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 = Af(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 ,
iN
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
jC 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
jC 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 b0ht 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 b0ht 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
jD 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 ˆ2 ( 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 xit 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 + Δt + 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