bank project team and bank management? • From funding sources? Is a credible evaluation feasible: • Is there in-country capacity (data collection, supervision)? • Are there existing or planned surveys that can be used or questionnaires that can be adapted? • Is there a potential sampling frame in the prospective zone of influence? • Is there time to prepare and field a baseline before the project begins? • ωan a counterfactual be identified under seemingly plausible assumptions? Step 2: Learn from the ex-ante evaluation Understanding programme placement: to understand biases in the ex-post evaluation and define an appropriate counterfactual Step 3: Set up the evaluation team • Finding a stable in-country home for the evaluation. • ωhoosing an evaluation team that is reasonably independent of executing agency yet can work with that agency as need be: local counterpart, interviewers, data processors. Step 4: The evaluation design: Deciding what data are needed Outcome variables (distributional impacts, traffic counts, time use, travel diaries) Control variables Project data Choice and definition of: • Zone of influence • ψeneficiariesμ communities, households, firms, individuals? • ωomparison areas Step 5: The evaluation design: Collecting the data: Identify data sources and data collection methods Sampling and sample size Designing survey instruments Deciding on timing of baseline and follow-up rounds Step 6: Analysis and writing up Plan adequate time for data processing: entry, cleaning, lessons for follow-up; analysis of baseline Plan for follow-up survey(s). Source: Van de Walle, 2009:33. 397 Annex: Chapter 4 Key Informants interviewed in Lusaka: Mr. Silwimba, Land Use Planning and Mapping Unit, Ministry of Agriculture, 30 August 2004. Mr. Nicholas Mwale, Statistician in the Database and Early Warning Unit, Ministry of Agriculture, 30 August 2004. Mr. Gibson Kapili, Contracts & Financial Manager, Smallholder Enterprise and Marketing Programme, 30 Aug 2004. Dr. Klaus Kroppelmann, Monitoring and Evaluation Expert at the Agricultural Consultative Forum, Sept. 2004. Mr. Masileso Sooka, Agriculture and Environment Department of the Central Statistical Office, 10 Sept. 2004. [Responsible for the PHS data collection and methodology] Mr. Ollings Chihana, Agriculture Statistical Division of the CSO, 10 Sept. 2004. Dr. Jones Govereh, Research Fellow, Food Security Research Project, 10 Sep. 2004. Mr. Ballard Zulu, Research Specialist, Food Security Research Project, 10 Sep. 2004. 398 Map A1: Zambia Eastern Province‟s District Roadmap Sourceμ Authors‘ based on DIVA-GIS 5.2. Table A1: Road and Population Density in the Districts of Eastern Province Density Population (people/km²) Districts Districts Province Capital Area (km²) Central Copperbelt Eastern Luapula Lusaka Northern North-Western Kabwe Ndola Chipata Mansa Lusaka Kasama Solwezi 94395 31328 69106 50567 21898 147826 125827 1,012,257 1,581,221 1,306,173 775,353 1,391,329 1,258,696 583,35 10.7 50.5 18.9 15.3 63.5 8.5 4.6 6 10 8 7 4 7 12 Southern Western Zambia Livingstone Mongu Lusaka 85283 126386 752616 1,212,124 765,088 9,885,591 14.2 6.1 13.1 11 7 72 Area Density Total Length Road Density (km) Population Road length/ Road length / (people/km²) of Feeder Shape (2000) 100 km2 1000 population Area (GIS) (2000) Roads (*) Chadiza 2410,18 83981 34,8 314,4 13,04 3,74 2170,16 Chama 16209,64 7489 0,5 700,1 4,32 93,48 15304,33 Chipata 8166,26 367539 45,0 402,35 4,93 1,09 4726,95 Katete 3223,02 18925 5,9 506,8 15,72 26,78 3187,19 Lundazi 10694,58 236833 22,1 727,6 6,80 3,07 11537,15 Mambwe 6124,69 70425 11,5 402,35 6,57 5,71 5112,32 Nyimba 12476,23 47376 3,8 404,05 3,24 8,53 6287,88 Petauke Eastern Area (km²) 9801,40 69106,00 Sourceμ Author‘s calculations. 399 235879 1068447 24,1 15,5 404,05 3861,7 4,12 5,59 1,71 3,61 9198,50 57524,48 Table A2: Cotton Production and Yield Trends Source: Zambia Food Security Research Project, 2000:3. Table A3: Total Observations in Eastern Province Zambia, 1996/1997 – 2001/2002 District 1996/1997 1997/1998 1998/1999 1999/2000 2000/2001 2001/2002 Chadiza (301) 96 88 89 100 88 100 Chipata (303) 303 295 304 338 307 330 Katete (304) 198 198 199 220 184 212 Lundazi (305) 224 225 229 260 233 261 Petauke (308) 267 262 271 320 262 305 Total Catchment Districts 1088 1068 1092 1238 1074 1208 Chama (302) 37 36 76 80 70 77 Mambwe (306) 52 55 34 59 51 59 Nyimba (307) 48 37 53 60 54 59 Total Control Districts 137 128 163 199 175 195 Total 1225 1196 1255 1437 1249 1403 Source: Authors‘ calculations based on the Post Harvest Surveys 1997-2002. Table A4: Total Average Land Area in Eastern Province Zambia, 1996/1997 – 2001/2002 District Chadiza (301) Chipata (303) Katete (304) Lundazi (305) Petauke (308) Total Catchment Districts Chama (302) Mambwe (306) Nyimba (307) Total Control Districts Total 1996/1997 2,121 2,266 1,827 1,946 1,937 1,995 0,765 1,318 1,535 1,624 1,939 1997/1998 2,094 1,841 1,789 1,844 2,010 1,870 0,734 1,375 1,691 1,599 1,829 1998/1999 1,840 1,899 1,627 1,939 2,043 1,862 1,048 1,702 1,286 1,514 1,807 Source: Authors‘ calculations based on the θost Harvest Surveys 1997-2002. 400 1999/2000 1,676 2,344 2,325 2,217 1,840 2,153 1,082 1,674 1,832 1,476 2,040 2000/2001 1,727 1,777 1,727 1,563 1,927 1,765 0,833 1,546 1,907 1,383 1,732 2001/2002 1,791 1,901 1,864 1,656 2,122 1,903 0,909 1,710 2,013 1,483 1,832 Table A5: Percentage of Households that Grow Maize Conditional on Growing Cotton in Eastern Province, 1997 - 2002 District Chadiza (301) Chipata (303) Katete (304) Lundazi (305) Petauke (308) Total Catchment Districts Chama (302) Mambwe (306) Nyimba (307) Total Control Districts Total Eastern Province 1996/97 100,00% 99,67% 99,49% 99,55% 99,25% 99,54% 100,00% 100,00% 100,00% 100,00% 99,59% 1997/98 100,00% 97,29% 98,48% 97,33% 98,47% 98,03% 97,22% 94,55% 100,00% 96,88% 97,91% 1998/99 100,00% 96,71% 98,49% 98,69% 98,15% 98,08% 97,37% 97,06% 100,00% 98,16% 98,09% 1999/2000 2000/2001 99,00% 97,73% 99,11% 99,67% 100,00% 99,46% 95,00% 97,01% 98,75% 99,62% 98,30% 98,88% 98,75% 100,00% 98,31% 100,00% 98,33% 100,00% 98,49% 100,00% 98,33% 99,04% 2001/2002 97,00% 98,79% 99,06% 97,70% 99,34% 98,59% 100,00% 100,00% 100,00% 100,00% 98,79% Source: Authors‘ calculations based on the θost Harvest Surveys 1997-2002. Table A6: Rainfall between 1994/95 and 2004/05 (in millimeter) Year 1994/1995 1995/1996 1996/1997 1997/1998 1998/1999 1999/2000 2000/2001 2001/2002 2002/2003 2003/2004 2004/2005 Long-term Mean Eastern 528,55 805,71 813,72 719,02 759,38 678,55 914,78 700,40 835,42 781,11 788,61 756,84 Chadiza (301) (i) 610,83 868,50 945,33 708,50 901,42 584,83 1165,17 771,92 871,33 915,92 1007,92 850,15 Chama (302) (iii) 494,08 688,58 708,00 760,00 574,00 695,25 562,58 564,67 685,50 746,00 750,17 657,17 Chipata (303) (i) 610,83 868,50 945,33 708,50 901,42 584,83 1165,17 771,92 871,33 915,92 1007,92 850,15 Katete (304) 214,17 936,83 826,00 874,75 791,08 711,83 1163,50 846,17 877,47 878,01 895,40 819,56 Lundazi (305) 693,58 721,67 609,08 696,58 543,42 563,75 761,58 607,67 815,42 658,37 681,36 668,41 Mambwe (306) (iii) 494,08 688,58 708,00 760,00 574,00 695,25 562,58 564,67 685,50 746,00 750,17 657,17 Nyimba (307) (ii) 555,42 836,50 884,00 621,92 894,83 796,33 968,83 738,08 938,42 694,33 608,00 776,06 Petauke (308) (ii) 555,42 836,50 884,00 621,92 894,83 796,33 968,83 738,08 938,42 694,33 608,00 776,06 Long-term Mean 756,84 756,84 756,84 756,84 756,84 756,84 756,84 756,84 756,84 756,84 756,84 Notes: We assume the following coverage for the five weather stations: Chipata covers Chipata and Chadiza districts; Lundazi covers Lundazi district; Petauke covers Petauke and Katete districts; Msekere covers Katete district; and Mfuwe covers Chama and Mambwe districts. Source: Author based on Zambia Meteorological Service data. Table A7: Implementation of the EPFRP, 1996/1997 to 2001/2002 Agricultural Season Control Districts Catchment Districts District Codes 1996/1997 302, 306, 307 Yes 301, 303, 304, 305, 308 No 1997/1998 1998/1999 Yes Yes No Yes 1999/2000 2000/2001 2001/2002 Yes Yes Yes Yes Yes Yes Source: Authors. Figure A1: Index of offering prices on the international raw cotton market, Index 1990 = 100 Cotton Outlook 'A Index' 120,0 100,0 80,0 60,0 'A Index' 40,0 20,0 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 0,0 Notes: It is an average of the cheapest five quotations from a selection (at present numbering nineteen) of the principal upland cottons traded internationally. Sourceμ Authors‘ calculation based upon Cotton Outlook‘s publishing of CIF prices. 401 Table A8: EPFRP treated roads in Catchment Districts Road No. Rd Length (km) Road Name RD405 D128 - Zingalume - Mwangala RD406 DR405 - D130 U3 T6 - Naviruli Total EPFRP Primary Road Length EPFRP Share of Primary of Feeder Roads in Chadiza EPFRP Share of Total Length of Feeder Roads in Chadiza 44 70 7,6 121,6 63,3% 38,7% Category District Chadiza P Chadiza P Chadiza P Chadiza Out of 192 km Out of 314 km P RD118 M12-Tamanda Mission 6,7 P Chipata RD121 D104 - Chipalamba - D104 14,7 P Chipata RD400 D124 - Chiguya 12,2 P Chipata RD401 T4 - Madzimawe - D124 15,6 P Chipata RD595 T4 - Nzamane - Kazimuli 19,3 P Chipata RD596 RD595 - Sayiri - D128 25,5 P Chipata U33 Link RD 402 - Madzimoyo 8,3 P Chipata Total Primary Road Length 102,3 P Chipata EPFRP Share of Primary Length of Feeder Roads in Chipata 19,3% 12,7% 14,9 18,8 10 33,2 17,2 69,2 14,9 10 94,1 30,1% 18,6% 25,5 27,6 23,9 35,4 10 8,2 17 17,2 9,5 25,1 10,9 154,7 27,9 27,7 210,3 31,1% 28,9% 21 21,2 19 25,1 86,3 16,5% 10,7% 534,1 27,1% 16,9% EPFRP Share of Total Length of Feeder Roads in Chipata RD409 Chikonza - Walilanji - T6 RD585 T6 - Kalambana School R292 Walilanji - Mbabala - T6 U23 Katete (T4) - Kazungulile - D598 U29 T6 - Mbinga - Katete Total Primary Road Length Total Secondary Road Length Total Tertiary Road Length Total EPFRP Road Length EPFRP Share of Primary Length of Feeder Roads in Katete EPFRP Share of Total Length of Feeder Roads in Katete RD107 D103 - Emusa - Chasefu - Chama Boundary RD110 Lundazi (M12) - Mwase RD110N D109 - Kapachila - Mwase (RD110) R243 Mphamba (D104) - Nyalubanga (D103) R246 Phikhamalaza (R245) - R248 R250 RD110 N - Kanyunya School R251 Mwase (RD 110 ) - Pono R254 RD110 - Gwaba - Kamtande R255 Mwase (RD110) - R254 U16 Gwaba (R254 ) - TBZ - Lumezi (M12) U18 Kapachila - RD110 Total Primary Road Length Total Secondary Road Length Total Tertiary Road Length Total Feeder Road Length EPFRP Share of Primary Length of Feeder Roads in Lundazi EPFRP Share of Total Length of Feeder Roads in Lundazi RD135 D139 - Sasare RD413 R12 Chataika - T4 RD415 Minga (T4) - Nyalukomba (D414): 'D' State Road R13 T4 - Chikalawa School (R12) Total Primary Road Length EPFRP Share of Sub-Total Length of Feeder Roads in Petauke EPFRP Share of Total Length of Feeder Roads in Petauke Total Primary Road Length in Eastern Province EPFRP Share of Primary Length of Feeder Roads in Eastern EPFRP Share of Total Length of Feeder Roads in Eastern Out of 529,4 km Out of 804,7 km Katete Katete T Katete P Katete P Katete P Katete S Katete T Katete Total Katete Out of 229 km Out of 506 km P Lundazi P Lundazi P Lundazi P Lundazi T Lundazi T Lundazi S Lundazi P Lundazi T Lundazi P Lundazi S Lundazi P Lundazi S Lundazi T Lundazi Total Lundazi Out of 497,9 km Out of 727,6 km P Petauke P Petauke P Petauke P Petauke P Petauke Out of 524 km Out of 808 km P Eastern Out of 1972 km Out of 3162 km S P Sourceμ Authors‘ calculations based upon EPFRP documents and Ministry of Local Government & Housing, 1998: Feeder Roads Support Programme. Volume 2: District Feeder Road Lists. Notes: P = Primary Feeder Roads; S = Secondary Feeder Roads; and T = Tertiary Feeder Roads. 402 Figure A2a: Trends in Maize Productivity EPFRP Catchment versus Control Districts 8,500 Log yield of maize 8,000 7,500 7,000 Total Catchment Districts Total Control Districts 6,500 6,000 Sourceμ Authors‘ estimations based on the Post Harvest Survey. Figure A2b: Trends in Cotton Productivity EPFRP Catchment versus Control Districts 7,5000 7,4000 Log yield of cotton 7,3000 7,2000 7,1000 7,0000 Total Catchment Districts 6,9000 Total Control Districts 6,8000 6,7000 6,6000 6,5000 1996/97 1997/98 1998/99 1999/2000 2000/2001 2001/2002 Sourceμ Authors‘ estimations based on the Post Harvest Survey. Table A9a: Descriptive Statistics of Dependent Variable: Log Cotton Productivity Treatment Treatment Control Treatment Control Control Period 1996/97-2001/02 1996/97-2001/02 1998/99-2001/02 1998/99-2001/02 1996/97-1997/98 Variable logyield logyield logyield logyield logyield Obs 1238 925 1238 196 729 Mean 6,650 6,726 6,650 6,803 6,705 Std.Dev 0,995 0,993 0,995 0,845 1,029 Min 2,286 0,811 2,286 2,773 0,811 Max 10,697 10,217 10,697 8,490 10,217 Table A9b: Descriptive Statistics of Dependent Variable: Cotton Productivity (kgs) Treatment Treatment Control Treatment Control Control Period 1996/97-2001/02 1996/97-2001/02 1998/99-2001/02 1998/99-2001/02 1996/97-1997/98 Variable productivity productivity productivity productivity productivity Obs 1238 925 1238 196 729 Sourceμ Authors‘ calculations. 403 Mean 1278,23 1380,41 1278,23 1174,38 1435,80 Std.Dev 2252,89 2051,91 2252,89 792,85 2271,79 Min 9,84 2,25 9,84 16,00 2,25 Max 44212,00 27372,97 44212,00 4864,87 27372,97 18 16 12 5.5 14 6 6.5 EPFRPpct 7 20 7.5 22 Figure A3: Time-series plot of log yield against year and EPFRPpct against year for ages 20-30 1996 1997 1998 1999 2000 2001 year 1998 1999 year 2000 2001 Sourceμ Authors‘ estimations based on the Post Harvest Survey. 4 5 6 7 8 Figure A4: Fitted Quadratic Regression and nonparametric Lowess Regression Curves added to Scatter plot of log yield on the key regressor – EPFRP‟s percentage share of total road network 0 5 10 EPFRPpct 15 logyield lowess logyield EPFRPpct 20 25 Fitted values Sourceμ Authors‘ estimations. Table A10: Matching and Propensity Score Estimators Propensity score matching methods (i) 1. One-to-One propensity score matching (ii) 2. k-nearest neighbors matching (iii) 3. radius matching (iv) 4. kernel (v) 5.local linear regression (vi) 6.'spline-smoothing' (vii) 7. Mahalanobis matching (viii) Variable Sample Treated Controls Difference S.E. productivity Unmatched 1278,227 1380,409 -102,182 94,277 ATT 1291,437 1660,469 -369,032 144,114 ATU 1640,836 1178,661 -462,176 ATE -415,930 productivity Unmatched 1278,227 1380,409 -102,182 94,277 ATT 1383,466 1599,065 -215,600 163,080 ATU 1602,739 1389,113 -213,626 ATE -214,717 productivity Unmatched 1278,227 1380,409 -102,182 94,277 ATT 1420,120 1689,646 -269,526 172,248 ATU 1533,129 1327,329 -205,800 ATE -242,665 productivity Unmatched 1278,227 1380,409 -102,182 94,277 ATT 1192,912 1184,001 8,911 78,477 ATU 1322,991 1317,266 -5,725 ATE 2,734 productivity Unmatched 1278,2270 1380,4087 -102,1817 94,2765 ATT 1126,3095 1196,0222 -69,7127 195,5590 ATU 1246,8842 1235,1213 -11,7629 ATE -44,943443 productivity Unmatched 1278,227 1380,409 -102,182 94,277 ATT 1383,466 1629,728 -246,263 , ATU 1602,739 1303,153 -299,587 ATE -270,129 productivity Unmatched 1278,227 1380,409 -102,182 94,277 ATT 1067,260 1080,069 -12,810 70,558 ATU 1075,650 1085,807 10,157 ATE -2,529 Sourceμ Authors‘ estimations. 404 T-stat -1,080 -2,560 -1,080 -1,320 -1,080 -1,560 -1,080 0,110 -1,0800 -0,3600 -1,080 , -1,080 -0,180 Fig. A5.1a. Histogram of estimated propensity score, Fig. A5.1b. Histogram of estimated propensity score, Controlled 0 0 2 2 4 4 6 6 Treated .2 .4 .6 Estimated propensity score .8 1 .2 .4 .6 Estimated propensity score .8 1 Source: Authors‘ calculations. Table A11: First-Differences (FD) estimator using differenced age cohort data Variable D.EPFRPpct FD1_Total FD2 (1996-98) FD3 (1996-99) FD4 (1996-2000) FD5 (2000-2001) FD6 (1999-2000) FD7 (1998-99) -0.0040 0.0026 -0.0039 -0.0049 0.0273 0.0231 -0.0115 (0.0051) (0.0060) (0.0053) (0.0053) (0.0271) (0.0358) (0.0064) D.Sex -0.7965 -0.7869 -0.7806 -0.7321 -1.0185 -0.8067 -0.6244 (0.5499) (0.6889) (0.7315) (0.6751) (0.7088) (1.0097) (0.7566) D.shareofm~e 0.5067 0.7318 0.5070 0.3536 1.4484 0.3233 0.7062 (0.8878) (0.9429) (1.0024) (1.0840) (1.3107) (1.7356) (1.2883) D.loghhsize -0.0009 0.3901 0.1403 -0.0031 -0.5444 -0.7765 -0.2239 (0.3582) (0.4939) (0.4361) (0.4077) (0.4661) (0.5816) (0.5532) D.stratum 0.2295 0.1676 0.0935 0.2490 1.0261 1.5921 2.1369* (0.5011) (0.7079) (0.5761) (0.6039) (0.9425) (1.1727) (0.9204) D.livestock -0.7136 0.2949 -0.2336 -0.6742 -1.0878 -0.6890 -0.4721 (0.4258) (0.5312) (0.4668) (0.5495) (0.6241) (0.8906) (0.5506) D.basalprha -0.0032 -0.0075 -0.0101 -0.0031 0.0001 0.0043 -0.0010 (0.0070) (0.0115) (0.0077) (0.0076) (0.0130) (0.0165) (0.0096) D.Topdresp~a -0.0033 -0.0004 -0.0003 -0.0040 -0.0060 -0.0168 -0.0127 (0.0080) (0.0118) (0.0086) (0.0088) (0.0151) (0.0211) (0.0115) D.Areapc 0.4385 0.4990 -0.6768 0.4089 0.7340 -0.4278 -2.2099* (0.6315) (0.6205) (0.7391) (0.8572) (0.7748) (1.1847) (1.0193) D.Clandfrac -1.1440 -3.6980** -1.6370 -1.5387 0.4397 -0.0326 -2.6424* (0.8889) (1.2742) (1.1588) (1.0396) (1.5226) (1.3452) (1.1233) D.rain_EP 0.0004 -0.0023* -0.0005 0.0005 0.0005 0.0006 0.0013 (0.0003) (0.0010) (0.0007) (0.0005) (0.0004) (0.0006) (0.0010) N 213 87 130 172 83 85 86 r2 0,069 0,3207 0,1121 0,0684 0,193 0,1502 0,2191 Notes: cluster-robust standard errors in parentheses * p<0.05, ** p<0.01, *** p<0.001 Source: Authors estimations. 405 Table A12a: Determinants of Land Cotton Shares (1) OLS1 Age of Head of Household -0.0038* (0.0016) Age Squared 0.0000 (0.0000) Sex -0.0340*** (0.0059) Male Household Head 0.0361** (0.0128) Household Size 0.0005 (0.0009) Farm Type (stratum) -0.0102 (0.0083) Livestock Area Rainfall District Level Observations R2 F ll (2) (3) (4) OLS2 OLS3 Tobit1 -0.0052 -0.0239* -0.0097 (0.0030) (0.0111) (0.0050) 0.0001 0.0002 0.0001 (0.0000) (0.0001) (0.0001) -0.0213 -0.2441*** -0.1345*** (0.0124) (0.0415) (0.0198) -0.0154 0.2775** 0.1650*** (0.0253) (0.0891) (0.0414) -0.0032* 0.0092 0.0048 (0.0015) (0.0060) (0.0026) -0.0141 -0.0738 -0.0223 (0.0126) (0.0577) (0.0241) 0.0145** -0.0248** 0.1504*** (0.0050) (0.0093) (0.0352) 0.0188*** -0.0027 0.1628*** (0.0017) (0.0027) (0.0117) 0.0000** -0.0000 0.0003** (0.0000) (0.0000) (0.0001) 6586 2013 6699 (5) Tobit2 -0.0650 (0.0368) 0.0004 (0.0004) -0.9774*** (0.1457) 1.2197*** (0.3042) 0.0399* (0.0192) -0.2140 (0.1773) 0.0723*** 0.6031*** (0.0158) (0.1162) 0.0592*** 0.4804*** (0.0049) (0.0358) 0.0002*** 0.0010** (0.0000) (0.0003) 6586 6699 0.0519 0.0222 0.0740 40.0195 5.0550 59.3634 1659.9147 521.0816 -1.139e+04 -3495.8158 -7610.8444 Notes: OLS1 with zeros; OLS2 without zeros; OLS3 on log-transformed dependent variable; Tobit1 on left-censored observations (69%); Tobit2 on lognormal dependent variable. Table A12b: Detailed Summary to show Skewness and Kurtosis of Dependent Variable Fraction of Land Allocated to Cotton uncensored Percentiles Smallest 1% 0 0 5% 0 0 10% 0 0 Obs 6587 25% 0 0 Sum of Wgt. 6587 Mean 0,108 Largest Std. Dev. 0,193 75% 0,166 1 90% 0,429 1 Variance 0,037 95% 0,535 1 Skewness 1,767 99% 0,721 1 Kurtosis 5,317 Logarithm of Fraction of Land Allocated to Cotton uncensored Percentiles Smallest 1% -3,219 -4,119 5% -2,546 -3,753 10% -2,197 -3,551 Obs 2013 50% 0 25% -1,609 50% -1,099 -3,548 75% 90% 95% -0,693 -0,509 -0,401 Largest 0 0 0 99% -0,154 0 Fraction of Land Allocated to Cotton censored Percentiles Smallest 1% 0,040 0,0163 5% 0,078 0,0234 10% 0,111 0,0287 Obs 2013 25% 0,200 0,0288 Sum of Wgt. 2013 Mean 0,353 Largest Std. Dev. 0,189 75% 0,500 1 90% 0,601 1 Variance 0,036 95% 0,669 1 Skewness 0,534 99% 0,857 1 Kurtosis 3,094 Logarithm of Fraction of Land Allocated to Cotton censored Percentiles Smallest 1% -3,219 -4,119 5% -2,546 -3,753 10% -2,197 -3,551 Obs 2013 50% 0,333 Sum of Wgt. 2013 25% -1,609 50% -1,099 Mean Std. Dev. -1,226 0,671 Variance Skewness 0,451 -0,976 75% 90% 95% Kurtosis 3,818 99% Sourceμ Authors‘ calculations. 406 -3,548 Sum of Wgt. 2013 Mean Std. Dev. -1,226 0,671 -0,693 -0,509 -0,401 Largest 0 0 0 Variance Skewness 0,451 -0,976 -0,154 0 Kurtosis 3,818 Table A12c: Determinants of Land Cotton Shares for each survey year (1) (2) (3) Tobit96 Tobit97 Tobit98 Age of HH Head -0.0032 -0.0078 -0.0270* (0.0117) (0.0120) (0.0117) Age Square -0.0000 0.0000 0.0003 (0.0001) (0.0001) (0.0001) Sex of HH Head -0.1673*** -0.1454** -0.1095* (M=1, F=0) (0.0482) (0.0481) (0.0476) Share of Males 0.2024* 0.2213* 0.2742** (0.0957) (0.0981) (0.1012) HH Size 0.0098 0.0168* 0.0008 (0.0059) (0.0066) (0.0064) Stratum -0.0377 -0.1199 0.1163 (0.0507) (0.0640) (0.0621) Livestock -0.0092 0.0065 0.0871* (0.0364) (0.0368) (0.0396) Area 0.0615*** 0.0802*** 0.0548*** (0.0124) (0.0123) (0.0111) Rainfall 0.0003* 0.0018*** -0.0007*** (0.0001) (0.0002) (0.0001) Constant -0.2657 -1.3009*** 0.6858* (0.2764) (0.3046) (0.2679) Sigma Constant 0.4436*** 0.4451*** 0.4590*** (0.0186) (0.0196) (0.0201) Observations 1053 1041 1079 Loglikelihood -584,81 -537,27 -564,40 (4) Tobit99 -0.0062 (0.0141) 0.0000 (0.0002) -0.0752 (0.0553) 0.1228 (0.1188) 0.0063 (0.0067) -0.2006* (0.0956) 0.1383** (0.0457) 0.0847*** (0.0175) -0.0004 (0.0002) 0.0290 (0.3378) (5) Tobit00 -0.0113 (0.0123) 0.0001 (0.0001) -0.1249* (0.0490) 0.1293 (0.0996) -0.0059 (0.0070) 0.0086 (0.0547) 0.1758*** (0.0410) 0.0485*** (0.0122) 0.0001 (0.0001) -0.0940 (0.2776) (6) Tobit01 -0.0031 (0.0111) -0.0000 (0.0001) -0.1283** (0.0417) 0.0801 (0.0902) 0.0063 (0.0059) -0.0518 (0.0504) 0.1167*** (0.0339) 0.0418*** (0.0101) 0.0005** (0.0002) -0.4212 (0.2773) 0.5359*** (0.0281) 1292 -585,55 0.4473*** (0.0216) 982 -496,43 0.4290*** (0.0185) 1139 -584,51 Notes: Separate Tobit regressions are run for each year e.g. to account to macro shocks and prices. Since no information on assets was collected in 2001, none of the Tobit specifications include the value of e.g. ploughs. Standard errors in parentheses: * p<0.05, ** p<0.01, *** p<0.001 Source: Authors Estimations. Table A13a: Model 1: Marginal Effects variable EPFRPpct Age Agesq Sex shareofmale hhsize stratum basal_qty Topdres_qty livestock(*) Area Clandfrac rain_EP Marginal effects after tobit MFX1 dy/dx (s.e.) -.0098016*** (.00187) -.0203272 (.01443) .0002196 (.00017) -.1446019** (.05968) -.1170192 (.12219) -.0004382 (.00736) -.0622634 (.06189) -.0004593 (.00032) .0004847 (.00041) .0960845** (.04526) -.0195372 (.01375) -1.358038*** (.10136) -.0000161 (.00012) 6.6824833 MFX2 dy/dx (s.e.) -.0098016*** (.00187) -.0203272 (.01443) .0002196 (.00017) -.1446019** (.05968) -.1170192 (.12219) -.0004382 (.00736) -.0622634 (.06189) -.0004593 (.00032) .0004847 (.00041) .0960845** (.04526) -.0195372 (.01375) -1.358038*** (.10136) -.0000161 (.00012) 6.6824833 MFX3 dy/dx (s.e.) -.0044233*** (.00084) -.0091732 (.00651) .0000991 (.00007) -.0652554** (.02694) -.052808 (.05515) -.0001978 (.00332) -.028098 (.02793) -.0002073 (.00014) .0002187 (.00018) .0436073** (.02067) -.0088167 (.00621) -.6128507 *** (.0464) -7.26e-06 (.00006) 6.1320435 Tobit1 -.0098016*** (.00187) -.0203272 (.01443) .0002196 (.00017) -.1446019** (.05968) -.1170192 (.12219) -.0004382 (.00736) -.0622634 (.06189) -.0004593 (.00032) .0004847 (.00041) .0960845** (.04526) -.0195372 (.01375) -1.358038*** (.10136) -.0000161 (.00012) Notes: MFX1: The Marginal Effects (ME) for the left-trunctaed mean, E(y|x,y>0); MFX2: The Marginal Effects for the Censored mean, E(y|x); MFX3: Right Censoring at the median value of y; Tobit1: Simple Tobit model in levels. (*) dy/dx is for discrete change of dummy variable from 0 to 1. 407 Figure A6: Deviation from Long-term Rainfall Mean in Natural Logarithm 0,6000 Deviation from long-term mean (in logs) 0,4000 0,2000 Eastern 0,0000 Chadiza (301) (i) -0,2000 Chama (302) (iii) Chipata (303) (i) -0,4000 Katete (304) -0,6000 Lundazi (305) -0,8000 Mambwe (306) (iii) -1,0000 Nyimba (307) (ii) -1,2000 Petauke (308) (ii) -1,4000 -1,6000 Sourceμ Authors‘ calculations. 0 2 4 6 8 10 Figure A7: Relationship between log cotton productivity and Rain deviation from LT-mean -.6 -.4 -.2 dlnrain_EP Sourceμ Authors‘ calculations. 408 0 .2 Table A14: Effects of EPFRP on Log Yield using method based on the propensity score (1) EPFRP -0.0185* (0.0109) 0.0002 (0.0001) 0.0589 (0.0397) 0.2974*** (0.0373) -0.1761*** (0.0533) 0.0075 (0.0344) 0.4155*** (0.0605) -0.5821*** (0.0833) 0.0011*** (0.0001) Age Agesq Sex loghhsize stratum livestock Areapc Clandfrac rain_EP EPFRP (2) logyield -0.1384*** (0.0439) 1.3710*** (0.2149) Propensity (3) logyield -0.0165 (0.0147) 0.0002 (0.0002) -0.1403** (0.0596) -0.0762 (0.0547) -0.0548 (0.0595) 0.0992** (0.0455) -0.1288* (0.0719) -1.3384*** (0.1017) 0.0001 (0.0001) -0.1765*** (0.0429) (4) logyield (5) logyield -0.0752* (0.0432) 7.8485*** (0.3196) 2163 6.7255*** (0.0327) 2163 -0.1901*** (0.0462) 2.4660*** (0.3757) -1.6234*** (0.4574) 5.3908*** (0.2059) 2163 EPFRP*(propensity-r(mean)) constant Number of Observations R2 F ll -0.6368*** (0.2364) 6586 - 4 344,48 5.9835*** (0.1207) 2163 0,02 21,90 - 3 035,40 0,08 19,12 - 2 965,03 0,00 3,03 - 3 055,59 0,03 18,87 - 3 029,11 Notes: Standard errors in parentheses * p<0.10, ** p<0.05, *** p<0.01. Source: Author based on Wooldridge(2002:619). Table A.15: One-to-One Matching: Sample characteristics and estimated impacts Control Sample Probit No. Of Mean Propensity Observation Score (ii) 5276 0,52088 Logit Without replacement: Random 5276 925 Low to high 925 High to low 925 With replacement: Nearest neighbor 1215 Caliper, 0.00001 48 Caliper, 0.00005 211 Caliper, 0.0001 408 0,47465 Treatment Regression Effect (Diff. In Treatment Clandfrac Means) (iii) Effect (iv) -0,218 0,088 -0,434 0,174 Age -0,021 0,012 -0,049 0,022 Agesq 0,000 0,000 0,000 0,000 Sex 0,084 0,045 0,123 0,081 loghhsize 0,329 0,041 0,545 0,078 stratum -0,202 0,059 -0,214 0,114 livestock 0,085 0,037 0,086 0,069 Areapc 0,433 0,066 0,921 0,163 -0,012 0,020 -0,012 0,020 -0,012 0,020 0,000 0,000 0,000 0,000 0,000 0,000 0,226 0,081 0,226 0,081 0,226 0,081 0,147 0,073 0,147 0,073 0,147 0,073 -0,021 0,080 -0,021 0,080 -0,021 0,080 0,252 0,059 0,252 0,059 0,252 0,059 0,166 0,098 0,166 0,098 0,166 0,098 -1,087 0,135 -1,087 0,135 -1,087 0,135 -0,120 0,045 -0,120 0,045 -0,030 0,047 -0,012 0,020 -0,012 0,020 -0,012 0,020 0,000 0,000 0,000 0,000 0,000 0,000 0,226 0,081 0,226 0,081 0,226 0,081 0,147 0,073 0,147 0,073 0,147 0,073 -0,021 0,080 -0,021 0,080 -0,021 0,080 0,252 0,059 0,252 0,059 0,252 0,059 0,166 0,098 0,166 0,098 0,166 0,098 -1,087 0,135 -1,087 0,135 -1,087 0,135 -0,216 0,063 -0,270 0,167 -0,205 0,100 -0,012 0,020 0,000 0,000 0,226 0,081 0,147 0,073 -0,021 0,080 0,252 0,059 0,166 0,098 -1,087 0,135 -0,174 0,077 Notes: (i) Variables: shareofmale; basalprha; Topdresprha and rain_EP are not balanced and therefore left out of the specification. (ii) The propensity score is estimated using a logit of treatment status on. Source: Author estimation based on psmatch2 (Leuven and Sianesi, 2003) available from ssc desc psmatch2. 409 Table A16: Estimation of ATEs using Different Matching Methods Number of (ii) Analytical Bootstrapped ATT estimation with Treated Controls ATT Std.Err. t Std.Err. t Nearest Neighbor 3870 530 -0.180 0.052 -3.447 0.080 -2.256 Matching method Radius Matching method Kernel Matching method 3870 2716 -0.199 n.a.(i) n.a. 0.050 -3.959 Stratification method 3869 2717 -0.298 0.026 -11.497 0.056 -5.360 Notes: (i) Analytical standard errors cannot be computed; (ii) Refer to actual nearest neighbour matches. 410 Annex: Chapter 5 Web Links Partnership in Statistics for Development in the 21st Century (PARIS21) http://www.paris21.org/ The International Household Survey Network (IHSN) http://www.internationalsurveynetwork.org/home/ Country questionnaires (database)317 http://www.internationalsurveynetwork.org/home/?lvl1=tools&lvl2=questionnaire&lvl3= country# Other survey databanks, incl. IFPRI http://www.internationalsurveynetwork.org/home/?lvl1=activities&lvl2=catalog&lvl3=da tabank Central Statistical Office. 2008. Poverty in Zambia -1991 – 2006. http://www.zamstats.gov.zm/lcm.php World Bank. 2008. Africa Household Survey Databank. http://www4.worldbank.org/afr/poverty/databank/default.cfm World Bank. 2008. ZAMBIA, 1996 & 1998. Living Conditions Monitoring Survey (LCMS). This CD-ROM contains 2 surveys. http://www4.worldbank.org/afr/poverty/databank/cdroms/in_stock_zmb_96_98.cfm?cd= zmb_96_98&CFID=2260647&CFTOKEN=99110ce0ffa34585-A8C54B66-06F7-FAC618EBE63C159DC9B4&jsessionid=98302753a9902a643156 UNCTAD electronic portal on commodities http://www.unctad.org/infocomm/ Information on cotton http://www.unctad.org/infocomm/anglais/cotton/sitemap.htm International Cotton Advisory Committee (ICAC) http://www.icac.org/ UNCTAD Commodity Price Statistics on-line http://www.unctad.org/Templates/Page.asp?intItemID=1889&lang=1 http://stats.unctad.org/handbook/ReportFolders/ReportFolders.aspx?CS_referer=&CS_C hosenLang=en STATA http://www.stata.com 317 Zambia, 1950-2008: Found 32 questionnaire(s) in 15 survey(s). 411 Table A1: Eastern Province Districts codes Sampled Households District 1998 2004 Code Catchment Control 301 Yes No 120 102 302 No Yes 90 23 303 Yes No 286 199 304 Yes No 150 139 305 Yes No 170 209 306 No Yes 135 74 307 No Yes 120 85 308 Yes No 235 168 1998 961 345 1306 2004 817 182 999 Total Source: CSO. δiving ωonditions εonitoring Survey II 1998 data user‘s guide. District Name Chadiza Chama Chipata Katete Lundazi Mambwe Nyimba Petauke Household‟s unique identification (1998): This variable is a concatenation of the first variables of the Survey minus the panel number. hid = province+district+const+ward+csa+sea+rururb+stratum+centrlty+hhn Household‟s unique identification (2004): This variable is a concatenation of the first variables of the Survey minus the centrality and the panel number. hid = province+district+const+ward+csa+sea+rururb+stratum+hhn 412 Map A1: The major soil types of Eastern Zambia Source: Samuel Simute, C. L. Phiri, and Bo Tengnäs, 1998: Agroforestry Extension Manual for Eastern Zambia. 413 Map A2: Agroecological zones and agricultural districts, Eastern Province, Zambia Source: ARPT (Adaptive Research Planning Team), Eastern Province Agricultural Development Project, "Annual Report, 1985-86" (Chipata, Zambia, 1986, mimeographed). 414 Table A.1.1: Poverty Lines, Current Prices, Zambian Kwacha (ZMK) OER UPL LPL CPI CPI Ch 1996 1275 28979 20181 175 43,1 1998 2195 47187 32861 100 24,5 2003 4737 92185 64530 87,3 21,4 2004 4848 111747 78223 73,5 18 Notes: Official Exchange Rate (OER=USD/ZMK); Upper Poverty Line (UPL) and Lower Poverty Line (LPL) per month; Consumer Price Index (CPI) (1998=100); CPI Annual percentage change (Ch). Sourceμ Author‘s calculations based upon the LCMS II and LCMS IV datasets. Table A.1.2: Poverty Lines, Constant Prices, Zambian Kwacha (ZMK) Implicit GDP Deflator Implicit GDP 2000=100* Deflator 1998=100 OER UPL LPL 1996 1919,18 43620,38 30377,27 42,10 66,43 1998 2195,00 47187 32861 63,37 100,00 2003 1681,57 32724,45 22907,29 178,51 281,70 1 2004 1428,82 32934,38 23054,1 215,01 339,30 28 Sourcesμ Author‘s calculations. * IεF, World Economic ηutlook (WEη) Database. 20 Table A.2: Rural Summary statistics on Outcome Indicators 1998 Treatment Control 64853,44 29071,57 (1184530) (47889.78) 962 346 15076,05 18845,14 (34193.13) (29965.03) 960 343 15076,05 18845,14 (34193.13) (29965.03) 960 343 15596,61 19603,87 (34647.19) (31105) 958 342 9 2004 Percentage change 6 Treatment Control Treatment Control 59651,33 46688,28 -8% 61% (250051.9) (65130.34) 817 182 37570,73 27289,34 149% 45% (53322.62) (40485.26) 817 182 58276,76 42329,07 287% 125% (82709.85) (62797.55) 817 182 59891,14 43848,03 284% 124% (86171.3) (64447.27) 817 182 Mean Income** Standard Deviation Observations Mean Consumption Standard Deviation* Observations Mean Consumption** Standard Deviation Observations Mean P.A.E. Consumption** Standard Deviation Observations Income Poverty Rate Upper Poverty Line = ZMK47187 0,918 0,917 0,550 0,655 -40% Lower (Food) Poverty Line = ZMK32861 0,855 0,853 0,404 0,489 -53% Consumption Poverty Rate Upper Poverty Line = ZMK47187 0,952 0,955 0,882 0,769 -7% Lower (Food) Poverty Line = ZMK32861 0,917 0,917 0,647 0,777 -29% Notes: * Using the CPI04 = 335,5 Deflator; ** Using the Foodbasket = 216,3 Deflator. What we are measuring when we do not take into account survey design is the moments (mean and standard deviation) of the sample distribution whist when survey design is been applied we get the moments of the population. Sourceμ Author‘s calculations based upon the δωεS II and LCMS IV datasets. 415 -29% -43% -19% -15% Table A3.a: Principal Economic Activity of Household Head, Rural Areas, Numbers of Household Heads by Quintile of Consumption, Catchment Districts 1 2 3 4 5 6 7 8 9 10 Main economic activities status In Wage Employment Running Business / Self-Employed Farming, Fishing, Forestry Not working but looking for works / Means to do business Not working not looking for works / Means to do business, but available to do so Full Time student Full Time at home / home duties Retired Too old to work Other Total All 2,83% 1,47% 65,45% 1,47% 1998 Lowest 20% Highest 40% 0,00% 6,64% 1,29% 4,15% 68,24% 57,26% 0,00% 2,49% 0,52% 14,03% 7,23% 0,00% 1,68% 5,34% 956 0,86% 10,73% 8,58% 0,00% 4,29% 6,01% 233 0,00% 18,67% 7,05% 0,00% 0,83% 2,90% 241 Sourceμ Author‘s calculations based upon the δωεS II and δωεS IV datasets. All 3,83% 1,36% 94,68% 0,00% 0,00% 0,12% 0,00% 0,00% 0,00% 0,00% 809 2004 Lowest 20% Highest 40% 3,24% 4,70% 0,54% 1,25% 95,68% 94,04% 0,00% 0,00% 0,00% 0,54% 0,00% 0,00% 0,00% 0,00% 185 0,00% 0,00% 0,00% 0,00% 0,00% 0,00% 319 Table A3.b: Principal Economic Activity of Household Head, Rural Areas, Numbers of Household Heads by Quintile of Consumption, Control Districts 1 2 3 4 5 6 7 8 9 10 Main economic activities status In Wage Employment Running Business / Self-Employed Farming, Fishing, Forestry Not working but looking for works / Means to do business Not working not looking for works / Means to do business, but available to do so Full Time student Full Time at home / home duties Retired Too old to work Other Total All 4,35% 7,25% 56,81% 0,29% 1998 Lowest 20% Highest 40% 0,00% 0,00% 9,33% 6,36% 60,00% 40,91% 0,00% 0,00% 0,00% 17,39% 8,41% 0,00% 0,87% 4,64% 345 0,00% 16,00% 8,00% 0,00% 0,00% 6,67% 75 0,00% 10,91% 5,45% 0,00% 0,00% 4,55% 110 Sourceμ Author‘s calculations based upon the δωεS II and δωεS IV datasets. All 8,33% 8,33% 27,22% 0,00% 0,00% 11,11% 4,44% 0,00% 0,00% 1,67% 180 2004 Lowest 20% Highest 40% 14,81% 12,24% 5,56% 2,04% 303,70% 93,88% 7,41% 2,04% 0,00% 1,85% 0,00% 0,00% 0,00% 0,00% 54 0,00% 0,00% 0,00% 0,00% 0,00% 0,00% 49 Table A4.a: Pct. of Households in Rural Catchment Areas Owning Particular Assets by Quintile 1 2 3 4 5 6 7 8 9 10 As s et Owners hip plough crop s prayer bicycle motorcycle motorvehicle tractor radio telephone s cotch cart donkey Total All 79,36% 85,71% 48,67% 98,19% 96,15% 99,14% 50,94% 98,59% 89,40% 99,37% 1274 1998 Lowes t 20% 81,86% 92,83% 62,03% 100,00% 99,58% 100,00% 75,95% 100,00% 91,56% 100,00% 237 Highes t 40% 80,57% 79,35% 43,52% 96,56% 90,89% 97,77% 29,15% 96,56% 88,66% 99,19% 494 All 66,71% 74,42% 41,25% 99,76% 97,67% 2004 Lowes t 20% 56,99% 70,43% 26,88% 100,00% 94,62% Highes t 40% 73,37% 77,09% 39,94% 99,69% 99,69% 81,76% 98,65% 817 78,49% 97,31% 186 87,62% 99,07% 323 Sourceμ Author‘s calculations based upon the δωεS II and δωεS IV datasets. Table A4.b: Pct. of Households in Rural Control Areas Owning Particular Assets by Quintile 1 2 3 4 5 6 7 8 9 10 Asset Ownership plough crop sprayer bicycle motorcycle motorvehicle tractor radio telephone scotch cart donkey Total All 93,22% 90,04% 52,12% 97,88% 97,03% 98,94% 49,15% 97,88% 96,40% 99,36% 472 1998 Lowest 20% Highest 40% 94,94% 90,55% 92,41% 88,56% 69,62% 45,27% 100,00% 96,02% 100,00% 94,03% 100,00% 97,51% 74,68% 30,85% 100,00% 95,02% 97,47% 94,53% 100,00% 98,51% 79 201 All 75,27% 80,22% 36,81% 98,90% 98,90% 84,07% 97,80% 182 Sourceμ Author‘s calculations based upon the LCMS II and LCMS IV datasets. 416 2004 Lowest 20% Highest 40% 77,78% 81,63% 79,63% 79,59% 29,63% 46,94% 96,30% 100,00% 96,30% 100,00% 83,33% 77,78% 54 91,84% 100,00% 49 Table A7.b: Descriptive Statistics of covariates, 1998 and 2004, Catchment Districts Type CV CV DV CV DV DV DV DV DV CV CV DV Variable Name Log pae monthly household expenditure Cotton Sales share of household income Stratum, excl. Large AHH Distance to Inputmarket EPFRP Treatment Plough Ownership Bicycle Ownership Scotchcart Ownership Motorvehicle Ownership Age of Head of Household Age Squared Head of HH ever attended School Variable Obs Mean 1998 Std.Dev. LNPAE98 948 8,942 1,215 cotincshare stratum124 Distiput infrastructure Plough Bicycle Scotchcart Motorvehicle Age Agesq s4q5 929 958 964 964 964 964 964 964 961 961 720 Min. Max. Obs Mean 3,912 13,534 817 10,338 0,110 0,207 0,010 1,553 1,049 1 23,498 22,825 0 0 0 0 0,756 0,430 0 0,470 0,499 0 0,880 0,326 0 0,975 0,156 0 43,751 15,843 15 2164,902 1545,234 225 0,461 0,499 0 1 4 99 0 1 1 1 1 99 9801 1 2004 Std.Dev. Min. 1,169 817 0,283 0,332 817 1,206 0,431 817 13,742 18,493 817 1 0 817 0,667 0,472 817 0,330 0,471 817 0,818 0,386 817 0,977 0,151 817 42,765 14,984 817 2053,103 1459,178 815 0,237 0,425 Max. 6,563 13,771 Percentage change 16% 0 1 0 1 0 0 0 0 20 400 0 1 4 99 1 1 1 1 1 90 8100 1 157% n.a. -42% n.a. n.a. n.a. n.a. n.a. -2% -5% n.a. 2004 Std.Dev. Min. Max. Percentage change Catchment Districts: Chadiza, Chipata, Katete, Lundazi, and Petauke districts. Sourceμ Author‘s calculations based upon the δωMS II and LCMS IV datasets. Table A7.c: Descriptive Statistics of covariates, 1998 and 2004, Control districts Type CV CV DV CV DV DV DV DV DV CV CV DV Variable Name Log pae monthly household expenditure Cotton Sales share of household income Stratum, excl, Large AHH Distance to Inputmarket EPFRP Treatment Plough Ownership Bicycle Ownership Scotchcart Ownership Motorvehicle Ownership Age of Head of Household Age Squared Head of HH ever attended School Variable Obs Mean 1998 Std.Dev. LNPAE98 339 9,146 1,276 Cotincshare Stratum124 Distiput Infrastructure Plough Bicycle Scotchcart Motorvehicle Age Agesq s4q5 345 346 347 347 347 347 347 347 345 345 248 Min. Max. Obs Mean 4,605 12,662 182 10,090 0,085 0,185 0,010 1,390 0,958 1 20,452 23,208 0 0 0 0 0,916 0,277 0 0,519 0,500 0 0,960 0,197 0 0,980 0,141 0 41,110 15,325 20 1924,217 1442,787 400 0,306 0,462 0 1 4 99 0 1 1 1 1 86 7396 1 1,101 182 0,336 0,362 182 1,137 0,345 182 4,945 9,793 182 0 0 182 0,753 0,433 182 0,368 0,484 182 0,841 0,367 182 0,989 0,105 182 43,440 15,232 182 2117,725 1491,381 182 0,126 0,333 6,629 13,226 0 1 0 0 0 0 0 0 21 441 0 1 2 43 0 1 1 1 1 80 6400 1 Control districts: Chama, Nyimba, and Mambwe. Sourceμ Author‘s calculations based upon the δωεS II and δωεS IV datasets. Table A9: Ramsey's null hypothesis of no omitted variables for the model 1998 F(3, 1236)* Using powers of the fitted values of Lnpae 2.48 F(7, 1232)*** 7.21 Using powers of the independent variables F(3, 907) 0.78 Using powers of the fitted values of Lnpae F(10, 900)*** 3.99 Using powers of the independent variables Sourceμ Author‘s calculations based upon the δωεS II and δωεS IV. 417 2004 F(3, 640) 1.99 F(7, 636)** 2.60 F(3, 635) * 2.22 F(10, 628) ** 2.18 10% 294% n.a. -76% n.a. n.a. n.a. n.a. n.a. 6% 10% n.a. Table A10.1: Quantile 2 (40%), 1998 stratum124 cotincshare (1) (2) .0306036 (.0291806) observations 183 distiput (3) LNpae98 .0838091 -.0020793* (.1214851) (.0010685) 183 183 LNpae98 .1052395 -.0012672* (.0847555) (.0007452) 183 183 LNpae98 2.930644 -.0017576 (4.380422) (.0021677) 183 183 -.0046445 (.0192763) observations 183 .0542128 (.1051758) observations 183 s10q8 (4) infrastruc~e s1q3b age2 (5) (6) (7) Panel A: Quantile Regression .1379662** .0440118 .0217552** -.0002333** (.0656821) (.0514329) (.0089519) (.0000917) 183 183 183 183 Panel B: OLS Regression .1649384 .0215499 .0165729*** -.0001865*** (.2112359) (.034651) (.0061266) (.000063) 183 183 183 183 Panel C: IV Regression -.0475819 .0696887 .0073809 -.0000588 (.6629515) (.1202833) (.0219358) (.0002619) 183 183 183 183 Sourceμ Author‘s calculations based upon the δωεS II. s10q6 (8) s4q5 (9) _cons (10) .0230573 -.0544589 7.756616 (.0490552) (.0487617) (.2266492) 183 183 183 .033443 -.0011833 7.893203 (.0330267) (.0333019) (.2580066) 183 183 183 .2413556 .141309 7.700467 (.3342344) (.2385394) (.7636825) 183 183 183 Table A10.2: Quantile 2 (40%), 2004 LNpae04 observations LNpae04 observations LNpae04 observations stratum124 cotincshare distance11 assetownd09 infrastruc~e s1q3b age2 assetownd07 s4q5 _cons (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Panel A: Quantile Regression .0111791 -.0643832 .000694 -.0299201 .0283091 .0036939 -.0000392 -.098744 -.1200001 10.45 (.0710609) (.0854265) (.0017299) (.1451867) (.0782728) (.0118197) (.0001212) (.0659405) (.0712755) (.3413457) 122 122 122 122 122 122 122 122 122 122 Panel B: OLS Regression .0035191 -.0071594 .0003387 -.0581507 .0200911 -.0003744 .0000113 -.0635094* -.1091965*** 10.5874 (.0380326) (.0451215) (.0009131) (.1184886) (.0418787) (.0061095) (.0000628) (.0356337) (.0369827) (.1987075) 121 121 121 121 121 121 121 121 121 121 Panel C: IV Regression .0013715 .0157283 .000291 -.0664228 .0212123 -.0003076 .0000113 -.061623 -.1050834** 10.58732 (.0421088) (.1968866) (.0009977) (.1373647) (.0429653) (.0061421) (.0000629) (.0390149) (.050566) (.1989388) 121 121 121 121 121 121 121 121 121 121 Sourceμ Author‘s calculations based upon the δωεS IV. Table A.10.3:Quantile 3 (60%), 1998 stratum124 cotincshare (1) (2) LNpae98 .0131241 (.0170845) observations 179 LNpae98 observations LNpae98 observations distiput (3) .0182534 (.0645665) 179 -.0004084 (.0007415) 179 .018657 (.0146426) 179 -.0039638 (.0544468) 179 .0002326 (.0005954) 179 .013529 .0172762 179 -.1442804 .2444354 179 .0002699 .0006102 179 s10q8 (4) infrastruc~e s1q3b (5) (6) Panel A: Quantile Regression (i) .0773357** .0111856** (.0328427) (.0051563) 179 179 179 Panel B: OLS Regression (dropped) .0733414*** .0080515* (.0282064) (.0045343) 179 179 179 Panel C: IV Regression (dropped) .0760096*** .0084445* .0291063 .0046699 179 179 179 age2 (7) s10q6 (8) s4q5 (9) _cons (10) -.000118** (.0000518) 179 .0118405 -.0285177 8.756667 (.0303626) (.0316825) (.130224) 179 179 179 -.0000778* (.000046) 179 -.0158918 -.0400004 8.760756 (.0256738) (.0259596) (.1119348) 179 179 179 -.0000842* .0000481 179 -.0191792 -.0460521 8.786202 .0267582 .0283837 .1219941 179 179 179 Note: (i) s10q8 dropped due to collinearity. Table A10.4:Quantile 3 (60%), 2004 LNpae04 observations LNpae04 observations LNpae04 observations stratum124 cotincshare (1) (2) distiput (3) -.1210904** .0650898 (.0578442) (.0704968) 120 120 -.0007462 (.0009632) 120 -.0771451* (.0434194) 120 .0281332 (.0533395) 120 .0001323 (.0007735) 120 -.0274513 (.1070676) 120 .3761425 (.6617499) 120 .0010553 (.0019704) 120 s10q8 (4) infrastruc~e s1q3b age2 (5) (6) (7) Panel A: Quantile Regression -.0753203 -.1920056*** .0036829 -.0000166 (.1274591) (.0671596) (.0098971) (.0001027) 120 120 120 120 Panel B: OLS Regression -.112156 -.1140004** -.0061481 .0000622 (.1044471) (.0520285) (.0078154) (.0000825) 120 120 120 120 Panel C: IV Regression -.1658994 -.1568088 -.0070213 .0000763 (.1596262) (.1015923) (.0093515) (.0001008) 120 120 120 120 Sourceμ Author‘s calculations based upon the δωεS IV. 418 s10q6 (8) s4q5 (9) _cons (10) .0100093 (.0483314) 120 -.0253618 11.4021 (.0496774) (.2700357) 120 120 .0055806 (.0369743) 120 .0063169 11.50263 (.0397268) (.2137675) 120 120 .0601966 (.1121809) 120 -.0007482 11.42693 (.0486602) (.2896831) 120 120 The response of rural household pae to changes in covariates, using one year lagged rainfall as an instrument for household total income, Table A11.1: 1998 Table A11.2: 2004 (1) (2) (3) (4) (5) (6) (7) (8) (1) (2) (3) (4) (5) (6) (7) (8) cotincshare -1.155389 -.8419328 -.9221791 -.7696741 -.8154724 -.774537 -1.134514 -.9443361 cotincshare 48.082 19.07374 18.74465 18.90642 20.14417 20.06111 56.85585 25.23182 (.9183744) (.871716) (.8619486) (.7984894) (.7807687) (.7721495) (.7808524) (.7749301) (602.5159) (48.60581) (46.7002) (46.14526) (52.30744) (51.80792) (410.6017) (78.44945) stratum124 .2498329*** .2539297*** .2458618*** .2495341*** .2255256*** .2352431*** .2448666*** .2234603*** stratum124 -2.526446 -1.223312 -1.477908 -1.490572 -1.99464 -1.882525 -4.445558 -1.894739 (.0430891) (.0422414) (.0418052) (.0408097) (.0403459) (.0399179) (.0389216) (.0442302) (31.0412) (2.828579) (3.496587) (3.463053) (5.08833) (4.79789) (31.83854) (5.802258) di s t i p ut .0089651 .0091082 .0091843 .0120271 .012552 .0349126 .0117013 distiput -.0047652*** -.0047825*** -.0046938*** -.0053502*** -.005375*** -.0057705*** -.0036463** (. 0286511) (.0284554) (.0281838) (.0371967) (.0381299) (.2637287) (.0438474) (.0016435) (.0016205) (.0016029) (.0015733) (.0015736) (.0015543) (.0017596) s10q8 -3.685252 -3.721869 -3.98182 -4.082824 -12.71405 -5.3806 s10q8 -1.665408*** -1.661231*** -1.664813*** -1.652807*** -1.50967*** -1.645161*** (10.7666) (10.64287) (11.98193) (12.1471) (96.06082) (18.64155) infrastruc~e .0078374 .1514407 .0398085 -.1452633 -.0308442 (.2647215) (.2635249) (.2572347) (.2566613) (.2512013) (.2934461) (.6860589) (.8639049) (.7511902) (2.171083) (.9074778) infrastruc~e -.0559229 -.0633752 -.0584501 -.0239623 -.0208506 s1q3b .055252 -.1423722 -.3531945 -.1686533 (.0778764) (.0760542) (.0759079) (.0743726) (.0872116) s1q3b -.0172192*** -.0013572 -.006715 -.0133198 (.1591745) (.3262393) (2.40122) (.4730236) (.0022048) (.01268) (.0124171) (.0138254) age2 .0020507 .0052984 .0024084 age2 -.0001667 -.0001123 -.0000347 (. 0049495) (.0370629) (.0070666) (.0001327) (.0001298) (.0001426) s10q6 3.241787 1.38617 s10q6 -.5411348*** -.403476*** (23.53907) (4.40603) (.075096) (.0835142) s4q5 1.154893 s4q5 -.4325963*** (3. 731657) (.0798648) Observations 999 649 649 649 649 649 649 647 1245 1245 1245 1245 919 Observations 1246 1245 1245 R2 n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. 0.0743 0.1531 R2 0.0178 0.0400 0.0677 0.1185 0.1209 0.1744 Sourceμ Author‘s calculations based upon the δωεS II and δωεS IV. 419 Figure A8.1: Non-Parametric and Quadratic function, 1998 baseline dataset Non-parametric function Quadratic function 6 5.8 5.6 5.4 5.2 5 0 .2 .4 .6 .8 1 cotton sales share of total household income Sourceμ Author‘s calculations based upon the δωεS II. Figure A8.2: Non-Parametric and Quadratic function, 2004 dataset Non-parametric function Quadratic function 6 5.8 5.6 5.4 5.2 5 0 .2 .4 .6 cotton sales share of total household income Sourceμ Author‘s calculations based upon the δωεS IV. 420 .8 1 Table A12: Fixed-effects (within) Panel IV regression with Cotton Income instrumented by external instrument (distance to input market) cotincshare highgrade rain rainlagged Landowners~c cotincshar~q distbank _cons R2 within R2 between R2 overall corr(u_i, Xb) N of obs N of groups Catchment Coef. Std.Err. 10.06956 20.95206 0.0721656 0.07569 -0.0250642 0.057645 0.0029041 0.065073 0.1493332 0.105086 -9.979608 22.84006 0.0068756 0.010582 9.706605 2.433925 0.7095 0.0145 0.4991 -0.2025 90 45 Obs per group Wald chi2(7) Prob>chi2 Sigma_u Sigma_e rho F tests that all u_i=0: F(44,38) Prob>F 2 43245.28 0 0.33814856 0.43122517 0.38076791 0.9 0.6405 Control Coef. Std.Err. -0.9114254 25.86078 -0.0020926 0.0889257 0.0078505 0.0504679 0.0429133 0.0995224 -0.0609962 0.332604 1.427766 27.42889 -0.0077815 0.0147865 5.738843 5.690627 0.4951 0.2798 0.3574 0.0215 78 44 1.8 12338.95 0 0.53700608 0.74367172 0.34272354 0.65 0.8993 Notes: Instrumented: cotincshare. Instruments: highgrade rain rainlagged Landownershippc cotincsharesq distbank distiput. Sourceμ Authors‘ estimations. 421 Annex: Chapter 6 Table A.1a: Equivalence Scale for The Calculation of p.a.e. Expenditure Age 0 1 2 3-4 5-6 7-9 10-11 12-13 14-15 16-17 18-29 30-59 60+ Table A.1b: Latham equivalence scales Weight Male Female 0,33 0,33 0,46 0,46 0,54 0,54 0,62 0,62 0,74 0,7 0,84 0,72 0,88 0,78 0,96 0,84 1,06 0,86 1,14 0,86 1,04 0,8 1 0,82 0,84 0,74 Weight Male Female 0,4 0,4 0,48 0,48 0,56 0,56 0,64 0,64 0,76 0,76 0,8 0,88 1 1 1,2 1 1 0,88 0,88 0,72 Age 0-2 3-4 5-6 7-8 9-10 11-12 13-14 15-18 19-59 > 60 Source: The equivalence scale is based on a World Health Organization equivalence scale. Source: (Latham, 1965). Table A.2a: Overall Poverty in Zambia‟s Eastern Province Poverty Headcount Rate (P0) Poverty Gap (P1) Pooled 2004 1998 change(i) Pooled 2004 Squared Poverty Gap (P2) 1998 change(i) Pooled 2004 1998 change(i) Poverty line = 33562.4(ii) Urban 54,1 46,4 69,6 23,2 24,5 21,4 30,8 9,4 14,7 13,0 18,1 5,1 Rural 74,3 49,7 94,4 44,7 50,3 24,7 71,4 46,8 39,2 15,4 58,6 43,2 Total 71,4 49,1 92,2 43,1 46,7 24,0 67,7 43,7 35,7 15,0 55,0 40,0 Poverty line = 24164.9(iii) Urban 38,6 33,0 50,0 17,1 15,7 13,9 19,3 5,4 9,4 8,5 11,2 2,7 Rural 65,8 38,6 88,1 49,4 42,5 17,1 63,3 46,2 32,1 10,1 50,1 40,0 Total 62,0 37,5 84,7 47,1 38,7 16,5 59,4 42,9 28,9 9,8 46,5 36,8 Notes: (i) Changes shown between years 2004 and 1998. (ii) Total Poverty Line. (iii) Food Poverty Line. The nominal 2004 p.a.e. figure is deflated into a real 1998 p.a.e. measure by dividing the nominal 2004 values by a CPI based on the foodbasket equal to 216.3 and multiplying it with 100. Source: Authors' calculations using Adept version 4.1. Table A.2b: Overall Poverty in Zambia‟s Eastern Province Poverty Headcount Rate (P0) Poverty Gap (P1) Squared Poverty Gap (P2) Pooled 2004 1998 change(i) Pooled 2004 1998 change(i) Pooled 2004 1998 change(i) Poverty line = 33562.4(ii) Urban 67,5 66,5 69,6 3,1 32,7 33,6 30,8 -2,8 20,6 21,9 18,1 -3,8 Rural 82,2 67,4 94,4 27,0 55,9 37,0 71,4 34,4 43,4 24,9 58,6 33,8 Total 80,2 67,2 92,2 25,0 52,6 36,4 67,7 31,4 40,2 24,3 55,0 30,7 Urban 50,8 51,2 50,0 -1,2 22,6 24,2 19,3 -4,9 13,7 15,0 11,2 -3,7 Rural 72,9 54,4 88,1 33,7 47,2 27,5 63,3 35,8 35,4 17,6 50,1 32,5 Total 69,8 53,8 84,7 30,9 43,7 26,9 59,4 32,5 32,3 17,1 46,5 29,5 Poverty line = 24164.9(iii) Notes: (i) Changes shown between years 2004 and 1998. (ii) Total Poverty Line. (iii) Food Poverty Line. The nominal 2004 p.a.e. figure is deflated into a real 1998 p.a.e. measure by dividing the nominal 2004 values by a CPI equal to 335.51 and multiplying it with 100. Source: Authors' calculations using Adept version 4.1. 422 Table A.γa: Poverty by Treatment and Control Regions in Zambia‟s Eastern Province Poverty Headcount Rate Distribution of the Poor Distribution of Population Pooled 2004 1998 change(i) Pooled 2004 1998 change(i) Pooled 2004 1998 change(i) Poverty line = 33562.4(ii) Urban 54,1 46,4 69,6 23,2 10,7 18,5 6,8 -11,7 14,1 19,6 9,0 -10,5 Rural 74,3 49,7 94,4 44,7 89,3 81,5 93,2 11,7 85,9 80,4 91,0 10,5 Control 79,2 64,4 92,2 27,7 17,9 20,6 16,6 -4,0 16,1 15,7 16,6 0,9 Treatment 69,9 46,2 92,2 46,0 82,1 79,4 83,4 4,0 83,9 84,3 83,4 -0,9 Total 71,4 49,1 92,2 43,1 100,0 100,0 100,0 0,0 100,0 100,0 100,0 0,0 Urban 38,6 33,0 50,0 17,1 8,8 17,2 5,3 -11,8 14,1 19,6 9,0 -10,5 Rural 65,8 38,6 88,1 49,4 91,2 82,8 94,7 11,8 85,9 80,4 91,0 10,5 Control 69,1 55,1 81,3 26,3 18,0 23,0 15,9 -7,1 16,1 15,7 16,6 0,9 Treatment 60,6 34,3 85,3 51,0 82,0 77,0 84,1 7,1 83,9 84,3 83,4 -0,9 Total 62,0 37,5 84,7 47,1 100,0 100,0 100,0 0,0 100,0 100,0 100,0 0,0 Infrastructure Treatment Poverty line = 24164.9(iii) Infrastructure Treatment Notes: (i) Changes shown between years 2004 and 1998. (ii) Total Poverty Line. (iii) Food Poverty Line. The nominal 2004 p.a.e. figure is deflated into a real 1998 p.a.e. measure by dividing the nominal 2004 values by a CPI based on the foodbasket equal to 216.3 and multiplying it with 100. Source: Authors' calculations using Adept version 4.1. Table A.3b: Poverty by Treatment and Control Regions in Zambia‟s Eastern Province Poverty Headcount Rate Distribution of the Poor Distribution of Population Pooled 2004 1998 change(i) Pooled 2004 1998 change(i) Pooled 2004 1998 change(i) Poverty line = 33562.4(ii) Urban 67,5 66,5 69,6 3,1 11,9 19,4 6,8 -12,5 14,1 19,6 9,0 -10,5 Rural 82,2 67,4 94,4 27,0 88,1 80,6 93,2 12,5 85,9 80,4 91,0 10,5 Infrastructure Treatment Control 85,3 77,4 92,2 14,8 17,2 18,0 16,6 -1,5 16,1 15,7 16,6 0,9 Treatment 79,2 65,3 92,2 26,8 82,8 82,0 83,4 1,5 83,9 84,3 83,4 -0,9 Total 80,2 67,2 92,2 25,0 100,0 100,0 100,0 0,0 100,0 100,0 100,0 0,0 Urban 50,8 51,2 50,0 -1,2 10,3 18,6 5,3 -13,3 14,1 19,6 9,0 -10,5 Rural 72,9 54,4 88,1 33,7 89,7 81,4 94,7 13,3 85,9 80,4 91,0 10,5 Poverty line = 24164.9(iii) Infrastructure Treatment Control 74,9 67,6 81,3 13,7 17,3 19,7 15,9 -3,8 16,1 15,7 16,6 0,9 Treatment 68,8 51,2 85,3 34,1 82,7 80,3 84,1 3,8 83,9 84,3 83,4 -0,9 Total 69,8 53,8 84,7 30,9 100,0 100,0 100,0 0,0 100,0 100,0 100,0 0,0 Notes: (i) Changes shown between years 2004 and 1998. (ii) Total Poverty Line. (iii) Food Poverty Line. The nominal 2004 p.a.e. figure is deflated into a real 1998 p.a.e. measure by dividing the nominal 2004 values by a CPI equal to 335.51 and multiplying it with 100. Source: Authors' calculations using Adept version 4.1. 423 Table A.4a: Poverty Decomposition Table A.4b: Poverty Decomposition Absolute Percentage change change Absolute Percentage change change Poverty line = 33562.4 Poverty line = 33562.4 Change in poverty (P0) 43,10 100,00 Change in poverty (P0) 24,95 100,00 Total Intra-sectoral effect 43,10 100,00 Total Intra-sectoral effect 24,95 100,00 Population-shift effect 0,16 0,38 Population-shift effect 0,11 0,43 Interaction effect -0,16 -0,38 Interaction effect -0,11 -0,43 Control 4,34 10,08 Control 2,31 9,27 Treatment 38,76 89,92 Treatment 22,64 90,73 Change in poverty (P0) 47,12 100,00 Change in poverty (P0) 30,87 100,00 Total Intra-sectoral effect 47,15 100,08 Total Intra-sectoral effect 30,90 100,12 Intra-sectoral effects: Intra-sectoral effects: Poverty line = 24164.9 Poverty line = 24164.9 Population-shift effect 0,19 0,40 Population-shift effect 0,15 0,48 Interaction effect -0,22 -0,47 Interaction effect -0,18 -0,59 Control 4,12 8,74 Control 2,15 6,96 Treatment 43,04 91,34 Treatment 28,76 93,16 Intra-sectoral effects: Intra-sectoral effects: Notes: The nominal 2004 p.a.e. figure is deflated into a real Notes: The nominal 2004 p.a.e. figure is deflated into a real 1998 p.a.e. measure by dividing the nominal 2004 values by 1998 p.a.e. measure by dividing the nominal 2004 values by based on the foodbasket equal to 216.3 and multiplying it with 100. based on the CPI equal to 335.51 and multiplying it with 100. Source: Authors' calculations using Adept version 4.1. Source: Authors' calculations using Adept version 4.1. 424 Table A.5a: Poverty by Household Head's Age in Zambia‟s Eastern Province Poverty Headcount Rate Distribution of the Poor Distribution of Population Pooled 2004 1998 change Pooled 2004 1998 change Pooled 2004 1998 change Poverty line = 33562.4 15-19 79,5 86,3 69,2 -17,1 0,2 0,4 0,1 -0,3 0,2 0,2 0,1 20-24 66,5 38,7 89,9 51,2 4,1 3,3 4,5 1,2 4,4 4,2 4,6 0,4 25-29 66,6 43,5 89,1 45,6 11,0 10,7 11,2 0,5 11,8 12,1 11,6 -0,5 30-34 67,8 52,0 88,5 36,5 14,1 18,5 11,9 -6,6 14,9 17,5 12,4 -5,1 35-39 73,0 53,5 90,9 37,4 15,0 15,9 14,6 -1,3 14,7 14,6 14,8 0,2 40-44 67,5 44,7 90,8 46,1 10,6 10,8 10,6 -0,2 11,3 11,8 10,7 -1,1 45-49 76,5 43,8 97,6 53,7 14,1 9,6 16,3 6,7 13,1 10,7 15,4 4,7 50-54 65,7 33,0 90,3 57,3 7,6 5,0 8,9 4,0 8,3 7,4 9,1 1,7 55-59 70,6 53,6 92,2 38,6 6,9 8,9 6,0 -2,9 7,0 8,1 6,0 -2,2 60-64 80,6 57,2 93,4 36,2 6,5 5,0 7,3 2,4 5,8 4,3 7,2 3,0 65+ 81,4 64,8 98,9 34,1 9,8 12,1 8,6 -3,4 8,6 9,1 8,1 -1,1 Total 71,4 49,1 92,2 43,1 100,0 0,0 100,0 15-19 67,3 66,0 69,2 3,2 0,2 0,4 0,1 -0,3 0,2 0,2 0,1 20-24 56,8 33,7 76,3 42,5 4,0 3,7 4,1 0,4 4,4 4,2 4,6 0,4 25-29 57,7 36,5 78,2 41,7 11,0 11,7 10,7 -1,0 11,8 12,1 11,6 -0,5 30-34 54,3 38,3 75,1 36,8 13,0 17,9 11,0 -6,9 14,9 17,5 12,4 -5,1 35-39 62,8 36,9 86,5 49,6 14,9 14,4 15,1 0,8 14,7 14,6 14,8 0,2 40-44 60,1 33,7 87,0 53,4 10,9 10,6 11,0 0,4 11,3 11,8 10,7 -1,1 45-49 69,5 34,2 92,3 58,1 14,7 9,7 16,8 7,0 13,1 10,7 15,4 4,7 50-54 58,4 28,2 81,1 52,9 7,8 5,5 8,7 3,2 8,3 7,4 9,1 1,7 55-59 60,7 40,2 86,8 46,7 6,9 8,7 6,1 -2,6 7,0 8,1 6,0 -2,2 60-64 73,5 52,4 85,1 32,8 6,9 6,0 7,3 1,3 5,8 4,3 7,2 3,0 65+ 69,9 46,6 94,4 47,8 9,7 11,4 9,0 -2,4 8,6 9,1 8,1 -1,1 Total 62,0 37,5 84,7 47,1 100,0 0,0 100,0 100,0 100,0 -0,1 100,0 100,0 0,0 Poverty line = 24164.9 100,0 100,0 -0,1 100,0 100,0 0,0 Notes: (i) Percentage Changes shown between years 2004 and 1998. (ii) The nominal 2004 p.a.e. figure is deflated into a real 1998 p.a.e. measure by dividing the nominal 2004 values by a CPI based on the foodbasket equal to 216.3 and multiplying it with 100. Source: Authors' calculations using Adept version 4.1. Table A.5b: Poverty by Household Head's Age in Zambia‟s Eastern Province Poverty Headcount Rate Pooled 2004 1998 Distribution of the Poor change Pooled 2004 1998 change Distribution of Population Pooled 2004 1998 change Poverty line = 33562.4 15-19 79,5 86,3 69,2 -17,1 0,2 0,3 0,1 -0,2 0,2 0,2 0,1 20-24 72,1 50,9 89,9 39,0 3,9 3,2 4,5 1,3 4,4 4,2 4,6 -0,1 0,4 25-29 79,5 69,7 89,1 19,4 11,7 12,5 11,2 -1,3 11,8 12,1 11,6 -0,5 30-34 75,9 66,3 88,5 22,3 14,1 17,2 11,9 -5,3 14,9 17,5 12,4 -5,1 35-39 80,5 69,1 90,9 21,7 14,8 15,0 14,6 -0,4 14,7 14,6 14,8 0,2 40-44 78,6 66,6 90,8 24,2 11,0 11,7 10,6 -1,1 11,3 11,8 10,7 -1,1 45-49 83,4 61,5 97,6 36,0 13,7 9,8 16,3 6,5 13,1 10,7 15,4 4,7 50-54 77,3 60,1 90,3 30,1 8,0 6,6 8,9 2,3 8,3 7,4 9,1 1,7 55-59 81,5 73,1 92,2 19,1 7,1 8,9 6,0 -2,9 7,0 8,1 6,0 -2,2 60-64 84,3 67,7 93,4 25,8 6,1 4,3 7,3 3,0 5,8 4,3 7,2 3,0 65+ 87,9 77,5 98,9 21,3 9,4 10,5 8,6 -1,9 8,6 9,1 8,1 -1,1 Total 80,2 67,2 92,2 25,0 100,0 0,0 100,0 15-19 79,5 86,3 69,2 -17,1 0,2 0,3 0,1 -0,2 0,2 0,2 0,1 20-24 60,5 41,9 76,3 34,4 3,8 3,2 4,1 0,9 4,4 4,2 4,6 0,4 25-29 67,7 56,9 78,2 21,4 11,5 12,8 10,7 -2,1 11,8 12,1 11,6 -0,5 30-34 63,5 54,7 75,1 20,4 13,5 17,8 11,0 -6,8 14,9 17,5 12,4 -5,1 35-39 72,9 58,0 86,5 28,6 15,4 15,8 15,1 -0,6 14,7 14,6 14,8 0,2 40-44 66,8 47,0 87,0 40,0 10,8 10,3 11,0 0,7 11,3 11,8 10,7 -1,1 45-49 76,1 51,0 92,3 41,3 14,3 10,1 16,8 6,6 13,1 10,7 15,4 4,7 50-54 61,2 34,8 81,1 46,3 7,3 4,8 8,7 3,9 8,3 7,4 9,1 1,7 55-59 70,1 56,9 86,8 30,0 7,0 8,6 6,1 -2,5 7,0 8,1 6,0 -2,2 100,0 100,0 100,0 100,0 0,0 Poverty line = 24164.9 -0,1 60-64 75,7 58,5 85,1 26,6 6,3 4,6 7,3 2,6 5,8 4,3 7,2 3,0 65+ 81,0 68,3 94,4 26,1 10,0 11,6 9,0 -2,6 8,6 9,1 8,1 -1,1 Total 69,8 53,8 84,7 30,9 100,0 100,0 100,0 0,0 100,0 100,0 100,0 Notes: (i) Percentage Changes shown between years 2004 and 1998. (ii) The nominal 2004 p.a.e. figure is deflated into a real 1998 p.a.e. measure by dividing the nominal 2004 values by a CPI equal to 335.51 and multiplying it with 100. Source: Authors' calculations using Adept version 4.1. 425 0,0 Table A.6a: Poverty by Household Head's Status of Employment of Eastern Province Poverty Headcount Rate Pooled 2004 1998 Distribution of the Poor change Distribution of Population Pooled 2004 1998 change Pooled 2004 1998 change Poverty line = 33562.4 Employment status SEδF EεθδηYED………………….……01 58,7 49,1 90,7 41,6 55,5 87,1 33,4 -53,7 64,5 86,8 34,6 -52,2 ωEζTRAδ GηVT EεθδηYEE………... 0β 51,3 47,6 75,4 27,8 3,1 6,0 1,0 -5,0 4,1 6,2 1,3 -4,9 δηωAδ GηVT EεθδηYEE……...…..…0γ 31,8 28,3 48,1 19,8 0,2 0,4 0,1 -0,3 0,5 0,6 0,2 -0,5 θARASTATAδ EεθδηYEE……….....04 78,1 78,5 0,0 -78,5 0,6 1,5 0,0 -1,5 0,6 1,0 0,0 -1,0 θRIVATE SEωTηR EεθδηYEE…...…0η 52,5 41,5 86,9 45,4 2,2 3,3 1,5 -1,7 2,9 3,8 1,6 -2,2 ζGη EεθδηYEE…………………...….0θ 37,2 36,8 100,0 63,2 0,2 0,4 0,0 -0,4 0,3 0,5 0,0 -0,5 Int.ηrg. / Embassy Employee…….…...07 83,7 EεθδηYER/θARTζER…………….....08 96,5 0,0 96,5 96,5 37,5 0,0 63,7 63,7 26,5 0,0 62,1 62,0 HηUSEHηδD EεθδηYEE…………....09 74,7 67,3 100,0 32,7 0,3 0,5 0,1 -0,3 0,3 0,3 0,1 -0,2 UζθAID FAεIδY WηRKER………...10 71,6 71,6 0,2 0,4 0,2 0,3 83,7 0,1 0,1 0,0 0,1 θIEωE WηRKER……….…………….…11 35,4 35,4 0,1 0,2 0,2 0,3 ηTHER SθEωIFY)…………………..…..1β Total 100,0 100,0 0,1 0,2 0,0 0,1 71,4 49,1 92,2 43,1 100,0 SEδF EεθδηYED………………….……01 48,4 37,8 84,1 46,4 52,8 87,8 ωEζTRAδ GηVT EεθδηYEE………... 0β 40,7 36,3 69,8 33,5 2,8 6,0 δηωAδ GηVT EεθδηYEE……...…..…0γ 20,6 14,7 48,1 33,4 0,2 θARASTATAδ EεθδηYEE……….....04 67,2 67,6 0,0 -67,6 θRIVATE SEωTηR EεθδηYEE…...…0η 32,2 26,1 51,4 ζGη EεθδηYEE…………………...….0θ 37,2 36,8 100,0 Int.ηrg. / Embassy Employee…….…...07 83,7 EεθδηYER/θARTζER…………….....08 92,0 0,0 92,0 92,0 41,3 0,0 64,7 64,7 26,5 0,0 62,1 62,0 HηUSEHηδD EεθδηYEE…………....09 42,3 25,3 100,0 74,7 0,2 0,2 0,2 -0,1 0,3 0,3 0,1 -0,2 UζθAID FAεIδY WηRKER………...10 54,9 54,9 0,1 0,4 0,2 0,3 θIEωE WηRKER……….…………….…11 33,1 33,1 0,1 0,3 0,2 0,3 ηTHER SθEωIFY)…………………..…..1β 6,8 6,8 0,0 0,0 0,0 62,0 37,5 100,0 100,0 0,0 100,0 100,0 100,0 0,0 33,0 -54,8 64,5 86,8 34,6 -52,2 1,0 -5,0 4,1 6,2 1,3 -4,9 0,3 0,1 -0,2 0,5 0,6 0,2 -0,5 0,6 1,7 0,0 -1,7 0,6 1,0 0,0 -1,0 25,3 1,6 2,7 1,0 -1,7 2,9 3,8 1,6 -2,2 63,2 0,2 0,5 0,0 -0,5 0,3 0,5 0,0 -0,5 Poverty line = 24164.9 Employment status Total 83,7 0,1 84,7 47,1 100,0 0,1 0,0 100,0 100,0 0,0 0,1 0,1 100,0 100,0 100,0 0,0 Notes: (i) Percentage Changes shown between years 2004 and 1998. (ii) The nominal 2004 p.a.e. figure is deflated into a real 1998 p.a.e. measure by dividing the nominal 2004 values by a CPI based on the foodbasket equal to 216.3 and multiplying it with 100. Source: Authors' calculations using Adept version 4.1. Table A.6b: Poverty by Household Head's Status of Employment of Eastern Province Poverty Headcount Rate Pooled 2004 1998 change Distribution of the Poor Pooled 2004 1998 change Distribution of Population Pooled 2004 1998 change Poverty line = 33562.4 Employment status SEδF EεθδηYED………………….……01 73,2 68,0 90,7 22,6 60,1 87,9 33,4 -54,5 64,5 86,8 34,6 ωEζTRAδ GηVT EεθδηYEE………... 0β 65,9 64,5 75,4 10,9 3,4 6,0 1,0 -4,9 4,1 6,2 1,3 -4,9 δηωAδ GηVT EεθδηYEE……...…..…0γ 42,2 41,0 48,1 7,1 0,2 0,4 0,1 -0,3 0,5 0,6 0,2 -0,5 θARASTATAδ EεθδηYEE……….....04 90,8 91,3 0,0 -91,3 0,6 1,3 0,0 -1,3 0,6 1,0 0,0 -1,0 θRIVATE SEωTηR EεθδηYEE…...…0η 62,2 54,4 86,9 32,5 2,3 3,1 1,5 -1,6 2,9 3,8 1,6 -2,2 ζGη EεθδηYEE…………………...….0θ 38,1 37,8 100,0 62,2 0,2 0,3 0,0 -0,3 0,3 0,5 0,0 -0,5 Int.ηrg. / Embassy Employee…….…...07 83,7 EεθδηYER/θARTζER…………….....08 96,5 17,1 96,5 79,4 32,5 0,0 63,7 63,7 26,5 0,0 62,1 62,0 HηUSEHηδD EεθδηYEE…………....09 77,8 71,2 100,0 28,8 0,3 0,4 0,1 -0,2 0,3 0,3 0,1 -0,2 UζθAID FAεIδY WηRKER………...10 80,8 80,8 0,2 0,3 0,2 0,3 0,3 83,7 0,0 0,1 0,0 -52,2 0,1 θIEωE WηRKER……….…………….…11 43,4 43,4 0,1 0,2 0,2 ηTHER SθEωIFY)…………………..…..1β Total 100,0 100,0 0,1 0,1 0,0 80,2 67,2 92,2 25,0 100,0 100,0 100,0 SEδF EεθδηYED………………….……01 61,1 54,2 84,1 29,9 57,5 87,6 ωEζTRAδ GηVT EεθδηYEE………... 0β 53,4 50,9 69,8 18,9 3,2 5,9 δηωAδ GηVT EεθδηYEE……...…..…0γ 31,8 28,3 48,1 19,8 0,2 θARASTATAδ EεθδηYEE……….....04 79,0 79,4 0,0 -79,4 0,6 θRIVATE SEωTηR EεθδηYEE…...…0η 45,6 43,8 51,4 7,6 1,9 3,1 1,0 -2,2 2,9 3,8 1,6 -2,2 ζGη EεθδηYEE…………………...….0θ 37,2 36,8 100,0 63,2 0,2 0,4 0,0 -0,4 0,3 0,5 0,0 -0,5 Int.ηrg. / Embassy Employee…….…...07 83,7 EεθδηYER/θARTζER…………….....08 92,0 0,0 92,0 92,0 35,6 0,0 64,7 64,7 26,5 0,0 62,1 62,0 HηUSEHηδD EεθδηYEE…………....09 77,8 71,2 100,0 28,8 0,3 0,5 0,2 -0,3 0,3 0,3 0,1 -0,2 UζθAID FAεIδY WηRKER………...10 80,8 80,8 0,2 0,4 0,2 0,3 0,3 0,1 0,0 100,0 100,0 100,0 0,0 33,0 -54,6 64,5 86,8 34,6 -52,2 1,0 -4,9 4,1 6,2 1,3 -4,9 0,3 0,1 -0,2 0,5 0,6 0,2 -0,5 1,4 0,0 -1,4 0,6 1,0 0,0 -1,0 Poverty line = 24164.9 Employment status 83,7 0,1 0,1 0,0 θIEωE WηRKER……….…………….…11 40,8 40,8 0,1 0,2 0,2 ηTHER SθEωIFY)…………………..…..1β 100,0 100,0 0,1 0,1 0,0 Total 69,8 53,8 84,7 30,9 100,0 100,0 100,0 0,0 100,0 0,1 0,1 100,0 100,0 Notes: (i) Percentage Changes shown between years 2004 and 1998. (ii) The nominal 2004 p.a.e. figure is deflated into a real 1998 p.a.e. measure by dividing the nominal 2004 values by a CPI equal to 335.51 and multiplying it with 100. Source: Authors' calculations using Adept version 4.1. 426 0,0 Table A.7a: Poverty by Household Head's Education Level in Eastern Province Poverty Headcount Rate Distribution of the Poor Distribution of Population Pooled 2004 1998 change Pooled 2004 1998 change Pooled 2004 1998 change Poverty line = 33562.4 Highest Grade 0 100,0 1 82,3 69,6 100,0 97,8 28,2 4,0 0,0 3,8 4,3 0,0 0,5 3,1 2,6 4,0 2 65,6 52,7 88,8 36,2 7,3 7,6 7,0 -0,6 7,0 6,9 7,3 0,3 3 69,0 53,3 90,6 37,2 5,9 5,4 6,5 1,1 5,4 4,8 6,6 1,8 4 70,7 45,9 99,3 53,3 14,3 10,1 18,6 8,5 12,8 10,5 17,2 6,8 5 68,1 42,9 96,5 53,6 10,2 6,9 13,4 6,5 9,4 7,6 12,8 5,1 6 68,5 54,4 95,4 41,0 13,0 13,7 12,4 -1,2 12,0 12,0 12,0 -0,0 7 64,4 46,2 95,9 49,7 21,7 19,9 23,5 3,6 21,2 20,6 22,5 1,9 8 51,1 41,9 86,2 44,3 3,5 4,6 2,4 -2,2 4,3 5,2 2,6 -2,6 9 57,5 51,5 76,1 24,6 9,3 12,6 6,0 -6,6 10,1 11,7 7,2 -4,5 10 44,0 29,8 91,8 62,0 2,9 3,1 2,8 -0,3 4,2 5,0 2,8 -2,2 11 56,4 50,6 92,0 41,4 0,6 1,0 0,3 -0,7 0,7 1,0 0,3 -0,7 12 49,8 50,3 48,0 -2,3 4,3 7,0 1,6 -5,4 5,5 6,7 3,1 -3,6 13 73,5 68,1 77,6 9,5 1,1 0,9 1,4 0,4 1,0 0,6 1,6 1,0 14 35,9 36,1 0,0 -36,1 1,3 2,6 0,0 -2,6 2,3 3,5 0,0 -3,5 15 39,9 39,9 0,3 0,7 0,5 0,8 16 19,8 19,8 17 21,0 21,4 0,0 -21,4 0,0 -0,2 Total 71,4 49,1 92,2 43,1 0,1 0,2 0,0 0,1 0,0 100,0 100,0 100,0 0,0 -0,1 0,0 0,0 0,3 0,4 0,1 0,2 100,0 100,0 100,0 1,5 0,0 Poverty line = 24164.9 Highest Grade 0 100,0 1 74,6 55,9 100,0 97,3 41,4 4,4 0,0 4,0 4,7 0,0 0,8 3,1 2,6 4,0 1,5 2 60,3 46,9 84,7 37,8 8,1 8,9 7,4 -1,5 7,0 6,9 7,3 0,3 3 62,6 45,5 86,1 40,6 6,5 6,0 6,8 0,8 5,4 4,8 6,6 1,8 4 53,6 26,9 84,3 57,4 13,1 7,7 17,5 9,8 12,8 10,5 17,2 6,8 5 56,9 26,1 91,5 65,5 10,2 5,5 14,1 8,6 9,4 7,6 12,8 5,1 6 58,0 43,1 86,1 42,9 13,2 14,2 12,4 -1,8 12,0 12,0 12,0 -0,0 7 55,8 38,8 85,1 46,3 22,6 21,9 23,1 1,2 21,2 20,6 22,5 1,9 8 40,7 31,3 76,9 45,6 3,3 4,5 2,4 -2,1 4,3 5,2 2,6 -2,6 9 45,4 37,3 70,0 32,7 8,8 12,0 6,1 -5,9 10,1 11,7 7,2 -4,5 10 31,0 17,5 76,4 58,9 2,5 2,4 2,6 0,2 4,2 5,0 2,8 -2,2 11 56,4 50,6 92,0 41,4 0,8 1,3 0,3 -1,0 0,7 1,0 0,3 -0,7 12 37,2 38,3 32,8 -5,5 3,9 7,0 1,2 -5,8 5,5 6,7 3,1 -3,6 13 66,5 68,1 65,2 -2,9 1,2 1,2 1,3 0,0 1,0 0,6 1,6 1,0 14 29,4 29,6 0,0 -29,6 1,3 2,8 0,0 -2,8 2,3 3,5 0,0 -3,5 15 14,3 14,3 0,1 0,3 0,5 0,8 16 19,8 19,8 17 0,0 0,0 0,0 0,0 0,0 -0,2 Total 62,0 37,5 84,7 47,1 0,1 0,2 0,0 0,0 0,0 100,0 100,0 100,0 0,0 0,0 0,0 0,0 0,3 0,4 0,1 0,2 100,0 100,0 100,0 0,0 Notes: (i) Percentage Changes shown between years 2004 and 1998. (ii) The nominal 2004 p.a.e. figure is deflated into a real 1998 p.a.e. measure by dividing the nominal 2004 values by a CPI based on the foodbasket equal to 216.3 and multiplying it with 100. Source: Authors' calculations using Adept version 4.1. 427 Table A.7b: Poverty by Household Head's Education Level in Eastern Province Poverty Headcount Rate Pooled 2004 1998 Distribution of the Poor change Pooled 2004 1998 change Distribution of Population Pooled 2004 1998 change Poverty line = 33562.4 Highest Grade 0 100,0 100,0 0,0 0,0 1 87,2 78,5 97,8 19,3 3,5 3,0 4,3 1,3 3,1 2,6 4,0 1,5 2 72,9 64,1 88,8 24,7 6,8 6,6 7,0 0,4 7,0 6,9 7,3 0,3 3 82,0 75,8 90,6 14,8 5,9 5,4 6,5 1,1 5,4 4,8 6,6 1,8 4 82,4 67,8 99,3 31,4 13,9 10,5 18,6 8,0 12,8 10,5 17,2 6,8 5 79,4 64,4 96,5 32,0 9,8 7,3 13,4 6,1 9,4 7,6 12,8 5,1 6 76,9 67,1 95,4 28,2 12,1 11,9 12,4 0,5 12,0 12,0 12,0 -0,0 7 76,1 64,6 95,9 31,2 21,3 19,7 23,5 3,7 21,3 20,6 22,5 1,9 8 69,5 65,2 86,2 21,1 3,9 5,0 2,4 -2,6 4,3 5,2 2,6 -2,6 9 70,5 68,7 76,1 7,4 9,4 11,9 6,0 -5,9 10,2 11,7 7,2 -4,5 10 79,6 75,9 91,8 15,9 4,4 5,6 2,8 -2,8 4,2 5,0 2,8 -2,2 11 90,7 90,5 92,0 1,6 0,9 1,3 0,3 -1,0 0,7 1,0 0,3 -0,7 12 65,7 70,0 48,0 -22,0 4,7 7,0 1,6 -5,3 5,5 6,7 3,1 -3,6 13 73,5 68,1 77,6 9,5 0,9 0,7 1,4 0,7 1,0 0,6 1,6 1,0 14 60,2 60,5 0,0 -60,5 1,8 3,1 0,0 -3,1 2,3 3,5 0,0 -3,5 15 46,2 46,2 0,3 0,5 0,5 0,8 16 53,3 53,3 0,2 0,3 0,3 0,4 17 45,7 46,5 0,0 -46,5 0,1 0,1 0,1 0,2 0,0 -0,2 Total 80,2 67,2 92,2 25,0 0,0 100,0 100,0 100,0 0,0 -0,1 0,0 0,0 100,0 100,0 100,0 0,0 Poverty line = 24164.9 Highest Grade 0 100,0 1 82,1 69,6 100,0 97,3 27,7 4,0 0,0 3,4 4,7 0,0 1,3 3,1 2,6 4,0 2 70,0 61,9 84,7 22,8 7,8 8,1 7,4 -0,7 7,0 6,9 7,3 0,3 3 70,4 59,0 86,1 27,1 6,0 5,4 6,8 1,5 5,4 4,8 6,6 1,8 4 67,0 52,0 84,3 32,3 13,5 10,3 17,5 7,2 12,8 10,5 17,2 6,8 5 67,7 46,8 91,5 44,7 10,1 6,7 14,1 7,3 9,4 7,6 12,8 5,1 6 65,5 54,7 86,1 31,3 12,4 12,4 12,4 0,0 12,0 12,0 12,0 -0,0 7 64,1 52,0 85,1 33,1 21,5 20,2 23,1 2,9 21,3 20,6 22,5 1,9 8 56,9 51,7 76,9 25,2 3,8 5,1 2,4 -2,7 4,3 5,2 2,6 -2,6 9 59,6 56,2 70,0 13,9 9,5 12,4 6,1 -6,3 10,2 11,7 7,2 -4,5 10 41,9 31,6 76,4 44,8 2,8 3,0 2,6 -0,4 4,2 5,0 2,8 -2,2 11 71,9 68,6 92,0 23,4 0,8 1,2 0,3 -0,9 0,7 1,0 0,3 -0,7 12 52,6 57,4 32,8 -24,6 4,5 7,3 1,2 -6,0 5,5 6,7 3,1 -3,6 13 66,5 68,1 65,2 -2,9 1,0 0,8 1,3 0,4 1,0 0,6 1,6 1,0 14 45,0 45,2 0,0 -45,2 1,6 3,0 0,0 -3,0 2,3 3,5 0,0 -3,5 15 39,9 39,9 0,3 0,6 0,5 0,8 16 19,8 19,8 17 21,0 21,4 0,0 -21,4 0,0 -0,2 Total 69,8 53,8 84,7 30,9 0,1 0,2 0,0 0,1 0,0 100,0 100,0 100,0 0,0 -0,1 0,0 0,0 0,3 0,4 0,1 0,2 100,0 100,0 100,0 Notes: (i) Percentage Changes shown between years 2004 and 1998. (ii) The nominal 2004 p.a.e. figure is deflated into a real 1998 p.a.e. measure by dividing the nominal 2004 values by a CPI equal to 335.51 and multiplying it with 100. Source: Authors' calculations using Adept version 4.1. 428 1,5 0,0 Table A.8a: Poverty by Household Head's Gender Poverty Headcount Rate Distribution of the Poor Distribution of Population Pooled 2004 1998 change Pooled 2004 1998 change Pooled 2004 1998 change Poverty line = 33562.4 1 = Male and 2 = Female Male 70,2 48,6 91,2 42,7 78,8 81,0 77,7 -3,3 80,1 81,9 78,5 Female 76,2 51,4 95,6 44,2 21,2 19,0 22,3 3,3 19,9 18,1 21,5 -3,3 3,3 Total 71,4 49,1 92,2 43,1 100,0 100,0 100,0 0,0 100,0 100,0 100,0 0,0 Male 60,0 36,7 82,5 45,8 77,5 80,0 76,6 -3,5 80,1 81,9 78,5 -3,3 Female 70,0 41,4 92,4 51,1 22,5 20,0 23,4 3,5 19,9 18,1 21,5 3,3 Total 62,0 37,5 84,7 47,1 100,0 100,0 100,0 0,0 100,0 100,0 100,0 0,0 Poverty line = 24164.9 1 = Male and 2 = Female Notes: (i) Percentage Changes shown between years 2004 and 1998. (ii) The nominal 2004 p.a.e. figure is deflated into a real 1998 p.a.e. measure by dividing the nominal 2004 values by a CPI based on the foodbasket equal to 216.3 and multiplying it with 100. Source: Authors' calculations using Adept version 4.1. Table A.8b: Poverty by Household Head's Gender Poverty Headcount Rate Pooled 2004 1998 Distribution of the Poor change Pooled 2004 1998 change Distribution of Population Pooled 2004 1998 change Poverty line = 33562.4 1 = Male and 2 = Female Male 79,5 67,4 91,2 23,8 79,5 82,1 77,7 -4,4 80,1 81,9 78,5 Female 82,8 66,4 95,6 29,2 20,5 17,9 22,3 4,4 19,9 18,1 21,5 -3,3 3,3 Total 80,2 67,2 92,2 25,0 100,0 100,0 100,0 0,0 100,0 100,0 100,0 0,0 Male 68,1 53,1 82,5 29,4 78,1 80,8 76,6 -4,3 80,1 81,9 78,5 -3,3 Female 76,8 56,8 92,4 35,6 21,9 19,2 23,4 4,3 19,9 18,1 21,5 3,3 Total 69,8 53,8 84,7 30,9 100,0 100,0 100,0 0,0 100,0 100,0 100,0 0,0 Poverty line = 24164.9 1 = Male and 2 = Female Notes: (i) Percentage Changes shown between years 2004 and 1998. (ii) The nominal 2004 p.a.e. figure is deflated into a real 1998 p.a.e. measure by dividing the nominal 2004 values by a CPI equal to 335.51 and multiplying it with 100. Source: Authors' calculations using Adept version 4.1. Table A9: Poverty by Rural and Urban Strata Poverty Headcount Rate Pooled 2004 1998 Distribution of the Poor change Pooled 2004 1998 change Distribution of Population Pooled 2004 1998 change Poverty line = 33562.4 1 Rural small Scale 82,5 67,1 95,1 28,1 81,6 74,2 86,6 12,4 79,3 74,4 83,9 2 Rural meduim Scale 82,6 72,6 91,9 19,3 4,4 4,6 4,2 -0,3 4,2 4,2 4,2 9,6 0,0 3 Rural large scale 83,7 96,3 59,3 -37,1 0,1 0,2 0,0 -0,2 0,1 0,2 0,1 -0,1 4 Rural non-agric 74,1 68,9 77,1 8,2 2,1 1,7 2,3 0,5 2,2 1,7 2,7 1,0 5 Urban Low Cost 70,3 65,7 75,8 10,1 5,9 7,4 4,9 -2,5 6,7 7,6 5,9 -1,7 6 Urban medium cost 65,6 66,7 60,7 -6,0 5,6 11,4 1,7 -9,8 6,9 11,5 2,6 -9,0 7 Urban high cost 53,0 66,9 43,4 -23,5 0,3 0,4 0,2 -0,1 0,5 0,4 0,5 0,1 Total 80,2 67,2 92,2 25,0 100,0 100,0 100,0 0,0 100,0 100,0 100,0 0,0 1 Rural small Scale 73,7 53,9 89,9 36,0 83,7 74,6 89,1 14,6 79,3 74,4 83,9 9,6 2 Rural meduim Scale 63,8 61,5 65,9 4,4 3,9 4,9 3,3 -1,5 4,2 4,2 4,2 0,0 3 Rural large scale 31,2 16,6 59,3 42,6 0,1 0,0 0,1 0,0 0,1 0,2 0,1 -0,1 Poverty line = 24164.9 4 Rural non-agric 64,6 58,6 68,0 9,4 2,1 1,9 2,2 0,3 2,2 1,7 2,7 1,0 5 Urban Low Cost 56,5 56,7 56,2 -0,4 5,5 8,0 3,9 -4,1 6,7 7,6 5,9 -1,7 6 Urban medium cost 47,2 49,0 39,5 -9,5 4,6 10,5 1,2 -9,3 6,9 11,5 2,6 -9,0 7 Urban high cost 27,5 21,9 31,4 9,5 0,2 0,2 0,2 0,0 0,5 0,4 0,5 0,1 Total 69,8 53,8 84,7 30,9 100,0 0,0 100,0 100,0 100,0 100,0 100,0 0,0 Notes: (i) Percentage Changes shown between years 2004 and 1998. (ii) The nominal 2004 p.a.e. figure is deflated into a real 1998 p.a.e. measure by dividing the nominal 2004 values by a CPI equal to 335.51 and multiplying it with 100. (iii) Since there was only 4 fishfarms in 2004 and none in 1998, they have been filtered away, and therefore not included in the calculations. (iv) In 1998 there was only 7 large-scale farm obs and in 2004 there was only 5 large-scale farm obs. Source: Authors' calculations using Adept version 4.1. 429 Table A.10a: Income-based poverty estimates Poverty Headcount Rate Distribution of the Poor Pooled 2004 1998 change Pooled 2004 1998 change Distribution of Population Pooled 2004 1998 change Poverty line = 33562.4 Urban 38,9 22,5 71,9 49,5 7,7 8,0 7,6 -0,4 14,1 19,6 9,0 -10,5 Rural 76,1 63,2 86,8 23,6 92,3 92,0 92,4 0,4 85,9 80,4 91,0 10,5 Infrastructure Treatment Control 62,5 37,8 84,2 46,4 14,2 10,7 16,3 5,6 16,1 15,7 16,6 0,9 Treatment 72,5 58,5 85,7 27,2 85,8 89,3 83,7 -5,6 83,9 84,3 83,4 -0,9 Total 70,9 55,2 85,4 30,2 100,0 100,0 100,0 0,0 100,0 100,0 100,0 0,0 Urban 30,8 15,1 62,2 47,1 6,8 6,1 7,2 1,1 14,1 19,6 9,0 -10,5 Rural 69,4 57,0 79,6 22,6 93,2 93,9 92,8 -1,1 85,9 80,4 91,0 10,5 Poverty line = 24164.9 Infrastructure Treatment Control 53,0 31,5 71,8 40,3 13,4 10,1 15,3 5,1 16,1 15,7 16,6 0,9 Treatment 66,1 52,0 79,2 27,2 86,6 89,9 84,7 -5,1 83,9 84,3 83,4 -0,9 Total 64,0 48,8 78,0 29,2 100,0 100,0 100,0 0,0 100,0 100,0 100,0 0,0 Notes: (i) Percentage Changes shown between years 2004 and 1998. (ii) The nominal 2004 total income figure is deflated into a real 1998 total income measure by dividing the nominal 2004 values by a CPI based on the foodbasket equal to 216.3 and multiplying it with 100. Source: Authors' calculations using Adept version 4.1. Table A.10b: Income-based poverty estimates Poverty Headcount Rate Pooled 2004 1998 Distribution of the Poor change Pooled 2004 1998 change Distribution of Population Pooled 2004 1998 change Poverty line = 33562.4 Urban 23,9 25,5 71,9 46,4 7,6 8,2 7,6 -0,6 14,1 19,6 9,0 -10,5 Rural 47,6 69,5 86,8 17,2 92,4 91,8 92,4 0,6 85,9 80,4 91,0 10,5 Control 44,8 43,5 84,2 40,7 16,3 11,2 16,3 5,1 16,1 15,7 16,6 0,9 Treatment 44,2 64,2 85,7 21,5 83,7 88,8 83,7 -5,1 83,9 84,3 83,4 -0,9 Total 44,3 60,9 85,4 24,5 100,0 100,0 100,0 0,0 100,0 100,0 100,0 0,0 Urban 20,7 23,6 62,2 38,6 7,2 8,2 7,2 -1,0 14,1 19,6 9,0 -10,5 Rural 43,7 64,6 79,6 14,9 92,8 91,8 92,8 1,0 85,9 80,4 91,0 10,5 Control 38,2 40,0 71,8 31,8 15,3 11,1 15,3 4,2 16,1 15,7 16,6 0,9 Treatment 40,9 59,7 79,2 19,6 84,7 88,9 84,7 -4,2 83,9 84,3 83,4 -0,9 Total 40,4 56,6 78,0 21,4 100,0 100,0 100,0 0,0 100,0 100,0 100,0 0,0 Infrastructure Treatment Poverty line = 24164.9 Infrastructure Treatment Notes: (i) Percentage Changes shown between years 2004 and 1998. (ii) The nominal 2004 total income figure is deflated into a real 1998 total income measure by dividing the nominal 2004 values by a CPI equal to 335.51 and multiplying it with 100. Source: Authors' calculations using Adept version 4.1. 430 Table A.11a: Decomposition of inequality by Treatment and Control regions Pooled 2004 1998 GE(0) GE(1) 87,8 76,9 141,7 54,4 52,2 81,9 69,4 65,8 152,4 75,8 74,0 155,2 59,7 62,0 113,8 62,1 55,7 95,2 Total GE(2) GE(0) GE(1) GE(2) GE(0) GE(1) GE(2) Infrastructure Treatment Control Treatment 89,6 76,8 138,4 52,6 50,2 77,1 70,8 67,9 165,6 Within-group inequality 87,3 76,5 141,3 53,7 51,5 81,2 69,4 65,8 152,4 Between-group inequality 0,5 0,5 0,4 0,8 0,7 0,6 0,0 0,0 0,0 Between as a share of maximum 0,5 0,6 0,3 1,4 1,3 0,8 0,1 0,1 0,0 Notes: The nominal 2004 p.a.e. figure is deflated into a real 1998 p.a.e. measure by dividing the nominal 2004 values by a CPI based on the foodbasket equal to 216.3 and multiplying it with 100. Source: Authors' calculations using Adept version 4.1. Table A.11b: Decomposition of inequality by Treatment and Control regions Pooled Total 2004 1998 GE(0) GE(1) GE(2) GE(0) GE(1) GE(2) GE(0) GE(1) GE(2) 75,3 68,1 123,3 54,4 52,2 81,9 69,4 65,8 152,4 Infrastructure Treatment Control 66,5 65,0 126,6 59,7 62,0 113,8 62,1 55,7 95,2 Treatment 76,6 68,2 121,7 52,6 50,2 77,1 70,8 67,9 165,6 Within-group inequality 75,0 67,8 123,0 53,7 51,5 81,2 69,4 65,8 152,4 Between-group inequality 0,3 0,3 0,3 0,8 0,7 0,6 0,0 0,0 0,0 Between as a share of maximum 0,5 0,5 0,3 1,4 1,3 0,8 0,1 0,1 0,0 Notes: The nominal 2004 p.a.e. figure is deflated into a real 1998 p.a.e. measure by dividing the nominal 2004 values by a CPI equal to 335.51 and multiplying it with 100. Source: Authors' calculations using Adept version 4.1. Table A.12a: Decomposition of inequality by urban and rural areas Pooled 2004 1998 GE(0) GE(1) GE(2) GE(0) GE(1) GE(2) GE(0) GE(1) GE(2) Total 87,8 76,9 141,7 54,4 52,2 81,9 69,4 65,8 152,4 Urban 53,6 53,0 85,1 56,2 52,0 75,4 37,0 40,6 88,1 Rural 91,1 80,8 155,2 53,9 52,2 83,5 64,7 61,3 132,2 Within-group inequality 85,8 74,7 139,2 54,3 52,1 81,8 62,2 56,4 139,4 Between-group inequality 2,0 2,2 2,5 0,1 0,1 0,1 7,3 9,4 13,1 Between as a share of maximum 2,2 2,9 1,8 0,2 0,2 0,1 10,5 14,3 8,6 Notes: The nominal 2004 p.a.e. figure is deflated into a real 1998 p.a.e. measure by dividing the nominal 2004 values by a CPI based on the foodbasket equal to 216.3 and multiplying it with 100. Source: Authors' calculations using Adept version 4.1. Table A.12b: Decomposition of inequality by urban and rural areas Pooled GE(0) GE(1) GE(2) 2004 1998 GE(0) GE(1) GE(2) GE(0) GE(1) GE(2) Total 75,3 68,1 123,3 54,4 52,2 81,9 69,4 65,8 Urban 50,1 49,0 79,5 56,2 52,0 75,4 37,0 40,6 88,1 Rural 77,1 70,6 131,9 53,9 52,2 83,5 64,7 61,3 132,2 Within-group inequality 73,2 65,8 120,7 54,3 52,1 81,8 62,2 56,4 139,4 Between-group inequality 2,1 2,3 2,6 0,1 0,1 0,1 7,3 9,4 13,1 Between as a share of maximum 2,7 3,4 2,1 0,2 0,2 0,1 10,5 14,3 8,6 Notes: The nominal 2004 p.a.e. figure is deflated into a real 1998 p.a.e. measure by dividing the nominal 2004 values by a CPI equal to 335.51 and multiplying it with 100. Source: Authors' calculations using Adept version 4.1. 431 152,4 Table A.13a: Inequality in per-capita expenditure distribution by urban and rural areas bottom Half of the Distribution p25/p10 p50/p25 Upper Half of the Distribution p75/p50 p90/p50 Interquarti le Range p75/p25 p90/p10 Gini Tails Total Pooled 2,69 2,70 2,44 5,26 6,61 38,30 63,40 2004 1,89 2,01 1,97 3,83 3,97 14,61 53,68 1998 2,47 2,06 2,27 4,32 4,69 22,07 58,04 Pooled 2,05 1,67 1,94 4,06 3,25 13,94 53,43 2004 1,87 1,75 1,87 4,53 3,28 14,85 53,84 1998 2,04 1,56 1,57 2,66 2,45 8,47 44,97 Pooled 2,73 2,53 2,63 5,62 6,65 38,86 64,63 2004 1,83 2,05 2,00 3,88 4,10 14,53 53,55 1998 2,25 2,15 2,14 4,19 4,61 20,22 56,30 Urban Rural Notes: The nominal 2004 p.a.e. figure is deflated into a real 1998 p.a.e. measure by dividing the nominal 2004 values by a CPI based on the foodbasket equal to 216.3 and multiplying it with 100. Source: Authors' calculations using Adept version 4.1. Table A.13b: Inequality in per-capita expenditure distribution by urban and rural areas bottom Half of the Distribution p25/p10 p50/p25 Upper Half of the Distribution p75/p50 p90/p50 Interquarti le Range p75/p25 Tails p90/p10 Gini Total Pooled 2,62 2,34 2,23 4,52 5,23 27,82 59,99 2004 1,89 2,01 1,97 3,83 3,97 14,61 53,68 1998 2,47 2,06 2,27 4,32 4,69 22,07 58,04 Urban Pooled 1,93 1,72 1,79 3,35 3,07 11,10 51,49 2004 1,87 1,75 1,87 4,53 3,28 14,85 53,84 1998 2,04 1,56 1,57 2,66 2,45 8,47 44,97 Rural Pooled 2,59 2,37 2,33 4,74 5,52 29,08 60,80 2004 1,83 2,05 2,00 3,88 4,10 14,53 53,55 1998 2,25 2,15 2,14 4,19 4,61 20,22 56,30 Notes: The nominal 2004 p.a.e. figure is deflated into a real 1998 p.a.e. measure by dividing the nominal 2004 values by a CPI equal to 335.51 and multiplying it with 100. Source: Authors' calculations using Adept version 4.1. 432 Table A.14a: Sensitivity of Headcount Poverty Rate with Respect to the Choice of Poverty Line Pooled 2004 1998 Poverty Change Poverty Change Poverty Change Headcount from actual Headcount from actual Headcount from actual Rate (P0) (% ) Rate (P0) (% ) Rate (P0) (% ) Poverty line = 33562.4 Actual 71,4 0,00 49,1 0,00 92,2 0,00 +5% 73,0 2,17 51,2 4,28 93,2 1,12 +10% 74,1 3,79 53,2 8,45 93,5 1,48 +20% 76,3 6,75 56,4 14,97 94,7 2,69 -5% 70,0 -1,95 47,3 -3,72 91,2 -1,07 -10% 68,4 -4,21 44,9 -8,55 90,3 -2,06 -20% 65,5 -8,37 41,0 -16,55 88,2 -4,33 Actual 62,0 0,00 37,5 0,00 84,7 0,00 +5% 63,6 2,57 38,8 3,50 86,5 2,19 +10% 64,9 4,78 40,7 8,38 87,4 3,30 +20% 67,5 8,93 43,4 15,72 89,9 6,14 -5% 59,8 -3,43 34,5 -8,16 83,4 -1,48 -10% 58,5 -5,59 32,3 -13,85 82,8 -2,20 -20% 55,7 -10,20 29,3 -22,00 80,1 -5,35 Poverty line = 24164.9 Notes: The nominal 2004 p.a.e. figure is deflated into a real 1998 p.a.e. measure by dividing the nominal 2004 values by a CPI based on the foodbasket equal to 216.3 and multiplying it with 100. Source: Authors' calculations using Adept version 4.1. Table A.14b: Sensitivity of Headcount Poverty Rate with Respect to the Choice of Poverty Line Pooled 2004 1998 Poverty Change Poverty Change Poverty Change Headcount from actual Headcount from actual Headcount from actual Rate (P0) (% ) Rate (P0) (% ) Rate (P0) (% ) Poverty line = 33562.4 Actual 80,2 0,00 67,2 0,00 92,2 0,00 +5% 81,2 1,33 68,3 1,63 93,2 1,12 +10% 81,9 2,15 69,3 3,12 93,5 1,48 +20% 83,8 4,51 72,1 7,20 94,7 2,69 -5% 78,6 -1,97 65,0 -3,28 91,2 -1,07 -10% 76,9 -4,06 62,5 -7,00 90,3 -2,06 -20% 73,6 -8,19 57,9 -13,90 88,2 -4,33 Poverty line = 24164.9 Actual 69,8 0,00 53,8 0,00 84,7 0,00 +5% 71,6 2,54 55,5 3,13 86,5 2,19 +10% 73,0 4,57 57,4 6,73 87,4 3,30 +20% 76,1 9,02 61,3 13,90 89,9 6,14 -5% 68,0 -2,50 51,5 -4,23 83,4 -1,48 -10% 66,7 -4,47 49,3 -8,32 82,8 -2,20 -20% 63,1 -9,64 44,7 -16,92 80,1 -5,35 Notes: The nominal 2004 p.a.e. figure is deflated into a real 1998 p.a.e. measure by dividing the nominal 2004 values by a CPI equal to 335.51 and multiplying it with 100. Source: Authors' calculations using Adept version 4.1. 433 Table A.15a: Elasticity of Poverty with Respect to the Consumption Poverty Headcount Rate Poverty Gap (P1) Squared Poverty Gap (P2) (P0) Pooled 2004 1998 change Pooled 2004 1998 change Pooled 2004 1998 change Poverty line = 33562.4 Urban -0,82 -0,68 -1,01 -0,32 -1,12 -1,09 -1,17 -0,08 -1,21 -1,17 -1,27 -0,10 Rural -0,34 -0,85 -0,12 0,72 -0,45 -0,94 -0,31 0,63 -0,54 -1,10 -0,42 0,68 Total -0,39 -0,82 -0,18 0,63 -0,50 -0,97 -0,35 0,62 -0,58 -1,11 -0,44 0,67 Urban -1,41 -1,25 -1,62 -0,37 -1,27 -1,25 -1,32 -0,07 -1,19 -1,13 -1,29 -0,16 Rural -0,42 -1,24 -0,13 1,11 -0,51 -1,09 -0,38 0,71 -0,62 -1,27 -0,51 0,76 Total -0,51 -1,24 -0,21 1,03 -0,55 -1,11 -0,41 0,71 -0,64 -1,25 -0,53 0,72 Poverty line = 24164.9 Notes: The nominal 2004 p.a.e. figure is deflated into a real 1998 p.a.e. measure by dividing the nominal 2004 values by a CPI based on the foodbasket equal to 216.3 and multiplying it with 100. Source: Authors' calculations using Adept version 4.1. Table A.15b: Elasticity of Poverty with Respect to the Consumption Poverty Headcount Rate Pooled 2004(P0) 1998 change Poverty Gap (P1) Pooled Squared Poverty Gap (P2) 2004 1998 change Pooled 2004 1998 change Poverty line = 33562.4 Urban -1,09 -1,13 -1,01 0,12 -0,97 -0,87 -1,17 -0,30 -1,07 -0,99 -1,27 Rural -0,28 -0,54 -0,12 0,41 -0,45 -0,78 -0,31 0,46 -0,55 -0,91 -0,42 -0,28 0,49 Total -0,37 -0,65 -0,18 0,47 -0,50 -0,79 -0,35 0,45 -0,58 -0,92 -0,44 0,48 -1,14 -0,90 -1,62 -0,72 -1,09 -1,00 -1,32 -0,31 -1,17 -1,13 -1,29 -0,16 Poverty line = 24164.9 Urban Rural -0,31 -0,66 -0,13 0,53 -0,52 -0,91 -0,38 0,53 -0,63 -1,05 -0,51 0,54 Total -0,39 -0,71 -0,21 0,50 -0,56 -0,93 -0,41 0,52 -0,66 -1,06 -0,53 0,53 Notes: The nominal 2004 p.a.e. figure is deflated into a real 1998 p.a.e. measure by dividing the nominal 2004 values by a CPI equal to 335.51 and multiplying it with 100. Source: Authors' calculations using Adept version 4.1. 434 Table A.16a: Elasticity of Poverty with Respect to the Inequality Poverty Headcount Rate (P0) Poverty Gap (P1) Squared Poverty Gap (P2) Pooled 2004 1998 change Pooled 2004 1998 change Pooled 2004 1998 change Urban 0,59 0,75 0,00 -0,75 2,33 2,89 1,08 -1,81 3,62 4,30 2,01 -2,29 Rural -0,06 0,71 -0,11 -0,82 0,92 2,36 0,11 -2,25 1,77 3,74 0,36 -3,38 Total 0,02 0,74 -0,11 -0,85 1,06 2,45 0,17 -2,28 1,99 3,85 0,49 -3,36 Poverty line = 33562.4 Poverty line = 24164.9 Urban 2,55 2,66 1,01 -1,65 3,70 4,34 2,05 -2,30 5,06 5,80 3,18 -2,62 Rural 0,24 1,54 -0,24 -1,79 1,48 3,75 0,24 -3,51 2,63 5,39 0,62 -4,77 Total 0,34 1,77 -0,22 -1,99 1,70 3,87 0,34 -3,52 2,93 5,48 0,82 -4,66 Notes: The nominal 2004 p.a.e. figure is deflated into a real 1998 p.a.e. measure by dividing the nominal 2004 values by a CPI based on the foodbasket equal to 216.3 and multiplying it with 100. The figures are elasticities of FGT poverty measures (P0, P1, P2) with respect to a simulated 10 percent decrease of Gini inequality index. The change of Gini is done via the following transformation of the actual income structure: 1. shift of all incomes by a fixed amount (lump-sum transfer); 2. normalization of incomes to bring the mean of the new distribution to the mean of the original distribution (tax on incomes). Table A.16b: Elasticity of Poverty with Respect to the Inequality Poverty Headcount Rate (P0) Poverty Gap (P1) Squared Poverty Gap (P2) Pooled 2004 1998 change Pooled 2004 1998 change Pooled 2004 1998 change Urban 0,11 0,14 0,00 -0,14 1,34 1,45 1,08 -0,37 2,36 2,52 2,01 -0,51 Rural -0,08 0,02 -0,11 -0,13 0,52 1,17 0,11 -1,06 1,12 2,15 0,36 -1,79 Total -0,06 0,04 -0,11 -0,16 0,61 1,22 0,17 -1,05 1,28 2,22 0,49 -1,72 Urban 0,94 0,87 1,01 0,15 2,31 2,43 2,05 -0,38 3,60 3,79 3,18 -0,61 Rural -0,07 0,46 -0,24 -0,70 0,90 2,00 0,24 -1,77 1,75 3,28 0,62 -2,66 Total 0,02 0,51 -0,22 -0,73 1,05 2,08 0,34 -1,74 1,98 3,38 0,82 -2,55 Poverty line = 33562.4 Poverty line = 24164.9 Notes: The nominal 2004 p.a.e. figure is deflated into a real 1998 p.a.e. measure by dividing the nominal 2004 values by a CPI equal to 335.51 and multiplying it with 100. The figures are elasticities of FGT poverty measures (P0, P1, P2) with respect to a simulated 10 percent decrease of Gini inequality index. The change of GINI is done via the following transformation of the actual income structure: 1. shift of all incomes by a fixed amount (lump-sum transfer); 2. normalization of incomes to bring the mean of the new distribution to the mean of the original distribution (tax on incomes). 435 Figure A.1a: Rural Growth-Incidence Rural Growth-incidence 95% confidence bounds Growth in mean Mean growth rate 49 36 23 10 -3 -16 1 10 20 30 40 50 60 70 80 90 100 Expenditure percentiles Notes: The nominal 2004 p.a.e. figure is deflated into a real 1998 p.a.e. measure by dividing the nominal 2004 values by a CPI based on the foodbasket equal to 216.3 and multiplying it with 100. Source: Authors' calculations using Adept version 4.1. Figure A.1b: Total, Urban and Rural Growth-Incidence Total (years 2004 and 1998) 95% confidence bounds Growth in mean Mean growth rate 39 Annual growth rate % 39 Urban Growth-incidence 27 15 3 -9 27 15 3 -9 -21 -21 1 10 20 30 40 50 60 70 80 90 100 Expenditure percentiles 1 10 20 30 40 50 60 70 80 90 100 Expenditure percentiles Rural 39 27 15 3 -9 -21 1 10 20 30 40 50 60 70 80 90 100 Expenditure percentiles Notes: The nominal 2004 p.a.e. figure is deflated into a real 1998 p.a.e. measure by dividing the nominal 2004 values by a CPI equal to 335.51 and multiplying it with 100. Source: Authors' calculations using Adept version 4.1. 436 Figure A.2a: Cumulative Distribution, Rural Rural 1 .8 .6 2004 1998 .4 .2 0 0 84 168 252 336 420 Welfare indicator, '000 Notes: The nominal 2004 p.a.e. figure is deflated into a real 1998 p.a.e. measure by dividing the nominal 2004 values by a CPI based on the foodbasket equal to 216.3 and multiplying it with 100. Source: Authors' calculations using Adept version 4.1. Figure A.2b: Cumulative Distribution, Total, Urban and Rural Total 1 .8 .6 2004 1998 .4 .2 0 0 56 112 168 224 280 Welfare indicator, '000 Urban Rural 1 Cumulative distribution 1 .8 .6 .4 .2 .8 .6 .4 .2 0 0 0 56 112 168 224 280 Welfare indicator, '000 0 56 112 168 224 280 Welfare indicator, '000 Notes: The nominal 2004 p.a.e. figure is deflated into a real 1998 p.a.e. measure by dividing the nominal 2004 values by a CPI equal to 335.51 and multiplying it with 100. Source: Authors' calculations using Adept version 4.1. 437 Figure A.3a: Probability Density of Per Adult Equivalent Expenditure, Pooled Data Pooled .04 Total Urban Rural .03 .02 Mean .01 0 0 70 140 210 280 350 per adult equivalent expenditure, '000 Figure A.3b: Probability Density of Per Adult Equivalent Expenditure, 2004 Dataset 2004 .02 Total Urban Rural .015 .01 Mean .005 0 0 80 160 240 320 400 PAE 2004 converted into 1998 values, '000 Figure A.3c: Probability Density of Per Adult Equivalent Expenditure, 1998 Dataset 1998 .08 Total Urban Rural .06 .04 Mean .02 0 0 40 80 120 per adult equivalent expenditure, '000 Source: Authors' calculations using Adept version 4.1. 438 160 200 Figure A.3d: Probability Density of Per Adult Equivalent Expenditure, Pooled, 2004 and 1998 Data Pooled 2004 .03 .02 Mean 0 0 50 .025 Urban Rural .02 .015 .01 .005 Mean .04 .01 Total Probability density function .05 0 100 150 200 250 0 per capita expenditure, '000 60 120 180 240 300 (pae04/335.5075)*100, '000 1998 .08 .06 .04 Mean .02 0 0 40 80 120 160 200 per capita expenditure, '000 Notes: The nominal 2004 p.a.e. figure is deflated into a real 1998 p.a.e. measure by dividing the nominal 2004 values by a CPI equal to 335.51 and multiplying it with 100. Source: Authors' calculations using Adept version 4.1. 439 Figure A.4a: Lorenz Curve (GINI coefficient), Rural Dataset Eastern Province Rural 1 2004, Gini=53.55 1998, Gini=56.3 .8 .6 .4 .2 0 0 .2 .4 .6 .8 1 Cumulative population proportion Figure A.4b: Lorenz Curve (GINI coefficient), Total, Urban & Rural Dataset Eastern Province 1 Total 2004, Gini=53.68 .8 1998, Gini=58.04 .6 .4 .2 0 0 .2 .4 .6 .8 1 Cumulative population proportion 1 Urban 1 Rural 2004, Gini=53.84 .8 2004, Gini=53.55 .8 Lorenz curve 1998, Gini=44.97 .6 .4 1998, Gini=56.3 .6 .4 .2 .2 0 0 0 .2 .4 .6 .8 1 Cumulative population proportion 0 .2 .4 .6 .8 Cumulative population proportion Notes: The nominal 2004 p.a.e. figure is deflated into a real 1998 p.a.e. measure by dividing the nominal 2004 values by a CPI equal to 335.51 and multiplying it with 100. Source: Authors' calculations using Adept version 4.1. 440 1 Figure A.5a: Relative Pro-poor Curve, Upper Bound Confidence Interval Relative propoor curve 0 .02 .04 .06 .08 (Order : s=1 | Dif. = P_2( (m2/m1)z, a=s-1) - P_1(z,a=s-1)) 0 6712.48 13424.96 20137.44 Poverty line (z) Difference Null horizontal line 26849.92 33562.4 Upper bound of 95% confidence interval Sourceμ Author‘s calculations based upon DASP 1.4 (Araar and Duclos, 2007a, Araar and Duclos, 2007b). Figure A.5b: Relative Pro-poor Curve, Lower Bound Confidence Interval Relative propoor curve 0 .02 .04 .06 .08 (Order : s=1 | Dif. = P_2( (m2/m1)z, a=s-1) - P_1(z,a=s-1)) 0 6712.48 13424.96 20137.44 Poverty line (z) Difference Null horizontal line 26849.92 33562.4 Lower bound of 95% confidence interval Sourceμ Author‘s calculations based upon DASθ 1.4 (Araar and Duclos, 2007a, Araar and Duclos, 2007b). 441 Annex: Chapter 7 Key Informants: Geneva: Mr. Terje Tessem, Head of Employment Intensive Investment Programme, ILO. Jan. 2004 & July 2004. Ms. Linda Deelen, Social Finance Programme, ILO, Geneva, Switzerland, 5 th April, 2004. Lusaka: Mr. Carl-Eric Hedstrom, Project Manager, Road Training School, Lusaka, Zambia. Sept. 2003 - Sept. 2004. Mr. Frans Blokhuis, Chief Technical Adviser, Roads Department Agency, Lusaka, Zambia, 25 th of July, 2005. Mr. Alfred S. Sakwiya, EPFRP Project Manager, Lusaka, Zambia. 26th of July 2005. Mr. Nkumbu Siame, Senior Engineer Roads, National Road Board, Lusaka, Zambia, 2005. Dr. Dennis Chiwele, Ruralnet & Institute of Economic Research, UNZA, Lusaka, Zambia. 30 Sept., 2005. Dr Henrietta Kalinda, Independent Consultant and UNZA, Lusaka, Zambia. 26th of July, 2005. Ms. Mashinkila, Economic and Social Research Institute UNZA, Lusaka, Zambia, 26th of July 2005. Mr. Kalinda, Head of School of Agricultural Sciences, UNZA, Lusaka, Zambia. 2107-2005. Professor Oliver Saasa, UNZA & Premier Consult, Lusaka, Zambia. 25th of July 2005. Mr. Valentine Mwanza, Department of Development Studies UNZA, Lusaka, Zambia, 25th of July 2005. Mr. Zagis, Mwanza School of Mines UNZA, Lusaka, Zambia, 25th of July 2005. Mr. Justin Mubanga, Ministry of Finance & Planning (MoFNP), Lusaka, Zambia, 2005. Mr. Solomon Tembo, Head of LCMS-Unit, CSO, Lusaka, Zambia, 27th of July 2005. Mr. Masileso Sooka, Principal Statistician, Agriculture and Environment, CSO, Lusaka, Zambia, 2004 & 2005. Mr. Ollings Chihana, Agricultural Statistical Division, CSO, Lusaka, Zambia. 2004; 21 July 2005 & 27 July 2005. Mr. Nyasulu, Agricultural Statistical Division, CSO, Lusaka, Zambia. 27 th of July & 30 Sept. 2005. Mr. Simbeye, Project Coordinator, SHEMP – AFRICARE, Lusaka, Zambia, 2005. Dr. Dick Siame, Director, SHEMP – AFRICARE Headquarters, Lusaka, Zambia. 30 Sept. 2005. Dr. Jones Govereh, Food Security Research Project, Lusaka, Zambia. 28th of July 2005. Mr. Nicholas F. Mwale, Statistician, Database and Early Warning Unit, MAFF, Lusaka, Zambia, 2005. Mr. Silwimba, Mr. Mulenga, & G. Mbozi, Land Use Planning & Mapping Unit, MAFF, Lusaka, Zambia, 2005. Mr. Tadeyo Lungu, Programme Coordinator, Smallholder Enterprise & Marketing Programme, MAFF, Lusaka, Zambia. Mr. Daka, Programme Officer World Food Programme, Lusaka, Zambia. 27th of July 2005. Mr. Alan Mulando, Assessing & Mapping Unit, WFP, Lusaka, Zambia. Mr. Bangwe, FAO, Lusaka, Zambia. 30 Sept. 2005. Mr. Davies Makasa, World Bank, Lusaka, Zambia. 27th of July 2005. Chipata District: Mr. Simon Tembo, East Consult, Chipata, Zambia. 24th September 2005. Mr. N. Kabwe, Secretary and Mr. Manda, Chairman, Eastern Province Labour Based Road Contractors Association, Chipata, Zambia, 2004 & 2005. Mr. Sylvester Mwanza, Zamsif Regional Office, NPF Building, Chipata, Zambia, 2005. Mr. Ronald Daka Director of Engineering Services Chipata Municipal Council, Zambia 27th of Sept 2005. Mr Alikhadio Maseko, District Agriculture Coordinator (DACO), Chipata, Zambia, Sept. 2005. Mr. Gershom Jere, Crops Husbandry Officer, DACO, MAFF, Chipata, Zambia, 19 th of Sept. 2005. Mr. V.Y. Mumba, Senior Field Service Coordinator, MAFF, Chipata, Zambia, 2005. Mr. Kapuka, Senior Agriculture Officer, DACO, Chipata, Zambia, 2005. Mr. Joseph Ngulube, District Marketing and Cooperative Officer (DMCO), MAFF, Chipata, Zambia, 2005. Dr. Hantuba, Deputy Director, Marketing & Cooperatives, DACO, Chipata, Zambia, 2005. Mr. G. Mbozi, Director of Planning, DACO, Chipata, Zambia, 2005. 442 Mr. Phillip Tembo, CSO, Chipata, Zambia, 2005. Mr. Agrippa Mwanza, Rapid Construction, Chipata, Zambia. 22 nd September, 2005. Mr. Ambroise Lufuma, African Development Bank Project Office, Chipata, Zambia. 28 th Sept. 2005. Mr. Bwembya, Provincial Local Government Officer, DDP, Chipata, Zambia. 26 th Sept. 2005. Mr. Clifford Nkhuma, Branch Manager Shoprite, Chipata, Zambia. 26 th September, 2005. Dr. Roy Musonda Chiti, Agriculture Support Programme, Chipata, Zambia. 21st of September, 2005. Mr. Henry Dane Zulu, Mtondo Building Contractors, Chipata, Zambia. 21st of September 2005. Mr. Isaac Manda, FRP RTS contractor, Chipata, Zambia. 25th of September 2005. Mr. Jani Ngoma, Regional Coordinator, Food Reserve Agency, Chipata, Zambia, 26th of Sept. 2005. Mr. John Kanenga, Regional Manager Africare Eastern Province, Chipata, Zambia. 27th Sept 2005. Mr. Kahwa Ruger, HSO UNV WFP Chipata Sub Office, Chipata, Zambia, Sept. 2005. Mr. Kamoto Ephraim, Wheeltrax Contractors, Chipata, Zambia. 28th Sept 2005. Mr. Mabaye Mpulu, Regional Manager PAM, Chipata, Zambia. 28th of Sept 2005. Mr. Michael Njobvu, CSO Chipata, Chipata, Zambia. 28th of September 2005. Mr. Mike Daka, Manager Radio Breeze, Chipata, Zambia. 21st of September 2005. Mr. Amos Phiri, Village Industry Accountant and Business trainer, Chipata, Zambia. 19th of Sept. 2005. Lundazi District: District Commissioner, Lundazi District, 2908-2005. District Agricultural Officer, 2908-2005. Mr. D.J. Sikazwe, Counsel Secretary, Lundazi District, 2908-2005. Mr. Jere, Planning Officer, District Planning Unit, Lundazi District, Zambia, 2908-2005. Mr. Munoni, Director of Works, Lundazi District Council. Most of the interviews were recorded and can be heard on Windows Media Player. These recordings are available upon request. 443 Table A1: Allocations of SEAs & Sample Villages in EPRHS 2005 PSU 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Survey Sites Mafuta Chiweteka Kwacha Jelo Farm John Shawer Kalume Kalinga Fulato Zimena Chimanga Chuni Kalunga Mulemba Bila Kapela Chikhumbi Chaloka Kauka Chimazuma Kapaika Kachindila District Chipata Chipata Chipata Chipata Chipata Chipata Chipata Chipata Chipata Chipata Chipata Chipata Lundazi Lundazi Lundazi Lundazi Lundazi Lundazi Lundazi Lundazi Constituency Chipangali Kasenengwa Chipata Central Chipata Central Chipata Central Chipata Central Chipata Central Luangeni Kasenengwa Chipata Central Chipangali Luangeni Chasefu Lundazi Central Lundazi Central Lundazi Central Chasefu Chasefu Lundazi Central Chasefu Ward Rukuzye Chiparamba Kanjala Kanjala Msanga Msanga Kanjala Khova Kwenje Dilika Chipangali Nsingo Nkhanga Msuzi Lunevwa Vuu Manda Hill Kajilime Vuu Membe CSA 6* 2* 3* 97 93 95 109 104 28 129 142 125 23 55 67 30 2 17 28 7** SEA 2* 4* 3* 2 4 1 3 2 4 3 1 3 1 1 1 1 1 2 1 1** Block Chanje Chipata Central Chipata Central Chipata Central Chipata Central Chipata Central Chipata Central Kalunga (Eastern) Kwenje I Chipata Central Chankhadze Kalunga (Eastern) Lundazi Central Lundazi Central Mwase Lundazi Central Emusa Emusa Lundazi Central Lundazi Central Camp Chief Mkanda Mkanda Mtaya Chikuwe Kanyanya Chinyaku Chinjara Chinyaku Katopola Kapata-Moyo Mtaya Chikuwe Chisifu Chinyaku Mtowe Sairi Kwenje I Mzamane Nsanjika Mpezeni Chankhadze Sairi Katambo Mpezeni Nkhanga Magodi Mapara Kapichila Pharaza Mwase-Lundazi Kaithinde Mphamba Mwata Magodi Mnuyukwa Magodi Vuu Mphamba Phikamalaza Phikamalaza Latitude -13,28 -13,33 -13,42 -13,68 -13,53 -13,57 -13,72 -13,87 -13,80 -13,68 -13,27 -13,84 -12,15 -12,34 -12,40 -12,32 -11,70 -12,14 -12,27 -12,23 Note: Convert: http://www.fcc.gov/mb/audio/bickel/DDDMMSS-decimal.html Degrees, Minutes, Seconds and Decimal Degrees Latitude/Longitude Conversions. Source: Authors. 444 Longi- Altitude tude Meter 32,47 1053 32,21 969 32,37 1075 32,50 993 32,61 973 32,50 1001 32,53 1047 32,50 1124 32,12 1032 32,77 1207 32,84 1052 32,75 1232 33,05 1126 33,22 1186 33,28 1166 33,13 1155 33,25 1153 33,03 1127 33,02 1096 33,28 1208 Map A1: LCMS-I 1996 & EPRHS 2005 SEAs -11 32 32,2 32,4 32,6 32,8 33 -11,5 L a t i t u d e 33,2 33,4 -11,4287 -12 -12,019 -12,09318 -12,11083 -12,16098 -12,19638 -12,20907 -12,29771822 -12,5 Lundazi District Centre -13 M12 Chipata - Lundazi District Road -13,16426 -13,5 -13,48914 -13,28015 -13,3153 -13,33407 -13,34645 -13,41159 -13,42031 -13,43685 -13,52987 -13,50809 -13,63087908 Chipata District Centre -14 Longitude Source: Authors. 445 EPRHS Field Sites Table A2a: Nutrition (calorie) based equivalence scales Years of Age Men Women 0-1 0,33 0,33 1-2 0,46 0,46 2-3 0,54 0,54 3-5 0,62 0,62 5-7 0,74 0,70 7-10 0,84 0,72 10-12 0,88 0,78 12-14 0,96 0,84 14-16 1,06 0,86 16-18 1,14 0,86 18-30 1,04 0,80 30-60 1,00 0,82 60 plus 0,84 0,74 Table A2b: Conversion table for area Acres and Limas to Hectares Acres Hectares Lima Hectares 0,33 0,25 0,50 0,67 0,75 1 2 3 4 5 6 7 8 9 10 15 20 .10 .13 .20 .27 .30 .40 .81 1.22 1.62 2.02 2.43 2.84 3.24 3.64 4.05 6.08 8.10 0,25 0,33 0,50 0,67 0,75 1 2 3 4 5 6 7 8 9 10 11 12 .06 .08 .12 .17 .19 .25 .50 .75 1.00 1.25 1.50 1.75 2.00 2.25 2.50 2.75 3.00 Source: Calculated from World Health Organisation Source: Data by Dearcon & Pramila Krishnan, 1998:40. http://www.metric-conversions.org/area/hectares-toacres.htm Table A3.1: Employment Generation by the EPFRP, 1996-2001 Direct Employment Generated by Rehabilitation and Maintenance Road Works Chipata District Lundazi District Chadiza District Katete District Petauke District Mambwe District Eastern Province Female Participation as % of total Rehabilitation Works Male Female Total 183505 26274 209779 126757 21103 147859 105809 13196 119005 114251 19942 134193 125674 27765 153439 28979 5350 34329 684975 113630 798604 Maintenance Works Male Female Total 28140 3365 31505 10719 805 11524 11818 975 12793 8525 1227 9752 6157 1733 7890 0 0 0 65359 8105 73464 14% 11% Source: Authors‘ calculations based upon Rwampororo et al., β00β. Status at the end of November 2001 Female Combined Total Worker as % of Days Generated Male Female Total Total 211645 29639 241284 12 137476 21908 159383 14 117627 14171 131798 11 122776 21169 143945 15 131831 29498 161329 18 28979 5350 34329 16 750334 121735 872068 14 14% Table A3.2: Total Earnings from the EPFRP, 1996-2001 Total Earnings in Wages over project Avg Earning per Avg Earning per Avg Earning per Percentage Average Total Earning Per Year Worker Per Year Worker Per Day Worker Per Month life, 1996-2001 (ZKW) Total Working years Full-time Jobs per Year Male Female Total Male Female Total Male Female Total of Total Male Female Total Total Total Total Chipata District 881,85 123,50 1005,35 146,98 20,58 167,56 561762020,3 78669775 640431795 28% 93627003,4 13111629 106738633 637023,72 2654,27 53085,31 Lundazi District 572,82 91,28 664,10 95,47 15,21 110,68 364897802,9 58149648 423044797 18% 60816300,5 9691608 70507466 637023,72 2654,27 53085,31 Chadiza District 490,11 59,05 549,16 81,69 9,84 91,53 312213287,2 37613596 349826883 15% 52035547,9 6268933 58304481 637023,72 2654,27 53085,31 Katete District 511,57 88,20 599,77 85,26 14,70 99,96 325880100,2 56188146 382068246 17% 54313350 9364691 63678041 637023,72 2654,27 53085,31 Petauke District 549,30 122,91 672,20 91,55 20,48 112,03 349914474,2 78295524 428209998 18% 58319079 13049254 71368333 637023,72 2654,27 53085,31 Mambwe District 120,75 22,29 143,04 20,12 3,72 23,84 76917959,72 14200320 91118280,1 4% 12819660 2366720 15186380 637023,72 2654,27 53085,31 Eastern Province 3126,39 507,23 3633,62 521,07 84,54 605,60 1991585644 323117010 2314700000 100% 331930941 53852835 385783333 637023,72 2654,27 53085,31 Percentage of total 86% 14% 100% 86% 14% 100% 86% 14% 100% Sourceμ Authors‘ calculations based upon Rwampororo et al., β00β. Notes: EPFRP project life = 6 years; Working weeks per year = 48; Working days per year = 240. 446 Map Aβ: Chipata District, Zambia‟s Eastern Province LUNDAZI DISTRICT RD115 RD115 M 12 RD 693 GW A R262 R261 RD693 RD115 Mwandauka D104 R264 Chinunde RD 694 U28 CHIPATA DISTRICT R 263 D791 R D697 R265 RD116 D104 E T R D113 U29 RD710 R258 R272 B28 R346 RD599 B29 CHIPAT A D123 R347 B66 B30 RD401 U33 B65 U27 RD400 RD402 B32 T4 D128 U31 RD595 D598 RD596 KATETE U32 DISTRICT MALAWI T4 B31 R280 D124 R270 RD121 D123/B64 B65 R269 RD116 R273 D123/B64 R268 RD119 R 271 R D712 M soro R 266 RD118 R267 RD706 RD 695 R345 RD709 R D711 RD694 RD695 RD117 R274 RD403 RD 403 R348 R275/R 349 D404 D125 D128 CHADIZA DISTRICT Source: Table A4: Chipata Road Network Road T4 U32 D128 T4 RD403 T4 D124 D123 D123/B64 D598 T4 D104 D791 M12 M12 RD116 RD118 X 32.39497 32.42010 32.58180 32.68887 32.77737 32.30101 32.28462 32.42556 32.13822 32.02896 32.20486 32.56104 32.14805 32.66920 32.82216 32.74022 32.85931 Y -13.81481 -13.92735 -13.88474 -13.67387 -13.79515 -13.98977 -13.76783 - 13.64109 -13.61487 -13.81263 -13.94046 - 13.52528 -13.29803 -13.51545 -13.41056 -13.39090 - 13.41275 MED_DESCRI Without Median RTT_DESCRI Primary Route Without Median Primary Route Without Median Secondary Route Without Median Primary Route Without Median Primary Route Without Median Secondary Route Without Median Secondary Route Without Median Secondary Route Without Median Secondary Route Source: Authors based on DIVA-GIS 5.2. 447 F_CODE_DES Description Road from D123 junction to U32 junction Trail from T4 junction to D128 junction Trail From U32 to T4 junction Road From D128 junction towards Malawi Border Road From T4 in the direction of D128 Junction Road From U33 to D124 Junction Trail From T4 to D123 Junction Trail From D124 to T4 Junction Trail From D124 to D598 junction Trail From D123/B64 to T4 Junction Road From D598 to D124 Junction Trail From M12 to D791 Junction Road From D104 Junction towards Luangwa Valley Road From D104 to RD116 Junction Road From DR119 to RD118 Junction Trail From M12 to RD117 Junction Road From M12 [EPFRP] Map A3: Lundazi District, Zambia‟s Eastern Province D103 NORTHERN PROVINCE CHAMA R238 RD107 DISTRICT R237 MPIKA RD107 DISTRICT R241 U19 U19 R240 R239 RD107 U19 MALAWI R243 RD105 R242 RD108 R245 R246 R247 R248 R244 R243 Luangwa LUNDAZI River RD110N R249 U18 D104 RD110N D104 R250 R251 RD110 R256 R255 U20 D104 R254 RD692 R252R253 U16 RD692 R254 Mwanya RD182 R257 MALAWI M12 D104 CHIPATA DISTRICT Source: Table A5: Lundazi Road network Road M12 RD110 RD110 RD110N R249 D103 D103 D104 D104 RD108 RD105 M12 RD692 X 33.17288 33.21658 33.28104 33.25919 33.24389 33.18490 33.21221 33.10405 32.99042 32.91285 32.58726 33.11060 32.74787 Y MED_DESCRI RTT_DESCRI F_CODE_DES -12.30051 Without Median Secondary Route Road -12.36934 Trail -12.41196 Trail -12.32455 Trail -12.26009 Without Median Secondary Route Road -12.18252 Without Median Secondary Route Road -11.81323 Without Median Secondary Route Road -12.32236 Without Median Secondary Route Road -12.40103 Without Median Secondary Route Road -12.24261 Trail -12.27975 Trail -12.44910 Without Median Secondary Route Road -12.68947 Trail Source: Authors based on DIVA-GIS 5.2. 448 Description From Lundazi to U18 Junction From M12 to RD110N Junction [EPFRP] From M12 to RD110N Junction [EPFRP] From RD110 to R249 [EPFRP] From Lundazi to Malawi From Lundazi to Chama From R237 in direction of Chama From Lundazi to RD108 From RD108 to RD105 From D104 to RD105 From RD108 to D104 From RD110 to RD692 From M12 to D104 Junction Map A4: Chipata Survey Sites Source: Authors based on Google Earth. Map A5: Lundazi Survey Sites Source: Authors based on Google Earth. 449 Table A6.1: JCTR Urban Basic Needs Basket (Cost of essential food and non-food items for a family of six), 2005 Luanshya Ndola Lusaka Kitwe Livingstone Kabwe Zambia (CSO) 2005 2005 Total Basic Needs Basket Cost of Basic Food Total Basic Needs Basket pc Cost of Basic Food pc 838410 511610 139735,0 85268,3 988280 501180 164713,3 83530,0 1340840 492940 223473,3 82156,7 973910 479210 162318,3 79868,3 1108750 453350 184791,7 75558,3 842900 424780 140483,3 70796,7 920441 640982 153406,8 106830,3 Sourceμ Authors‘ based on Jesuit ωentre for Theological Reflection. Table A6.2: JCTR Price Deflator Lusaka Weight Weight JCTR The food basket Jul. 1996 176660 Jul. 1998 206150 Jul. 2000 277475 Jul. 2001 332520 Jan. 2002 324650 Apr. 2003 403576 Jan. 2004 417298 Jul. 2004 432645 Jan. 2005 492940 Jul. 2005 513870 Price Deflator: 2005/1996 2,9088 Food & Basic Needs The food basket 451240,74 39,15% 526566,73 39,15% 708751,41 39,15% 849352,26 39,15% 829250 39,15% 1030850 39,15% 1065900 39,15% 1105100 39,15% 1340840 36,76% 1361770 37,74% 3,0178 Basic Needs 60,85% 60,85% 60,85% 60,85% 60,85% 60,85% 60,85% 60,85% 63,24% 62,26% The food basket 424780 494710 Food & Basic Needs The food basket Food & Basic Needs The food basket Food & Basic Needs The food basket Food & Basic Needs 842900 50,40% 49,60% 479210 973910 49,20% 50,80% 981410 50,41% 49,59% 496890 1042990 47,64% 52,36% The food basket 511610 497100,0 Ndola Weight Weight Food & Basic Needs The food basket Food & Basic Needs The food basket Food & Basic Needs The food basket Food & Basic Needs 838410 61,0% 39,0% 501180 988280 50,71% 49,29% 889120,0 55,91% 44,09% 557830 1142030 48,85% 51,15% The food basket 453350 501700 Livingstone Weight Weight Average Average Weight Weight Food & Basic Needs The food basket Food & Basic Needs The food basket Food & Basic Needs The food basket Food & Basic Needs 175207,86 347093,22 50,48% 49,52% 1108750 40,89% 59,11% 474026 950450 49,87% 50,13% 1181800 42,45% 57,55% 509646 1047470 48,65% 51,35% Average/Lusaka The food basket 99,18% 96,16% 99,18% 96,16% 99,18% Average/Lusaka Food & Basic Needs 76,92% 70,88% 76,92% 70,88% 76,92% Kabwe Weight Weight Kitwe Weight Weight Luanshya Weight Weight Source: Authors‘ calculations based upon JωTR datasets. 450 2,9088 3,0178 Table A7: Poverty Lines & Food Baskets in National currency and US Dollars, 1996 - 2006 Per Month Lower (Extreme) Poverty Line Upper (Moderate) Poverty Line Lower (Extreme) Poverty Line Upper (Moderate) Poverty Line Currency ZMK ZMK USD USD Jul. 1996 28777,45 41324,01 22,65 32,53 Jul. 1998 36345,91 52192,22 18,71 26,87 Exchange Rate OER = ZMK/USD 1270,29 1942,65 3112,25 3699 4779,33 4590,94 3679,445 The food basket ZMK 172664,68 218075,49 274557,04 321506,30 396738,77 476880,00 569724,00 676187,00 665132,00 Food & Basic Needs ZMK 247944,06 313153,35 394260,07 461678,54 569711,31 684793,00 818116,00 970995,00 1055041,00 The food basket Food & Basic Needs Per Day Lower (Extreme) Poverty Line Upper (Moderate) Poverty Line Lower (Extreme) Poverty Line Upper (Moderate) Poverty Line USD USD 135,93 195,19 112,26 161,20 88,22 126,68 86,92 124,81 92,12 132,28 100,73 144,65 119,21 171,18 147,29 211,50 180,77 286,74 ZMK ZMK USD USD 959,25 1377,47 0,755 1,084 1211,53 1739,74 0,624 0,896 1525,32 2190,33 0,490 0,704 1786,15 2564,88 0,483 0,693 2204,10 3165,06 0,512 0,735 2649,33 3804,41 0,560 0,804 3165,13 4545,09 0,662 0,951 3756,59 5394,42 0,818 1,175 3695,18 5861,34 1,004 1,593 Exchange Rate OER = ZMK/USD 1270,29 1942,65 3112,25 3699 4306,92 4734,01 4779,33 4590,94 3679,445 The food basket Food & Basic Needs The food basket Food & Basic Needs ZMK ZMK USD USD 5755,49 8264,80 4,531 6,506 7269,18 10438,44 3,742 5,373 9151,90 13142,00 2,941 4,223 10716,88 15389,28 2,897 4,160 13224,63 18990,38 3,071 4,409 15896,00 22826,43 3,358 4,822 18990,80 27270,53 3,974 5,706 22539,57 32366,50 4,910 7,050 22171,07 35168,03 6,026 9,558 Sourceμ Authors‘ calculations based on data from ψηZ, ωSη & IεF WEη. Jul. 2000 45759,51 65710,01 14,70 21,11 451 Jul. 2001 53584,38 76946,42 14,49 20,80 Jul. 2002 66123,13 94951,89 15,35 22,05 Jul. 2003 79480,00 114132,17 16,79 24,11 4306,92 4734,01 Jul. 2004 94954,00 136352,67 19,87 28,53 Jul. 2005 112697,83 161832,50 24,55 35,25 Jul. 2006 110855,33 175840,17 30,13 47,79 Table A8a: Changes in Comprehensive Total Expenditure and Poverty using Food basket deflator, 1996 and 2005 (%) Food Pove rty Line 19132,95 Price De fla tor 3,9162 1996 2005 Nomina l Va lue s 1996 2005 Re e l Va lue s 1996 Grow th (% p.a .) Consumption Grow th: Δln cons = ln C2005 - ln C1996 1996 2005 obse rva ti ons 2005 Loga rithm Comprehe nsive Ex pe nditure, pa e Comprehe nsive Ex pe nditure, pa e Comprehe nsive Ex pe nditure, pa e Comprehe nsive Ex pe nditure, pa e Comprehe nsive Ex pe nditure, pa e Comprehe nsive Ex pe nditure, pa e Comprehe nsive Ex pe nditure, pa e Comprehe nsive Ex pe nditure, pa e Trea tme nt 30783,77 47263,83 30783,77 12068,84 10,33 9,40 -9,88% -0,936 104 43 Re gions Compa rison 35727,46 66136,07 35727,46 16887,88 10,48 9,73 -7,99% -0,749 137 44 ma le -he a de d house holds 35363,87 61986,71 35363,87 15828,34 10,47 9,67 -8,54% -0,804 186 63 by gender of head: fe ma le -he a de d house holds 27608,98 43215,36 27608,98 11035,06 10,23 9,31 -9,69% -0,917 55 24 Tota l Ex pe nditure pa e including Ow n-consumption Age he a d (20-40) 32716,17 69027,5 32716,17 17626,21 10,40 9,78 -6,64% -0,618 126 47 by age of head: Age he a d (>40) 34555,97 42450,98 34555,97 10839,88 10,45 9,29 -12,09% -1,159 115 40 36917,82 58232,86 36917,82 14869,79 10,52 9,61 -9,61% -0,909 177 74 by education level of head: he a d ha s more tha n 5 ye a rs of e duca tion he a d ha s le ss tha n 5 or no e duca tion 24401,86 48699,98 24401,86 12435,56 10,10 9,43 -7,22% -0,674 64 13 la rge house hold (>5) 40013,8 46639,23 40013,8 11909,35 10,60 9,39 -12,60% -1,212 82 53 Sma ll house hold (<=5) 30283,29 72660,36 30283,29 18553,86 10,32 9,83 -5,30% -0,490 159 34 La rge la nd ow ne rship (>0.5/pe rson) 36688,43 74679,37 36688,43 19069,41 10,51 9,86 -7,01% -0,654 128 25 by household size: Sma ll la ndow ne rship (<=0.5/pe rson) or la ndle ss by land ownership: Whole Source: Authors‘ calculations. 30088,99 49602,38 30088,99 12665,99 10,31 9,45 -9,17% -0,865 113 69 33594,08 56808,41 33594,08 14506,06 10,42 9,58 -8,91% -0,840 241 87 Table A8b: Changes in Comprehensive Total Expenditure and Poverty using food basket deflator adjusted for US-dollars changes, 1996 and 2005 (%) Food Pove rty Line 19132,95 Price De fla tor 1,1185 Tota l Ex pe nditure pa e including Ow nconsumption Comprehe nsive Comprehe nsive Ex pe nditure, Ex pe nditure, pa e pa e 1996 Grow th (% p.a .) Consumption Grow th: Δln cons = ln C2005 - ln C1996 Comprehe nsive Ex pe nditure, pa e Comprehe nsive Ex pe nditure, pa e 2005 Loga rithm Comprehe nsive Comprehe nsive Comprehe nsive Ex pe nditure, Ex pe nditure, Ex pe nditure, pa e pa e pa e 1996 2005 obse rva tions 47263,83 30783,77 42255,15 10,33 10,65 3,58% 0,317 104 Compa rison 35727,46 66136,07 35727,46 59127,44 10,48 10,99 5,76% 0,504 137 44 ma le -he a de d house holds 35363,87 61986,71 35363,87 55417,80 10,47 10,92 5,12% 0,449 186 63 fe ma le -he a de d house holds 27608,98 43215,36 27608,98 38635,70 10,23 10,56 3,80% 0,336 55 24 Age he a d (20-40) 32716,17 69027,5 32716,17 61712,46 10,40 11,03 7,31% 0,635 126 47 Age he a d (>40) by education level of head: he a d ha s more tha n 5 ye a rs of e duca tion he a d ha s le ss tha n 5 or no e duca tion by land ownership: 2005 Re e l Va lue s 30783,77 by age of head: by household size: Comprehe nsive Ex pe nditure, pa e 1996 Trea tme nt Re gions by gender of head: 1996 2005 Nomina l Va lue s 43 34555,97 42450,98 34555,97 37952,33 10,45 10,54 1,05% 0,094 115 40 36917,82 58232,86 36917,82 52061,76 10,52 10,86 3,89% 0,344 177 74 13 24401,86 48699,98 24401,86 43539,10 10,10 10,68 6,64% 0,579 64 la rge house hold (>5) 40013,8 46639,23 40013,8 41696,74 10,60 10,64 0,46% 0,041 82 53 Sma ll house hold (<=5) 30283,29 72660,36 30283,29 64960,33 10,32 11,08 8,85% 0,763 159 34 La rge la nd ow ne rship (>0.5/pe rson) 36688,43 74679,37 36688,43 66765,38 10,51 11,11 6,88% 0,599 128 25 Sma ll la ndow ne rship (<=0.5/pe rson) or la ndle ss 30088,99 49602,38 30088,99 44345,87 10,31 10,70 4,40% 0,388 113 69 33594,08 56808,41 33594,08 50788,26 10,422 10,835 4,70% 0,413 241 87 Whole Source: Authors‘ calculations. 452 Figure A1-A2: Poverty Deficit Curves for Treatment and Comparisons Regions, 1996 and 2005 0 0 5 20 10 40 15 60 20 80 Poverty Deficit Curves for Treatment and Comparison Regions, 1996Poverty Deficit Curves for Treatment and Comparison Regions, 2005 0 50000 100000 Totexppae_Com Treatment Regions 150000 200000 0 50000 Comparison Regions 100000 150000 Totexppae_Com Treatment Regions 200000 250000 Comparison Regions Source: 0 0 10000 20000 20000 40000 30000 60000 40000 Figure A3-A4 Generalized Lorenz curve ordinates, 1996 and 2005 0 .2 .4 .6 Cumulative population proportion Treatment==0 .8 1 0 .2 Treatment==1 .4 .6 Cumulative population proportion Treatment==0 .8 1 Treatment==1 Source: Table A9: Income Distribution by Population Deciles (LCMS 1996) & Income Distribution by Population Deciles (EPRHS 2005), p.a.e., in nominal values Rural Eastern Province, 1996* Indicator Minimum First Decile Second Decile Third Decile Fourth Decile Fifth Decile Sixth Decile Seventh Decile Eight Decile Nineth Decile Max Mean Skewness Kurtosis Lower (Extreme) Poverty Line % Below Line Upper (Moderate) Poverty Line % Below Line Gini Coefficient Observations Rural Eastern Province, 2005* Total Food Share of Total Food Share of Expenditure Expenditure Food Expenditure Expenditure Food 2516,99 1755,32 69,74% 950,87 950,87 100% 8555,13 6622,64 77,41% 13109,76 7402,913 56,47% 10413,44 8485,48 81,49% 20737,34 12929,16 62,35% 13563,41 10190,27 75,13% 26512,15 16127,97 60,83% 16616,67 12800 77,03% 29959,24 18868,2 62,98% 19917,23 15248,38 76,56% 34551,86 22942,64 66,40% 24412,96 18224,8 74,65% 41362,3 27533,54 66,57% 30858,91 22400 72,59% 49786,32 32000 64,27% 41805,55 28700 68,65% 66250 44198,05 66,71% 66349,1 45268,54 68,23% 91602,4 58248,47 63,59% 1276187 1258976 98,65% 689241,7 441700 64,08% 38189,79 29166,04 76,37% 54263,48 34268,05 63,15% 10,480 11,658 5,782 6,045 129,193 153,413 42,362 46,847 20181 79032,55 75,26% 82,93% 88,78% 93,88% 28979,4 113487,15 83,97% 90,24% 93,88% 95,92% 287 287 98 Sourceμ Authors‘ calculations. Notes: * Chipata and Lundazi Districts; ** CSO calculations. 453 98 All Zambia, 1996** Total Expenditure Food Expenditure Share of Food 4127,6 7110,9 9758,1 12272,6 15353,6 19114,5 24022,5 31211,2 44121,4 114574,4 28261 3324,1 5366,4 7231,3 8456,4 10009,6 12124,4 14293,1 17976,3 23283,8 47625,6 15016,6 80,5% 75,5% 74,1% 68,9% 65,2% 63,4% 59,5% 57,6% 52,8% 41,6% 53,1% 20181 28979,4 0 .2 .4 .6 .8 1 Figure A5.1: Poverty Incidence Curves, 1996 & 2005. 0 50000 100000 150000 Welfare indicator Poverty incidence curve in year 1 200000 250000 Poverty incidence curve in year 2 Notes: Year 1 = 1996 and Year 2 = 2005. Source: 0 .2 .4 .6 .8 1 Figure A5.2: Poverty Incidence Curves for Treatment Areas, 1996 & 2005. 0 20000 40000 Welfare indicator Poverty incidence curve in year 1 60000 80000 Poverty incidence curve in year 2 Notes: Year 1 = 1996 and Year 2 = 2005. Source: 0 .2 .4 .6 .8 1 Figure A5.3: Poverty Incidence Curves for Comparison Areas, 1996 & 2005. 0 20000 40000 60000 W elfare indicator Poverty incidence curve in year 1 Notes: Year 1 = 1996 and Year 2 = 2005. Source: Authors estimations. 454 80000 100000 Poverty incidence curve in year 2 Figure A6a: Difference, Between FGT Curves, Alpha=0 -.05 0 .05 .1 .15 Difference between FGT Curves (alpha=0) 0 3826.59 7653.18 11479.77 15306.36 19132.95 Poverty line (z) Null horizontal line FGT_1 - FGT_0 Sourceμ Authors‘ based on DASP version 2.0. Figure A6b: Difference, Between FGT Curves, Alpha=1 0 .01 .02 .03 .04 Difference between FGT Curves (alpha=1) 0 3826.59 7653.18 11479.77 15306.36 19132.95 Poverty line (z) Null horizontal line FGT_1 - FGT_0 Sourceμ Authors‘ based on DASP version 2.0. Figure A6c: Difference, Between FGT Curves, Alpha=2 0 .005 .01 .015 .02 Difference between FGT Curves (alpha=2) 0 3826.59 7653.18 11479.77 15306.36 Poverty line (z) Null horizontal line Sourceμ Authors‘ based on DASP version 2.0. 455 FGT_1 - FGT_0 19132.95 Fig. A7b: Censored Nature of data, 2005 0 0 100 50 200 300 100 400 150 500 Fig. A7a: Censored Nature of data, 1996 0 5000 10000 Poverty_Gap 15000 0 20000 40000 Poverty_Gap 60000 80000 Sourceμ Authors‘ computations Table A10: Summary of Function Specifications Tobit on Log Poverty Gap Transformation Linear Tobit Two-part model in logs 1996 2005 Merged Data 1996 2005 Merged Data 1996 2005 Merged Data Pseudo R-Squared 0,0072 0,0179 0,0076 0,0215 0,0734 0,0314 0,058 0,251 0,0701 Log-likelihood -1346,78 -700,39 -2177,67 -453,88 -167,14 -651,54 -155,97 -25,77 -197,55 Observations 239 74 313 239 74 313 239 74 313 Notes: The value of the log-likelihood function when the model is linear is -2178. It can be observed that the value of the log-likelihood function increases at each new specification to a final value of -198. This latter value is higher than the value reported for the first model, which give some evidence that both the log-transformed and two-part model in logs give a better fit than the linear Tobit model. Sourceμ Author‘s estimations. 456 Table A11.1: Censored Data: Tobit Estimate & Marginal Effect, 1996 & 2005 Poverty Gap: pr(a, b) 1996 Variable DV Treatment I (*) Logarithm of Distance to District Road Distance to District Centre above/below 20 km (*) DV Ever Attended School or not (*) DV Bicycle Ownership (*) CV Age CV DV Age Squared Gender: Male or Female headed household (*) CV Landownership per household member (ha) DV Rainfall lagged (above/below avg) (*) CV Merge d Data 1996 2005 Merge d Data dy/dx (s.e.) X dy/dx (s.e.) X dy/dx (s.e.) X dy/dx (s.e.) X dy/dx (s.e.) X dy/dx (s.e.) X I II III IV III IV V VI VII VIII III IV Type DV Poverty Gap: e(a, b) 2005 -,03321 (,02021) ,0097134 (,05927) ,02409 (,06478) ,164494*** (,06323) ,0029287 (,01067) -,0000153 (,00011) -,0656014 (,07183) ,0514034 ,03586 ,0007498 ,01228 -,0342687 ,03256 ,0359614 ,03347 1,25 0,57 0,33 0,23 -,0389552 ,03214 ,0104983 ,00763 -,0001074 ,00008 -,0158172 ,04124 -,0257239** (,01259) ,0354661 (,0632) Margina l effects after tobit: dy/dx ,4392983*** (,03146) -,0150568 ,01752 ,0648412 (,05137) ,0206716 (,05399) 0,39 1,04 0,74 0,41 0,27 ,0034074 (,0518) ,0100451 (,00942) -,0000944 (,0001) ,0228324 (,05806) 1,54 -,1637599** (,06733) 0,57 ,011012 (,0399) 0,59 42,66 2067,10 0,535 0,09 1,20 0,61 0,35 -248.0749 (150.93) 72.47671 (441.74) 181.2966 491.11 0,24 1208.412*** (460.22) 21.87707 79.725 -.1141447 .82321 -476.5016 508.25 0,38 -,0394079*** (,01369) 0,72 ,0634506 (,05367) 0,57 40,86 1854,86 0,936 1,24862 0,56904 0,32636 6551.882 4563.4 88.56254 1449.9 -4629.494 4811.5 4453.11 4132.3 0,23013 -4760.223 3840 1240.081 850.88 -12.69056 9.41664 -1772.732 4368.2 - 1,27 -192.1546** (93.313) 0,60 263.9314 (468.65) 0,59 42,24 2016,93 0,581 0,39 1,04 0,74 0,41 18855.14*** (3089.1) -295.3517 (343.82) 1255.646 (983.01) 408.2051 (1073.3) 0,27 66.79867 (1014.9) 197.0437 (184.69) -1.85247 (1.92844) 453.4475 (1167.7) 1,54212 19343.74*** (5779) 0,38 -773.0198 (263.76) 0,56904 1256.199 (4391.1) 0,72 1230.001 (1028.7) 0,59414 42,6611 2067,1 6437,90 0,57 40,86 1854,86 31350,29 0,09 1,20 0,61 0,35 0,59 42,24 2016,93 0,24 1,27 0,60 16381,80 Table A11.2: Censored Data: Tobit Estimate & Marginal Effect, 1996 & 2005 Pove rty Ga p 1996 Va ria ble Type DV Treatment I CV DV Logarithm of Distance to District Road Distance to District Centre above/below 20 km DV Ever Attended School or not DV Bicycle Ownership CV Age CV DV Age Squared Gender: Male or Female headed household CV Landownership per household member (ha) DV Rainfall lagged (above/below avg) Constant Obse rva tions Log-like lihood Sigma R-Squa re d Pse udo R2 Tobit sca le Fa ctor: cdf e va lua te d a t zi 2005 OLS Tobit* Estima te Ma rgina l Effe ct I II III -462,7889 (417,7817) 75,96071 (1094,801) 965,6987 (747,8768) -647,5263 (394,48) 189,3678 (1155,422) 470,2412 (1265,393) 1368,343* (797,3156) -76,70118 (94,81407) ,7602623 (,9347434) 181,894 (1001,673) 3221,657** (1258,191) 57,10364 (208,1209) -,297941 (2,148791) -1276,77 (1398,933) -407,3515*** (106,2999) -501,563** (245,4533) -26,69316 (973,1264) 5546,243* (2987,671) Me rge d Da ta OLS Tobit* Estima te Ma rgina l Effe ct OLS Tobit* Estima te Ma rgina l Effe ct IV V VI VII VIII IX 27013,26*** (2607,787) -175,4543 (546,181) 2322,469 (1556,033) 780,4333 (1660,207) 31465,91*** (3861,213) -720,1154 (838,9257) 3093,55 (2445,53) 990,4495 (2591,391) -2098,061 -1438,187 (1569,468) 301,4767 288,2901 -3,194888 (3,01771) 1268,193 (1793,149) 162,9392 (2476,947) 480,4246 (450,3273) -4,516622 (4,700892) 1095,671 (2795,53) -22674,844 -579,152** (282,111) -1884,748*** (650,6379) 7695,577** (3315,017) 4,182597 (1267,773) -4473,495 (3539,113) 4109,972 4188,176 8289,656 (5684,958) 113,6646 (1860,841) -5802,606 (5898,336) 5665,165 (5216,264) -734,115 -5003,779 5471,063 1435,264** (571,9818) -14,9114** (5,851522) -1987,676 (4794,407) - -6069,701 (4863,628) 1591,569 (1089,234) -16,28756 (12,05694) -2297,151 (5715,119) -288,388 16056,23*** (3603,368) -372,314 108,882 270,379 1852,384 32,833 -0,171 691,3791 (1231,931) 397,528 -1827,863 (4949,898) -1050,982 239 -1346,792 7748,177 (556,4379) 0,0939 1974,032 (3727,72) 6665,847 (16358,07) -24826,51*** (7485,779) 7571,208 103,814 -5299,705 5174,176 -5543,652 1453,631 -14,876 1622,518 (5706,139) 1481,897 4916,375 (25453,77) 4490,282 74 -700,387 18906,24 (1790,774) 0,2652 0,007 0,575 457 2446,567 (1616,999) -257,3911 (6750,762) 16409,51 -375,54 1613,29 516,52 84,97 250,54 -2,36 571,39 -982,90 3027,869 (2554,399) 1579,04 -11456,64 (10629,62) -5974,65 313 -2151,047 18688,4 (1050,966) 0,3125 0,018 0,913 0,0197 0,522 Annex: Chapter 8 Annex 1a: Organisations surveyed in Chipata, Eastern Province, Zambia, 2005 Name of Organisation in Chipata Town Yielding Tree Milling Ltd Johabie and Braston Enterprise Hansim Agricultural Services MSP Farmers Shop CROP SERVE POTC Producer Owned Trading Co-Operative (Ltd) Mwanka Rural Development and Environmental Foundation ADRA Sheni Agricultural Suppliers Consolidated Farming Ltd Erisons Farmers House Shifa Trading Centre New life Investment Eastern Seed and Vet Nezi Investments Co. Ltd. Kondwelani Millers Katokoli Millers Willima Millers Lunkhwakwa Millers Mogra Tobacco (Agent for Zambia Leaf Co. Ltd.) Reuben Jere Sinoya Mwale Muthunzi Development Foundation SUPA Eastern Enterprises SK Store Modern Bazaar PAM M. Daka Fruit Shop Khulichi General Dealers Sohani Shopping Centre Chimwemwe Tonga Down Town Trading Centre Lukuzye Farms Lutheran World Federation (LWF) Listing Number T Main activity Interviewed 1 Provide animal stock feeds 13/09/2005 2 Agricultural trade / Marketing - supplies farmers with inputs of tobacco seed, fertliser and training and bu 13/09/2005 Agricultural Trade & Marketing; Suppliers of inputs; machinery and Marketing Services of Tobacco, 3 maize, G/nuts, soyabeans and training 13/09/2005 4 Supplies farm inputs, chemicals and tools 20/09/2005 6 Suppliers of inputs , seed, fertilisers, chemicals 19/09/2005 7 Suppliers of inputs and sells produce on behalf of member farmers 19/09/2005 11 Give rural communities support in terms of training, relief and sustainable projects 15/09/2005 12 A relief and Development Organisation involved in health, agricultural recovery problems etc. 15/09/2005 14 0 Deal in agricultural inputs, chemicals and machinery 14/09/2005 15 Supplies inputs, buys from farmers, transport agro products from point to point 21/09/2005 17 Buys from vendors and farmers: agricultural produce and suppliers inputs and chemicals and equipment 16/09/2005 18 Deals in farm produce, fertiliser and seed 20/09/2005 19 Sell fertiliser, seed, farm equipments, tools and farm produce 14/09/2005 21 Agricultural Inputs suppliers of seed, fertilisers, medicines and chemicals 13/09/2005 25 small scale maize mill 19/09/2005 26 Millers of maize and rice 15-Sep 27 Milling of maize and soyabeans 16/09/2005 28 Maize and Rice Milling 19/09/2005 29 Miller of mealie meal, soya oil, rice 14/09/2005 31 Buys tobacco from farmers on behalf of zamleaf. Also supplies inputs to these tobacco farmers 21/09/2005 Buying of agricultural products (tobacco, soyabeans, rice, cotton, beef cattle, maize, g/nuts and beans) 32 from farmers in rural areas and sell these products to big companies i Chipata urban. 13/09/2005 Agricultural marketing: buys farm produce (g/nuts; tobacco; soya beans, maize) from farmers in rural 33 areas and sell them to big companies Health and social work: primary helath care; food security; education; environmental management and 34 capacity building of local communities 35 Buys Sunflower from farmers and process it into cooking oil, buys maize, soyabeans 36 Suppliers of inputs such as fertilisers, seeds, chemicals and equipments 37 Agricultural Trading and Transport 38 06-221113 39 Buys fresh fruits from various suppliers and sell them in a shop 40 Suppliers of hammer mills, water pumps and oil expellers / extractors 41 Buying of produce and selling of farm inputs 42 Buys agricultural produce from rural areaas and sells them in town 43 General Retail and Input Supplier 44 support tobacco farmers and buys tobacco for export International NGO implementing rural community development. Rural community development motivation project in four districts with the aim of empowering people to get sustainable capacities to 46 meet their needs in dignity. Source: Authors. 458 15/09/2005 14/09/2005 20/09/2005 15/09/2005 14/09/2005 20/09/2005 16/09/2005 16/09/2005 16/09/2005 21/09/2005 17/09/2005 22/09/2005 Annex 1b: Organisations surveyed in Lundazi, Eastern Province, Zambia, 2005 Name of Organisations in Lundazi Town MSP Farmers Shop Aliboo Trading CROP Save Agri-COP PTC Mike Nyirongo Japhet Mkandawire Lutheran World Federation (LWF) YWCA Mthunzi Development Foundation Zambia Agricultural Small Scale Project Programme Against Malnutrition Central Church for African Presbutery (CCAP) Wildlife Conservation Society AFRICARE THANDIZANI ZAMLEAF STANCOM/DIMON - ALLIANCE ONE INTERNATIONAL DUNAVANT CLARK COTTON John Ngwenya Jumbo & Sons Investments Supernova Brewing Cooperation Sources: Authors. Listing Number 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 Interviewed yes (yes) yes yes yes Yes (yes) yes yes yes yes yes yes yes yes yes yes Web-links Directory of Development Organizations. Edition 2008. Volume I.B. / Africa. http://www.icnrd.org/africa/Zambia.PDF Zambia Yellow Pages: http://www.yellow.com/894.html Zambia Business Yellow Pages: http://www.thezambian.biz/ Zambia Official Yellow Pages: http://www.yellowpages.co.zm/ Zambia Business Web Directory: http://www.zambiz.co.zm/ , Zambia Business Links http://www.zambiz.co.zm/directory/businesslinks.htm Chipata Small Business Directory http://www.smallbusinessbigworld.com/afr/zambia/chipata/chipata.php Zambia Directory Services ILO BDS - Business Development Services: http://www.ilo.org/public/english/region/afpro/lusaka/tc/bds/index.htm At provincial level, MSE business directories (The Livingstone Directory and The Chipata Business Directory) have been established in partnership with commercial publishers. Community radio station: Breeze FM Chipata have begun radio programmes on HIV and AIDS targeting micro and small farmers and enterprises (MSEs). Cargill Cotton. http://www.cargillcotton.com/ Dunavant Enterprises. http://www.dunavant.com/ The Business Case for the Nacala Development Corridor. http://www.nacalacorridor.com/ http://www.nacalacorridor.com/pdf/the_regional_business_case.pdf PANA. Zambian insurance firm takes interest in growth triangle. Lusaka, Zambia. Shoprite. Available at http://www.shoprite.co.za/pages/127416071/About.asp 459 Zambia Country Commercial Guide FY 2004: Invest Climate. http://strategis.ic.gc.ca/epic/internet/inimr-ri.nsf/en/gr120406e.html 460 Annex 2a: Key Informants interviewed in August 2004 and July 2005 in Lusaka 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. Mr.James Nquluwe, District Marketing & Cooperative Officer, Ministry of Agriculture, 23rd of August 2004. Mr.Solomon Tembo, LCMS Unit, Central Statistical Office: 22nd July 2005. Mr.Klaus Droppleman, Agriculture Consultative Forum: 21st July 2005. Dr.Dick Siame, Smallholder Enterprise and marketing programme (SHEMP): 21 st July 2005. Dr.Jones Govereh & Ballard Zulu, Food Security Project & Michigan State University: 21st July 2005. Dr.Dennis Chiwele, Ruralnet: 21st July 2005. Professor Oliver Saasa, Institute of Economic and Social Research, UNZA: 22nd July 2005. Dr. Kalwinda, School of Agricultural Science, UNZA: 22nd July 2005. Mr.Mwanza, Lecturer project appraisal, Development Studies Department, UNZA: 22 nd July 2005. Mr.Alik Manda, Zambia Association of Geographic Information Systems, School of Mines: 25 th July 2005. Mr.Carl-Eric Hedstrom, Road Training School, ILO: 23rd of July 2005. Mr.Frans Blokhuis, Roads Department: 25th of July 2005. Mr.Alfred Takweya, former project manager of the EPFRP, and Mr. Chilufya, programme officer at the UNDP/UNCDF during the implementation of EPFRP, both Department of Decentralization: 22nd July 2005. Mr.Kabala, Chief Planner, policy and planning, Ministry of Agriculture, food and Fisheries: 25 th July 2005. Mr.Alex Mwanakasale, World Bank Office in Zambia: 28 th July 2005. Mr.Lewis Bangwe, Assistant FAO Representative: 28th July 2005. Mr.Masiliso Msooka, Agriculture & Environment, CSO and Mr.Ollings Chihana, Agriculture Statistical Division, CSO: 27th July 2005. Mr. G. Mbozi, Chief planner Marketing Coops: July 2005. Mr.Nkumbu Siame, Senior Engineer (Roads) National Road Board: September 2005. Mr. Alan Mulando, Assessing & Mapping unit, WFP, Zambia Country Office: 28 July 2005. Dr. Henrietta K. C. Kalinda, Consultant: 26 July 2005. Annex 2b: Key Informants interviewed in August 2004 & Sept. 2005 in Chipata 1. 2. 3. 4. 5. Mr. Gershom Jere, Crops Husbandry Officer, District Agriculture Co-ηrdinator‘s ηffice, εinistry of Agriculture and Fisheries, Chipata: 23rd August, 2004. African Development Bank Project Office in Chipata. Food Reserve Agency in Chipata. Programme Against Malnutrition Office in Chipata. World Food Programme Office in Chipata. Traders interviewed by Chiwele et al. in 1996 Number of traders Location Lundazi 16 Chipata North 9 Eastern-Province to Lusaka 10(28) Lusaka Urban 39 Annex 3. Traders Interviewed by Chiwele et al. in 1996 Questionnaire for traders used by Chiwele et al. 1996 Modules Number of questions Identification (metadata) Grain Purchases Prices Transportation Total number of questions 461 7 17 3 3 30 Table A.1: List of Market participants interviewed in Lundazi, 1996 Name of organisation Name of organisation Number Credit coordinators Mulla 1 DICE Enterprise 2 Kandanga 3 Crop financing and/or purchasing institutions Aliboo Trading Co. 1 Sable Transport 2 R.C. Chungu 3 Lonrho 4 LWF 5 Daud Mussa 6 Nyimba Super Market 7 Khalid Dalal 8 Clark Cotton 9 Number Crop Financing and/or purchasing institutions Sable Transport 1 Lonrho 2 Clark Cotton 3 Tobacco Development Company 4 Senegalia Farms and Fodya Investments 5 Tobacco Board of Zambia 6 Tobacco Growers Assocatian 7 Shiffa 8 Eastern Cooperative Union 9 Chipata District Cooperative Union Ltd 10 R.J. Angroindustries 11 Adam R. Macher 12 Nthaka Farmers Development Company 13 Ismail S. Ahmed 14 Quives Investments Ltd 15 Nezi Investments Company Ltd 16 Credit coordinators in Chipata Mosali 1 Mupa Investments 2 Mwobina 3 Mvuvye Development Company 4 Chalimbana Farmers Assocation 5 Source: Chiwele et al., 1998:95. 462 Table A2: Correspondence between GATS and RTAs PTA/COMESA (1981) COMESA (1994) ECOWAS ECOWAS* (1975) (1993) SADC (1992) SADC* (2001) EAC (1967) EAC* (2000) X X X X X X X X X X X X X X X X General provisions 1.1. Free movement of capital 1.2. Free movement of labour 1.3. National Treatment 1.4. Right of establishment 1.5. Policy Harmonization X X X X X X X X X X X X X X X X X X X X X X X X X X X X Direct provisions on services General Mandate Sectoral Coverage 2.1. Business Services 2.2. Communication Services X 2.3. Construction Services X X X X 2.4. Distribution Services X 2.5. Educational Services X 2.6. Environmental Services X 2.7. Financial Services 2.9. Tourism & Travel Services 2.12. Other Services not incl. X X X X X X X X X X X X X X X X X X X X X X X X 2.10. Cultural Services 2.11. Transport Services X X X 2.8. Health & Social Services X X X X X Source: Authors. 463 X X X X X X X X X X Table A3a: Zambia Doing Business Reports, 2004-2009 Starting a Business Year Ease of Doing Business Rank 2004 2005 2006 2007 2008 2009 .. .. .. .. 101 100 Dealing with Construction Permits Min. Cost (% of capital (% Procedures Time (days) income per of income (number) capita) per capita) Rank .. .. .. .. 84 71 6 6 6 6 6 6 35 35 35 35 33 18 33.2 31.7 30.3 29.9 30.5 28.6 Cost (% of Procedures Time (days) income per (number) capita) Rank 3.3 2.7 2.1 1.9 2.2 1.5 Employing Workers .. .. .. .. 149 146 .. .. 17 17 17 17 .. .. 254 254 254 254 Difficulty of Rigidity of Difficulty of Rigidity of Firing costs Hiring Hours Firing Employme (weeks of Index Index Index nt Index wages) Rank .. .. 1,638.4 1,342.8 1,518.0 1,023.1 .. .. .. .. 139 135 0 0 0 0 33 22 40 40 40 40 60 60 20 20 20 20 20 20 20 20 20 20 38 34 178 178 178 178 178 178 Source: http://www.doingbusiness.org/CustomQuery/ Table A3b: Zambia Doing Business Reports, 2004-2009 Registering Property Year Rank 2004 2005 2006 2007 2008 2009 .. .. .. .. 122 91 Procedures Time (days) (number) .. 6 6 6 6 6 .. 70 70 70 70 39 Getting Credit Cost (% of property value) Rank Legal Rights Index .. 9.7 9.6 9.6 9.6 6.6 .. .. .. .. 61 68 .. 9 9 9 9 9 Protecting Investors Public Private Credit registry bureau Informatio coverage (% coverage n Index adults) (% adults) .. .. .. 0 0.0 0.0 0 0.0 0.0 0 0.0 0.0 0 0.0 0.0 0 0.0 0.1 Rank Disclosure Index Director Liability Index .. .. 0 0 66 70 .. .. 3 3 3 3 .. .. 6 6 6 6 Shareholde Investor r Suits Protection Index Index .. .. 7 7 7 7 .. .. 5.3 5.3 5.3 5.3 Source: http://www.doingbusiness.org/CustomQuery/ Table A3c: Zambia Doing Business Reports, 2004-2009 Paying Taxes Year 2004 2005 2006 2007 2008 2009 Rank .. .. .. .. 33 38 Labor tax Payments Profit tax Other and Time (hours) (number) (%) contributio taxes (%) ns (%) .. .. .. .. .. .. .. .. .. .. 37 132 .. .. .. 37 132 .. .. .. 37 132 .. .. .. 37 132 1.7 10.4 4.0 Total tax rate (% profit) Rank .. .. 16.5 16.5 16.1 16.1 .. .. .. .. 155 153 Trading Across Borders Cost to Documents Time for export for import import (US$ per (number) (days) container) .. .. .. .. .. .. .. .. 53 2,098 9 64 53 2,098 9 64 53 2,098 9 64 53 2,664 9 64 Documents Time for for export export (number) (days) .. .. 8 8 6 6 Source: http://www.doingbusiness.org/CustomQuery/ 464 Enforcing Contracts Cost to import (US$ per container) .. .. 2,84 2,84 2,84 3,335 Rank .. .. .. .. 85 87 Closing a Business Procedures Cost (% of Time (days) (number) debt) 35 35 35 35 35 35 471 471 471 471 471 471 38.7 38.7 38.7 38.7 38.7 38.7 Rank Time (years) .. .. .. .. 87 80 2.7 2.7 2.7 2.7 2.7 2.7 Recovery Cost (% of rate (cents estate) on the dollar) 9 17.3 9 0.1 9 23.0 9 25.1 9 28.4 9 30.2 Figure A1: Zambia, 1996-2007 Aggregate Indicator: Regulatory Quality Source: Country Data Report for ZAMBIA, 1996-2007. Figure A2: ZAMBIA, 1996-2007 Aggregate Indicator: Control of Corruption Source: Country Data Report for ZAMBIA, 1996-2007. 465 .5 .6 .7 .8 .9 Figure A3a: Predicted Relocate values against „Decrease of Cost of Doing Business‟ 0 20 40 60 80 percentage decrease of cost of doing business 100 .6 .7 .8 .9 Figure A3b: Predicted Relocate values against „Increase in Output‟ 0 20 40 60 percentage increase in output 80 100 Table A3: Test of Model Specification Error Term Variable full_model1 full_model2 full_model3 full_model4 full_model5 full_model6 full_model7 full_model8 full_model9 full_model10 Stratum (S) 1.35 0.59 0.18 1.64 1.20 1.34 1.33 0.00 0.00 0.58 stratumb (Sb) 1.41 0.67 1.03 1.41 0.25 250498.75 1.26 0.00* 0.00* 0.58 . . mainacti (M) 1.13 1.15 0.29 1.13 0.27 1.19 0.00 District (D) 0.30** 0.28** 0.29** 0.30 0.19** 0.65 0.13* Interaction SxS Interaction SxM Interaction SxD Interaction SbxM Interaction SbxD Interaction MxD Constant _cons 1.51 5.13e+20*** 1.01e+17*** 4152.82 2.82 2444.64 0.00 0.11* 1.49 26.61 0.99 3.12 9.11e+21*** 9.03e+17*** 0.70 3.19e+17** 9.90e+18** 1.23e+19** 2.14e+17 5.94e+24** 1.18e+06 1.68 0.00*** 0.00*** 1.74 1.13e+30* 0.00*** 0.00*** 1.31e+32* N 41.00 41.00 41.00 41.00 41.00 41.00 41.00 41.00 41.00 41.00 ll -18.29 -17.68 -16.89 -18.29 -16.68 -18.09 -17.94 -12.31 -14.97 -17.31 df_m 4.00 5.00 5.00 5.00 5.00 5.00 5.00 8.00 7.00 6.00 chi2 8.97 10.20 11.77 8.97 12.19 9.38 9.68 20.94 15.62 10.94 Note: legend: * p<.1; ** p<.05; *** p<.01 Sourceμ Author‘s estimation. 466 Table A.4: ZMM-GT & LDC Indicators Country Malawi Mozambique Zambia LDCs Other developing countries Real GDP per capita (2006 dollars)* 1990 2006 110 164 198 349 944 938 322 454 1464 2580 Annual Average Growth Rates Population Annual Avg Growth (%) Rates (%) 1990-2000 2000-2006 1990-2000 2000-2006 4.9 1.2 2.0 2.6 2.7 5.6 3.1 2.4 -2.0 3.0 2.6 1.0 1.3 4.0 2.6 2.4 3.4 4.3 1.6 1.3 Percentage share of Net per capita food Road agriculture in production networks Total Labour Share of (Avg growth rates Density Force GDP p.a.) (km/1000km2) 1990 2004 1990 2006 1990-1996 2000-2006 2004 86.6 81.3 45.0 38.3 2.9 -4.4 164.2 83.4 80.3 37.1 21.5 -0.4 5.2 38.8 74.4 67.0 20.6 21.8 -0.1 -0.1 123.0 74.9 68.4 35.7 28.0 -1.1 0.8 n.a. n.a. n.a. n.a. n.a. Notes: * Real GDP data has been rebased using an implicit GDP deflator. Source: UNCTAD secretariat calculations. 467 n.a. n.a. n.a. Annex: Chapter 9 Table A1: Annual work plan for the Road Sector Investment Programme, Maintenance Funds for Feeder Roads, 2002 Total Funding Funding per Km Maximum no. of Maximum no. of Max no. Km per Type of Maintenance Anticipated Km (ZKW Billions) (ZKW millions) Km in Zambia Km per Province EPFRP District in EP Emergency 700 1.5 2.1 N/A N/A N/A Routine 4282 10.7 2.5 4282 475,78 833 59,472 Periodic Only partly available 55 25 2200 244,44 30,556 Allowance for For periodic maintenance only consultants 5.8 N/A N/A N/A N/A N/A Sourceμ Author‘s calculations based upon Rwampororo et al., 2002. Table A2: Road Fund Projects as of Nov 2003 Project Nameof Maintenance feeder roads in 8 districts Routine maintenance of trunk and district roads Nyimba - Petauke Katete Nyimba - Petauke Katete Feeder Roads Maintenance Contractor Agency Province 31 contracts MLGH Eastern Scope of Work Periodic unpaved 14 Vegetation contracts MoWS Eastern Control Sable Periodic Transport MWS Eastern Maintenance Ng'andu UWP MWS Eastern Supervision Road Type km Status Original Project Total Cost (ZMK) Expenditure Balance Year Feeder 484 On-going 5,984,535,409 2,642,901,292 3,305,634,117 2003 Trunk Road Trunk Road 35 Performance Feeder contractors MLGH Eastern maintenance Roads 1,561 On-going 1,412,336,637 435,659,241 976,677,396 2003 New Contract 19,611,525,148 0 19,611,525,148 2003 New Contract 1,604,481,250 0 1,604,481,250 2003 New Contract 6,242,969,579 0 6,242,969,579 2003 Source: National Road Fund Agency, 2010. Table A3: Disbursement of road fund for road maintenance works 1995 to 2005 Province 1995 - 8 1999 2000 2001 2002 2003 2004 2005 Eastern ZMK (million) ZMK (million) ZMK (million) ZMK (million) ZMK (million) ZMK (million) ZMK (million) ZMK (million) Provincial Rd Engineer 3,679,620 1,596,688 11,458,945 9,424,405 1,454.873 3,738.350 16,455.715 9,725.660 Chipata 820.640 140.020 219.472 2.670.634 848.404 566.577 2,366.185 1,765.232 Nyimba 170.000 0 0 455.616 429.919 362.416 497.995 593.433 Petauke 1.310.000 0 0 0 144.824 379.789 661.939 601.777 Lundazi 153.700 0 0 302.862 31.690 368.437 516.023 1,134.821 Katete 106.500 94.964 0 15.000 0 328.974 663.382 141.602 Chadiza 334.000 0 0 24.500 118.282 285.766 561.756 551.998 Chama 135.000 20.900 0 159.997 324.080 254.422 480.490 659.019 Mambwe 0 0 153.712 0 45.750 213.093 630.215 776.617 Consultancy Fees 15.860 0 0 0 0 0 0 0 Sub-Total 6,725.320 1,852.572 11,859.129 13,053.014 3,397.822 6,497.844 22,833.700 15.940.159 Sources: National Road Board (1995-2001) and National Road Fund Agency (2002-2005). 468 Table A4a: National Programme for Road Maintenance in Eastern Province from 1995 to Sept. 2007 Road Project Province Performance based maintenance of Lundazi - Chama Road D103 and Chipata - Luambe - Lundazi Road D104 for 4 Years: Package 5 Eastern Province Supervision of Performance based maintenance of Lundazi Chama Road D103 and Chipata - Luambe - Lundazi Road D104 for 4 Years: Package 5 Performance based maintenance of Lundazi - Chama Road D103 and Chipata - Luambe - Lundazi Road D104 for 4 Years Eastern Province Eastern Province Supervision of Performance based maintenance of Lundazi Chama Road D103 and Chipata - Luambe - Lundazi Road D104 for 4 Years Pothole patching of Chipata- Lundazi Road Eastern Province Eastern Province Maintenance of Nyimba, T4 - Utotwe Road 33,106,441,852.00 463.74 1,166,533,500.00 n.a. 33,106,441,852.00 463.74 1,166,533,500.00 2,000,000,000.00 Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Source: National Road Fund Agency, 2010. 469 5,275,487,679.16 131,362,803.43 12,528,144,926 116,653,350 172.00 250,000,000.00 Maintenance of Kazimule - Katete Road Supervision of road maintenance in the Eastern Province Periodic Maintenance of Nyimba- Katete Section of Great East Road Supervision of periodic maintenance of Great East Road from Nyimba to Katete Maintenance of Katete Dual Bridge on Katete - Mozambique Road T6 One-off routine maintenance: T4, Sinda - Msoro Turn-off One-off routine maintenance: T4, Luangwa Bridge - Kacholola One-off routine maintenance: T4, Mawanda - Petauke One-off routine maintenance: Lundazi - Mbeya One-off routine maintenance: Mbeya - Chama One-off routine maintenance: Katete - Chanida Border One-off routine maintenance: T4, Petauke - Sinda One-off routine maintenance: Chipata - Chadiza One-off routine maintenance: Mambwe - Mfuwe Road D791 One-off routine maintenance: T4, Msoro Turn-off - Mwami Border One-off routine maintenance: T4, Kacholola - Mawanda One-off routine maintenance: D104/D791, Chipata - Mambwe Supervision of BOQ Routin Maintenance of feeder and urban roads in Eastern Province Performance maintenance of feeder roads in Chipata District (Tamanda Loop) Maintenance of feeder roads in Chipata District (Chitandika Road Lot 1) Maintenance of feeder roads in Chipata District (Chiparamba Loop) Maintenance of feeder roads in Chipata District (Madzimoyo Eastern Dairy) Maintenance of feeder roads in Chipata District Chitandika Road Lot 2) Maintenance of feeder roads in Chipata District (Nzamane Kazimule Road) Maintenance of feeder roads in Nyimba District (T4 - Mambo (ZNS) School) Maintenance of feeder roads in Nyimba District (Nyimba Luembe Road Lot 1) Contract Sum Lenght Disbursement 2,000,000,000.00 151,994,096.00 410,414,904.00 45.20 160,762,090.00 325,636,811 n.a. 20,413,862.00 196,719,146 4.00 7,861,246,067.00 2,226,325,000 n.a. 861,275,000.00 394,988,000 n.a. 180,256,750.00 58,474,251 55.00 51,582,435.73 64,768,491 56.00 63,426,280.00 90,668,781 66.00 65,847,000.00 106,085,228 80.00 86,901,389.08 77,982,200 80.00 74,404,177.20 82,270,539 55.00 69,557,650.00 62,920,668 50.00 58,259,878.13 75,756,762 70.00 69,195,280.00 88,573,662 55.00 75,869,750.00 97,020,126 80.00 96,953,832.50 59,626,858 54.00 49,267,750.00 103,288,986 90.00 61,617,000.00 259,131,563 n.a. 46,848,424.50 182,329,098 20.00 64,097,190.18 196,110,896 25.00 110,098,470.25 280,683,249 15.60 103,947,614.62 242,242,832 26.10 148,932,113.74 162,806,402 25.00 86,560,670.80 239,545,567 24.60 11,867,925.35 116,780,631 10.40 100,250,755.60 304,009,959 19.00 189,799,345.64 Period Sources of funding JanuarySept 2007 EU JanuarySept 2007 JanuaryDec 2006 JanuaryDec 2006 JanuaryDec 2006 JanuaryDec 2006 JanuaryDec 2006 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 EU EU EU GRZ GRZ GRZ Table A4b: National Programme for Road Maintenance in Eastern Province from 1995 to Sept. 2007 Road Project Performance maintenance of feeer roads in Katete District (T6 - Mbinga Road) Maintenance of feeder roads in Nyimba District (Nyimba Luembe Road Lot 2) Performance maintenance of feeer roads in Petauke District (RD135, Sichilima - Mawanda) Performance maintenance of feeder roads in Petauke District (T4 - Chikalawa) Performance maintenance of feeder roads in Petauke District (T4 - Nyalukomba) Performance maintenance of feeder roads in Petauke District (T4 - Chitaika) Performance routine maintenance of feeder roads in Petauke District (Chikumbi - Kasonde) Performance routine maintenance of feeder roads in Petauke District (T4 - Merwe Mission School) Performance routine maintenance of feeder roads in Lundazi District Emusa - Munyukwa) Performance routine: maintenance of feeder roads in Lundazi District (Mphamba - Chitungulu) Perfromance routine maintenance of feeder roads in Lundazi District Performance routine: maintenance of feeder roads in Lundazi District (Tigone - Kapekesa) Performance routine: maintenance of feeder roads in Lundazi District (Hoya - Chiginya Road Lot 1) Performance routine: maintenance of feeder roads in Lundazi District (Hoya - Chiginya Road Lot 2) Maintenance of feeder roads in Katete District (T4 - Adoni) Performance routine: maintenance of feeder roads in Chadiza District (Mwami - Vubwi) Performance maintenance of feeder roads in Chadiza District (Chadiza - Tafelansoni) Performance maintenance of feeder roads in Chadiza District (Chadiza - T6 via Mlolo) Maintenance of Zozwe - Vubwi Rd Maintenance of Zozwe - Vubwi Rd Maintenance of Tembwe Khulamayembe Road Maintenance of feeder roads in Chama District (Chama Katangalika) Maintenance of feeder roads in Chama District (Chama - Sitwe) Performance routine maintenance of feeder roads in Chama District (Cham - Tembwe) Performance maintenance of feeder roads in Chama district (Chama - Sitwe) Maintenance of feeder roads in Mambwe district (Lugomo Road) Performance maintenance of feeder roads in Mambwe District (Masumba - Nsefu) Performance maintenance of feeder roads in Mambwe District (Msoro - Mambwe) Performance maintenance of feeder roads in Mambwe District (Nyamundela - Chikowa) Performance maintenance of feeder roads in Mambwe District (Kasamanda Road) Performance maintenance of feeder roads in Mambwe District Ncheka Road) Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Eastern Province Source: National Road Fund Agency, 2010. 470 Contract Sum Lenght Disbursement 174,700,597 17.50 77,869,736.84 303,050,278 20.00 204,668,554.71 209,864,259 11.00 117,873,631.24 183,839,048 25.00 116,183,573.93 156,872,761 17.50 93,796,425.67 167,106,084 22.00 40,429,114.75 62,458,770 4.5 33,313,658.75 108,733,372 6.00 70,427,121.80 190,540,350 24.00 102,511,547.40 288,055,339 20.00 118,220,765.38 677,825,775 9.30 513,780,880.78 276,719,856 16.00 228,445,597.38 100,914,264 8.00 41,240,739.73 140,502,505 20.00 130,621,052.50 293,076,482 19.50 13,084,239.06 268,858,095 32.50 80,955,573.00 149,265,293 20.00 94,649,192.20 152,912,444 13.00 108,980,780.00 167,092,783 17.00 100,982,760.63 140,497,903 17.00 90,415,803.50 427,621,557 20.20 137,727,444.40 175,911,189 9.10 62,311.54 178,215,070 26.00 25,664,127.96 173,677,878 25.00 128,074,485.38 201,373,012 30.00 339,871,068.41 238,212,920 2.00 97,293,196.00 118,378,783 9.40 34,707,582.87 155,047,008 20.00 50,644,074.50 204,478,412 28.00 167,397,565.51 69,889,353 4.00 49,395,237.50 197,973,776 18.00 131,155,756.00 Period JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 JanuarySept 2005 Fieldwork Documentation 471 Sourceμ Author‘s copyright. 472

RELATED PAPERS

RELATED TOPICS

Log In

An Enquiry into the Causes and Nature of the Transmission Mechanisms between Labour-Based Rural Roads, Sustainable Growth, and Agricultural Trade in Zambia‟s Eastern Province

An Enquiry into the Causes and Nature of the Transmission Mechanisms between Labour-Based Rural Roads, Sustainable Growth, and Agricultural Trade in Zambia‟s Eastern Province

Related Papers

RELATED PAPERS

RELATED TOPICS