Public Disclosure Authorized Public Disclosure Authorized Public Disclosure Authorized Public Disclosure Authorized Poey Rmroha WORK.ING PAPERS SoleoconomioData InternationalEconomicsDepartment The World Bank April 1993 WPS 1124 A New Database on Human Capital Stock Sources, Methodology, and Results VikramNehru EricSwanson and AshutoshDubey A briefreviewof the methodanddatasourcesusedto preparethe IntemationalEconomicsDepartment'sestimatesof the stock of education.Analysis suggeststhat it is not unreasonableto use education stock as a proxy for human capital in production functionanalysis. P0 LcyResashWofingPapers disaefinatcetendilgs of wo*kinpogress andeawnogcth c=changefides am<g BaDk ff and allthictihdiond ThespapastnedbytheRschAdvisyStaffcamythenames of fthensrdled aiythcrviws,andshoddbeandsitedaccordingly.lefindinpsiont sandconnsarteautho'eown.Theysoul4 not be aunbted to the Woid Bank, its Doud of DireoB. its managan. or any of its manbercm ydibm POoly Ro"rh WPS 1124 This paperdescribesanew databaseonhumancapitalstockin developingandindustrialcountriesprepared by the IntemationalEconomicsDepartment(IEC)and undertakenas part of a largerIEC researchproject on total factorproductivitygrowth.Copiesof this paperare availablefree from the WorldBank, 1818H Street NW, Washington,DC 20433. Please contact Moira Coleridge-Taylor,room S8-049, extension 33704(April 1993,19 pages). Nehru,Swanson,and Dubeydescribethe techniquesand data adoptedfor the construction of a new series of estimatesof the stock of educationin 85 countriesover 28 years (196087). It covers all the importantdeveloping regionsexcept the republicsof the formerSoviet Union. IEC continuesa well-establishedtrend in growth researchof using educationalstock (measuredas meanschool years of educationof the labor force)as a proxy for humancapital. The series are built from enrollmentdata using the perpetualinventorymethod,adjustedfor mortality. Estimatesare corrected for grade repetition amongschool-goersand country-specificdropout rates for primaryand secondarystudents. Enrollmentdata series used start as far back as 1930for most countries,and even earlier for others.This reducesthe need for backward extrapolationof enrollmentsto providethe initial estimatesof the investmentinventory. ThePolicyResearchWorkingPapSeriesdisseniinates tfindingsofwokurider wayintheBank Anobjectiveofthes cries is to get these findingsout quickly, even if presentationsare less than fully polished.The fmdings, interpretations, and conclusionsin these papersdo not necessarilyrepresentofficialBankpolicy. Producedby the Policy ResearchDisseninationCenter A NewDatabaseon HumanCalital Stock: Sources.Methodolofv. and Results VikramNehni EricSwanson AshutoshDubey Abriefreviewofthe methodology anddatasourcesusedin thepreparationof IEC'seducationstockestimates,and an analysisof the results. A New Database on Human Capital Steckin DevelogingandIndustrial Countries: A Brief Commentara on the Methodoloevy Sources,and Results Tableof Contents Introduction I MeasuringEducationStock 2 DataSources 5 A BriefDescriptionoftheResults 8 Conclusion 13 Appendix 14 41- A New Databaseon HumanCapital Stockin Develoningand Industrial Countries: A Brief Commentaryon the Methodology. Sources,and Results Introduction 1. Increasingimportanceis beinggivento humancapitalin economictheoryas reflected.n the literatureon the contribution of humancapitalto economicgrowththat hs appearedin recentyears.1 This literature emphasizes theimportanceof humancapitalformationin thelongtermgrowthof output,especiallyin developing countries.In addition,it highlightstheinfluenceof nationaleconomicpolicieson longtermeconomicgrowth,a featureabsentin neoclassical models.According to thisvicw,changesin therateof humancapitalinvestment leadto changesin thelongtermrateofoutputgrowthratherthansimplyto changesin thelevelof output.The allocationofexpenditures betweenhumanand physicalinvestments, includinggovernment expenditures, therefore has criticalramifications forfuturegrowthand development. 2. Thequestfor a betterunderstanding of thedeterminants of growthhasstimulatedfreshinterestin improvingestimatesof humancapitalstock. Strictlyspeaking,themeasurement of humancapitalshouldcoverthe rangeof investmcnts that humanbeingsmakein themselves and in others,includingformalandinformal education,on-the-job-training, health,nutrition,and socialservices.Sofar, noresearchershaveput togethersuch a compositemeasure,althougheffortsareunderway.2 Instead,proxiesfor humancapitalusedin gSiowth research includesuchvariablesas enrollmentrates,adultliteracyrates,and healthindicators.Eachofthesefacesseveral conceptualand empiricaldrawbacks.To overcome them,thetrendhasbeento developeducationstockestimates basedon themeanschoolyearsof educationperworkingpersonin an economy.3 Whilethis measureis alsoonedimensionalin natureandsubjectto otherweaknesses, it hastheadvantagesof beinga stockmeasureand of requiringfor its construction datasetsthatarerelativelymorecompleteand extensive 3. Theseadvantageshavepromptedseveralresearcheffortsaimedat estimatingtheeducationstockof 4 coun'Aies usingdifferentdatasetsand techniques. Thispaperpresentstheresultsof onesucheffort.It describes thetechniquesand thedataadoptedfor theconstnctionof a newseriesofeducationstockestimatescovering85 countriesfor28 years. Theresearchis partofa projectto calculatetotalfactorproductivity growthfor a large numberofindustrialanddevelopingcountries.Theeducationstockestimatesproducedin this studywillbe used togetherwithphysicalcapitalstockestimates(forthcoming) to estimateproductionfunctionsthatcanbe usedto derivetotalfactorproductivity growthforthesecountries.It is intendedto putin placesystemsthat willupdate theseestimatesas and whenadditionaldatabecomeavailable. 4. Thebasicapproachto measuringhumancapitalinvestmentin this paperis similarto that ofLau, Jairison,and Louat(1991),Psacharopolous and Arriagada(1986,1992),and otherswhotakeyearsof schoolingas a proxymeasureof humancapital.Theseriesarebuiltfromenrollmentdatausingthe perpetualinventorymethod I Seebibliography. Forexample,Lavy,Victor(1991)proposesan aggregatemeasureof humancapitalinvestments basedon total nationalexpenditures oneducation,health,nutrition,and familyplanning. 2 3 Educationmaybe the mostimportantcomponent ofhumancapitalpartlybecauseit alsoincreasestheabilityof peopleto livehealthierlivesand learnmorerapidlyon-the-joboncetheyentetthe laborforce. TheparallelresearcheffortsarebyKyriacou(1992),Barroand Lee(1992),Psacharopoulos and Arriagada (1992),and Lau,Jamison,andLouat(1991). 4 -2- adjusted for mortality. Lau, BhaUia,and Loust (1991) madesimilar calculationsas backgroundfor the WDR 1991, but we correctfor estimatedrates of grade repetitionamongschool-goersand employcountry-specificdrop-cut rates at the primaryand secondarylevels. Accountingfor grade repetitionis particularlyimportantin developing countrieswhere enrollmentsmay otherwisebe overstatedby as much as 25 percent. In additionwe have located new sourcesof historicalschoolenrollments- as far bsck as 1930for many countriesand even earlier for others. This reducesthe needfor backwardextrapolationof enrollmentsto providethe initial estimatesof the investment "inventory." S. The followingsection describesthe perpetual inventorymethod and its applicationto the estimation of educationstock. A complete,computational"model' is presentedin order to show clearlywhere primarydata have been used and where, because of lack of data, estimates, averages, and simplifying assumptions havc been employed. 6. The section after that describesthe principal sourcesof data used to construct the educationstock series. The results presented in the final section include estimates of the average years of schooling by region. The detailedestimates for primary, secondary,and tertiary stages, by country, coveringthe period 1960 through 1987 will be publishedin electronicforrnat. Measuring Education Stocks 8. We followLau, Jamison,and Louat (1990),Psacharopolousand Arriagada(1986),and others in associatinghumancapital with the accumulatedyears of schoolingpresentin the workingage population.The stock of human capital is, therefore,built up from past "investments"in schooling.Unlikephysicalcapital, educationalinvestmentis not placedimmediatelyinto service. It enters the capital stockwhen its bearer enters the labor force and is withdrawnwhen he or she retires. Becausewe have no evidenceon the rate of obsolescenceof human capital - it is generallyassumedto be very long-livedcomparedto physicalcapital - we discount investment,prior to its plannedretirement,only by the rate of morality. 5 9. Psacharopoulosand Arriagadaprovide estimatesof the meanyears of schoolingin the labor force for 99 countriesusing census data. For a given country,the generic form of their measureof educationstock is given as n L= S, ,L i=) (1) where, 1i is the share of persons in the labor force with the ith levelof schooling;Si is the averagenumber of years of educationreceivedin the ith level of schooling;i designatesthe classificationsof illiterates(or no education), primaryincomplete,primary complete,secondaryincomplete,secondarycomplete,and tertiaryeducation. In this analysis,all levels of schoolingare weightedequally. 5 This assumptionbears a striking resemblanceto the "one-hossshay" assumptionoften assumedin the calculationof the grossphysical capital stock. But such an assumptionappears to ignore two importantfactors affectinghuman capital formationand decay. First, "learning-by-doing"can be an importantsourceof human capital acquisition. And second,the quality of educationtends to improvewith time, therebyleadingto a productivitydifferentialbetweenyoung and old workers. It is assumedin this analysisthat these two factorsare offsetting. -3- 10. The advantageof their methodis that in 66 countriestheyare able to obtain informationdirectlyon the educationalcharacteristicsof the current labor force. In the remaining33 they use informationon the educational attainmentof the populationby age and sex to estimate the schoolingprofileof the labor force. Furthermore, becausetheyare measuringeducationalattainment and labor force participadoncontemporaneously,no correction for expectedmortalityis required. 11. One problemwith census-basedmeasuresis that the true value of Si is not knownfor those who completedonly part of each schoolingstage; consequently,analystsare forced to makearbitrary estimates (Psacharopoulosand Arriagada(1986, 1992),Barro and Lee, (1993)).6 Becauserepeater ratesand drop-outrates tend to vary considerablyacross countries,educationstockestimatesbasedon census surveydata are subjectto measurementerror. Anotherproblemwith census-basedmeasuresis that theyare availableonly at discrete intervals. In only 34 countriesdo Psacharapoulosand Arriagadaobtain more than one year of data. Barro and Lee (1993) haveestimatedtime seriesfor 129 countriesusingsimilar census data, but in their case 77 countries havethree or more observations.Perhapsthe most seriousconcernwith the Barro and Lee e.timates is that they refer to the populationaged 25 and over. This can lead to a seriousdownwardbias in the estimatesof the educationstockbecausein most developingcountriesthe segmentof the populationbetweenthe age, of 15 and 25 is usuallylarge and growing over time. 12. It was notedearlier (para. 3) that the ultimateobjectiveof estimatingeducationstocks in developing countrieswas to derive total factor productivitygrowthestimatesfor these countries. And measuringthe productivityof investmentin educationover long periodsof time requiresan unbrokentime series of estimatesof the educationstock. Givensufficientlylong serieson enrollments,the perpetualinventorymethodcan be used to accumulatea continuousseries of estimatesof the stock of education.In this paper, the stock of educationis definedas the sum of person-schoolyears. Let Sgtbe the additionto our educationstock as a result of 1 year of educationin grade g in year t, then the cumuladveinvestmentin educationthat takes place in grades G = [g1 , g21 betweenthe years T = [tl, t2 ] is HGT =EZSgI (2) GT wherethe summationoperatorsact over the range of index sets G and T. 13. Note that SO is not necessarilya count of enrollmentsbut could insteadmeasure the "quality"of educationor human capital investment. For example,one might specifythat Sg, =qg,,Eg, (3) whereEgt are the enrollmentsin grade g in year t and q is a measureof the "quality' of the additionalyear of educationreceivedin grade g in yeart. LackingplausiBle,a priori measuresof quality (either betweengradesand years in a given countzy,or betweencountries),we measure Sgtas the total of net enrollments: S, =E; (4) 6 Barro and Lee (1993)have recentlyproducedquinquennialestimatesof averageyears of schooling based on census data supplementedby the perpetualinventoRymethod.Their estimates,however,like those of Psacharopoulosand Arriagada(1986),dependupon an arbitraryassumptionconcerningSi - that census respondentswho say theyhave attendeda particularstage of schoolhavecompletedit. In some censuses, respondentsanswer that they havepartially completeda particular stage of schooling. In such cases, Barro and Lee assumethat half the numberof years of schoolinghavebeen completedin that stage. Since Si tends not only to fluctuateover timnebut also differsconsiderablybetweencountries,these assumptionscan lead to an overestimationor underestimationof the level of the educationstock. -4- 14. The differeAlce between"gross"enrollmentsand "net" enrollmentsis the number of repeatersand dropouts in each grade 7: E* =E8r-Rgi-D.t (S) whereRgt and Dgt are measuresof the number of repeatersand dropoutsrespectivelyby grade and year. If we have repetitionand drop-Jut rates, equation(5) can be rewrittenas E,Eg (I -r, -dg,) (6) wherergt is the ratio of repeatersto total (gross) enrollmentsin grade g in yeart and dgt is the drop-outrate from grade g in ear t. We assume that repeatersacquirethe equivalentof one fulIyear of schoolingno matterhow long they 1S. spend in a grade. For simplicityin accounting,we attribute all of the "credit"for completingthe grade to the first year of enrollmentand deductall subsequentre-enrollmentsin the same grade. 16. The proper accountingof dropoutsis somewhatproblematic. A studentwhoattends part of a year should presumablybe creditedwith a partial enrollment,in whichcase dgt mightmeasure the averageportionof a school year attendedby dropouts. But it is open to questionwhethera dropouthas acquiredany nseful educationduring the year in which he or she leavesschool(Hartleyand Swanson(1988)),in which casedropouts shouldbe treated like repeatersand be fullynetted out of enrollmentsin the year in whichthey dropout. In many schoolsystems, however,dropoutsare not reporteduntil the beginningof the followingyear, at whichpoint they are no longer countedas enrolled.Becausepracticesare so unevenand reportingso imprecise,our procedureis to treat dropouts as if they had completedthe year in whichthey were last enrolled in cases whereenrollmentdata are availableby grade. In such instances,no adjustmentfor dropoutsis requiredin equations(5) or (6) to obtain nat enrollments8 This mayresult in a slight overestimateof total years of schooling, dropoutsare simplynon-enrollments. especiallyin schoolsystemscharacterizedby very high ratesof droppingout. 9 To determinewhen human capital is put into serviceand thereforerelevantto determiningaggregate 17. output, we needto knowwhen an individualenters or becomeseligibleto enter the labor force. In developed countriesthere are well establishedstatutorylimits on the age of entry into the labor force and, in many, the retrement age is also definedeither by statuteor custom. Praw;cesdiffer widelyin developingcountries (Ps2charapoulos and Arriagada,1991).In all countries,thereare systematic differenceswithrespectto thesex, education,and socialclassof the individual.To simplifymatters,wecountall personsbetweentheagesof 15and 64 inclusiveas constitutingthelaborforce. Foi thispurpose,weignorecyclicalcontractions and expansionsofthe "economically active"populationas wellas differences in theparticipationratesof differentsubpopulations. 7 In manyestimates,onlyrepeatersarenettedout ofgrossenroliments.However,givenour treatmentof dropouts in thefollowinganalysis,it is convenientto excludethemalsofromnetenrllments. Givendataon enollmentsand repetitionby jradeand assumingthattransfersintoandout of thesystem are negligible,thenumberofdropoutsin eachyearcanbe estimatedusingthe "gradetransition"or "reconstructed cohort"method. 8 However,forseveralcountriesand overdifferentperiods,enrollmentdatabygradeare notavailable.In suc.i cases,weusethe drop-outrate in thecalculationof netenrollments(seeequation10). 9 18. To calculatethetotalstockof educationcreatedin a particularstageof schoolingaad embodiedin the laborforcein a givenyearT we mustbeginwiththefirstyearin whichtheoldestcohortenrolledin thefirst grade and continuesummingE*gtbycohortthroughto the lastyearin whichtheyoungestcohortenteredthefinalgrade of thestage. Forthe moment,letus consideronlythe stageofprimaryeducation.In mostschoolsystems,the primatystageincludesthefirst sixgradesandtypicallychildrenenterschoolat agesix. In yearT, the oldest cohortin the laborforcebeganschoolin theyearT-64+6,whiletheyoungestcohortbeganin T-1S+6.Thetotal netenrollmentsof the50 cohortswhoenteredtheprimarystagebetweenT-58and Tr9arecalcutatedby T-9 6 SPT=£2ZEE.T-g-J (7) T-SCg") 19. Equation(7)givesthetotalnumberofyearsof schooling(aftercorrectingfordropiutsand repeaters)that wereacquiredbythepopulationwholivedandenrolledin schoolbetweentheyearsT-S8andT-9. Butweneedto "depreciate" thestockof educationby theexpectedlossesin eachyeardueto mortality.Assumingthat weknow theage-specific mortalityratesforpopulationin eachof theyearsT-58 4troughT-9, wecancalculatethe ofeachenrolleesurvivingunt.ltheyearT. Becauseagein schoolis closelyrelatedto grade,wecan probability assciate a probabilityofsurvivalto theyearT witheachenrolleein gradeg in theyeart. Letthis probabilitybe thentheexpectednumberof survivingenrollments embodiedin theworkforcein theyearT is givenby .= (8) T9 Z 6 T-S8g=1 whichprovidesthe measureof primaryeducationstockin yearT. ExpandingnetenrollmentsE, whichis a functionof thegrossenrollmentlevel,the retentionrate,and 20. thedrop-outrate(seeequation6), equation(8)canbe nowrewrittenas T-9 6 SPr = 2 Ees,.T-g-(I -rg.r-g-d)Eg.T-..-S( -dg.T--g.J) (9) T-58g=J 21. Assumingrt=rforall t anddg,t=dfor allgt, equation(9)canbe simplifiedto T-9 6 d FT = .- 9YZTS,-(I-r)Eg.r-gS( -d) (10) T-58g-I Thisis theequationusedto estimateprimaryeducationstock. Thesameapproachis usedto calculatesecondaryand tertiaryeducationstock. Theseresultsweresubsequently normalizedbythe wordcngagepopulationto obtainthemeanschoolyearsofeducadon Data Sources 22. Theeducationstockestimatesdiscussedin this paperarebasedon enrollmentdataacquiredfrom UNESCOsources.Educationsystemsvaryfromcountryto cowuntry, butUNESCOhasdrawnup a standard classification - the International StandardClassification of Education(ISCED)- and recommendations concerningstatisticalpresentation to ensurethatinternational statisticsareas comparableas possible.Primary(or firstlevel)education- ISCEDlevel1 - is definedas havingits mainfunctionas providingthebasicelementsof education.Secondaryeducation- ISCEDlevels2 and 3 - is baseduponprimaryeducationof at leastfouryears, andcanbe generalor specialized.Therefore,in additionto middleand highschools,secondaryeducationcanalso -6- cover vocationaland technicalcoursesand teachertraining of non-universitylevel. Tertiaryeducation- ISCED levels 5, 6, and 7 -- is definedas requiringa minimumconditionof admission,successfulcompletionof secondary education, or proofof equivalentqualif:cations(for example,from a university,teachers'college, or higher proissional school). 23. If educationstockestimatesbased on the perpetualinventorymethodare to start in 1960,and it is assumedthat the labor force comprisesall thosebetweenthe ages of 15 and 64, then enrollmenitdata seriesneed to begin in 1902. The only previoussystema' effort to developeducationstockestimatesfor developingcountries basedon the perpetualinventorymethodwas by Lau, Jamison,and Louat (1991). They used enrollmentdata fron. UNESCOsources,kept 1960as their first year for mostcountries,and createdthe seriesbetwcen 1902and 1960 using statistical methods. 24. One of the key departuresof this studywas to try and use actual data on enrollmentfor the years before 1960,to the extent that this was feasible. Fortunately,data sourceswere foundfor primaryand secondary educationthat allowedthe constructionof a grossenrollmentseries from 1902onwardfor 50 countriesand from 1930onwardfor 26 more countries. 10 Interpolationtechniqueswereused to fill gaps in the data, but the use of this approachwas kept to a minimum. Data from 1950onwardwere used for 6 countries,and from 1960onward for 4 countries - most of these being in Sub-SaharanAfrica. 25. In putting togetherthe longerseries,due care was takento accountfor national boundarychanges. For example,data on pre-independentIndia had to be dividedinto componentsfor Bangladeshand Pakistan,and pre1971 data for Pakistan had to be split into Westand East Pakistan,with the latter being added to the Bangladesh series. Data on East Africa,whereboundarychangesoccurredfollowingthe SecondWorldWar, already incorporatedthese adjustmentsin the original data sources,and thereforeno further adjustmentswere made to these series. 26. Wheregaps existedbetween 1902and the startingyear of the series,country-specfficgrowth rateswere used to extrapolatethe series. The referenceperiods for the calculationof these growth ratesusuallycoveredmore than a decade,and were chosen carefullyto avoid the inclusionof unusual conditions(suchas wars, suddenpolicy changes,etc.). In no case did the series reach zerowhen extrapolatedbackward. 11 27. The tertiary enrollmentseries were more difficultto put togetherand requiredconsiderablygreater use of interpolatedand extrapolatedestimatesas well as spliceddata seriesfrom differentdata sources. Tht anchoring serieswas obtainedfrom UNESCOdata sourceshousedin the Bank's Economicand Social Database(BESD). These were supplementedby various UNESCOyearbooks,Mitchell(1982), UNESCO(1958),UNESCO(1961), 10 For primaryand secondaryenrollmentdata, the principaldata sourceswere: * UNESCOEducational Statisticsavailablein the WorldBank Economicand Social Database(BESD); * UNESCO.1958.WorldSurvevof Education VolumeIII (NewYork: UnitedNations); • UNESCO.1961.WorldSurveyof Education VolumeIV (NewYork: UnitedNations); * M;tchell,B.R 1980.EuropeanHistoricalStatistics. 1750-1970.SecondEdition (NewYork: Facts on File); * Mitchell,B.R. 1982.InternationalHistoricalStatistics.Africaand Asia. SecondEdition (NewYork: New York UniversityPress); * Mitchell,B.R 1983.InternationalHistoricalStatistics.The Americasand Australasia. (Detroit:Gale ResearchCmpany); In the paper by Lau, Jamison and Louat (1991),the enrollmentseries reachedzero in a numberof cases. -7- and a varietyofnationalsources.12 In caseswheretheiiationalsourcesand UNESCOseriesdid not match (urtallyfordefinitionalreasons),thetwoweresplicedbyapplyingtheannualgrowthrateimplicitin thc national series. Again,dueaccountwastakenof changesto nationalboundaries. 28. Thegreateruseof "statistically manufactured" grossenrollmnent datain creatingthe tertiaryeducation stockseriecgivesthesedata.owerinformation contentand makesthemlessreliablein regressionestimates. Althoughthedataappearto behavein linewitha pr. rI reasoning,as subsequent sectionsof thispaperwillshow, theyshouldbe usedwithsomecaution. 29. Dataon repeaterratesbygradewereavailablein five-yearly intervalsbetween1960and 1985formost countriesand wererestrictedto primaryand secondaryeducation.Usingthesedataas benchmarks,annualdata serieswereconstructed bysimpleinterpolation for theyearsbetween1965ai.d 1985.It wasassumedthat forthe yearsbefore1965,the repeaterrateremainedat the 1965rate,andfor theyears&fler1985at the 1985rate. Finally,owingto thedearthofdataon enrollments bygrade,it wasnecessaryto constructa weightedaverage repeaterrate(theweightsbeingtheenrollmentsbygradein thefewyearswhensuchdatawereavailable).It shouldbe notedthat educationstockestimatesareparticularlysensitiveto repeaterratelevels- if repeaterrates wcreto havedoubledthroughouttheperiodfrom 1902to 1985,theeducationstockestimatewouldhavehalved, givingan elasticityof-0.5. Obtaininga longerand moreaccurateseriesof observations on repeaterratesis, 3 therefore,of someimportancein futureresearchworkon humancapitalstockestimation.1 30. Dataon theothertwovariablesusedin theconstruction ofthe netenrollmentseries- age-specific mortalityratesanddrop-outrates- wereevenmoredifficultto acquire.Dataonage-specificmortalityfatesare sparse,soseriesweredeveloped fora representative countryin eachregionand thenappliedto all thecountriesof that region.14 Whencomparedto errorsin the repeaterrateestimates,errorsin themortalityrateestimatesareof lessconsequence to thefinaleducationstockestimates- a doublingofthe mortalityrate,for instance,tendsto reducetheeducationstockbybetween2 and 3 percent,an elasticityof -0.02to -0.03. In the absenceof dataon drop-outlevels,drop-outrateswerecalculatedusingavailableinformationon grossenrollments,mortalityrates, and repeaterrates. 31. Apartfromshortcomings in themethodology and thedearthof dataon enrollments,mortalityrates, repeaterrates,anddrop-obtrates(discussedabove),thedataaresubjectto threeimportant,butwellknown, weaknesses that needto be keptin mind. First,thedatado not measurethequalityofeducation,and this makes intertemporl as wellas cross-country comparisons difficultto interpret.Unfortunately, no goodindicatorofthe qualityofeducationis availableeasilyfordeveloping countries.OinopopularmeasureofRen usedfor this purpose, theteacher-student ratio,doesnotappearto be stronglyrelatedto thevalueaddedof theschoolingsystem(see Barro,1991;Behrmanand Rosenzweig, 1992).1 Second,enrollmentdatasufferfromthe sameproblemas other developingcountrydata - theirreportingtendsto getmoreaccuratewithdevelopment, makingintertemporal 12 Detailscan beprovideduponrequest. 13 It shouldbe notedthatthe paperby Jamison,Lau,andLouat(1991)did not userepeaterrate ratedatato derive netenrollmentlevelsfromthegrossenrollmentseries. Giventhelargedifferences in repeaterratesacross countriesthis obviously leadsto somedifferences betweentheirdataseriesandours. 14 The representative countriesarethesameas theonesusedbyLau,Louat,and Bhalla(1991)- Egyptfor the MiddleEastand NorthAfrica,thePhilippines forEastAsia,SriLankafor SouthAsia,andBrazilfor mostLatin Americaneconomies.Thedatasourcefor mortalityrateswasvariousissuesof theUnitedNationsDemographic Yearbookbetween1945and 1987. Is Wewouldliketo explorefurthertheuseofdrop-outratesas a proxyfor educationquality. *8- comparisons subjectto error. In thecaseof theeducationstockseries,withitsdatabaseon enrollmentsstretchlr.g backto 1902,thisproblemcouldbe potentiallyserious.Andthird,yearsof schoolingas a proxyfor educationis subjectto orrorin cross-country analysisbecausethenumberofdaysand hoursof schoolingper yearcanvary byUNESCOattemptsto takeintoaccount substantially acrosscountries.Althoughthe ISCEDsystemdeveloped suchdifferencesin its datacompilation, somedifferences in nationaldefinitionsinevitablyremain. problems,however.Otherstudies This studyis not alonein havingto dealwithdataand methodological 32. usingalternativadatabasesand methodological approachesto estimatingeducationstockalsoconfrontmajor, perhapsprohibitive,shortcomings.Forexample,studiesthatestimateeducationstockon thebasisofUNESCO dataon thedistributionof thepopulation25 yearsof ageandolderby levelsof educational attainment(see,for and Arriagada(1986))haveseveralweaknesses.These example,Barroand Lee(1992);andPsacharopoulos problemsaredocumentedin BehrmanandRosenzweig (1992),whonotethat suchstudiesarebasedon a limited numberof nationalsurveysandcensusesconductedovera widevarianceofyearsand usinga widevarietyof definitions.Theseproblemsarecataloguedin detailednotesto thedatapresentedbyUNESCO.Forexample,in 14countriestheagerangesdifferfromthestandarddefinitions,in 5 countriestheanswer"notstated"is combined of smaller with"noschooling",and in 9 countriesilliteracyi6interpretedto be "noschooling".A wumber definitional differencesalsorendercomparisons difficult.Moreover,thisdataset suffersfromthestandard problemsof ignoringqualitydifferences acrosscountries,variationsin lengthof schooldaysand schoolyears,and theimportanceofnon-schooling education.Andfinally,censussurveysdo notreportthenumbe ofyearsof educationalinstitutions.The schoolingofindividuals,merelywhethertheyattendedprimary/secondary/tertiary highincidenceof repeatersand dropoutsapparentin enrollmentdatabut notcapturedin censussurveyssuggests that educationstockfiguresbasedon censussurveydataalonewouldtendto beoverestimated. 33. Sinceeducationstockestimatestendto be basedon sparsedataof unevenquality,testingthefinalresults wouldprovidesomeindicatioi.of theirqualityanc eliability.In thesectionthatfollows,thedatabaseon educationstockpresentedin thispaperis analyzed:rieflyand comparedto measuresofeducationstockprepared byotherresearchers. A BriefDescriptionofthe Results 34. Theeducationstockseriespresentedin this paperareavailablefor85 countriesfor theyears1960-87. TkJeaverageeducationstockmeasuresthe meanschoolyearsof educationoftheworkingagepopulation(defined average as thepopulationbetweentheages1Sand 64),andis the sumof primary,secondary,andpost-secondany regionarecoveredexceptforthe republicsof the educationstock. Alltheimportantcountriesin eachdeveloping formerSovietUnion. Thedataareparticularlyweakfor Sub-Saharan Africa,especiallyin the caseofpostsecondayeducation,but alsofor primaryandsecondaryeducation.Forthemostpart,datafor theother regionsarebasedon longerti ne serieson enrollmentratesandtendto havestrongerbackground developing documentation. countries,and between 35. A comparisonof theaverageeducationstockbetweenindustrialand developing differentdevelopingregions,providessomeinterestinginsights l'able1). Asonewouldexpect,themeanschool yearsof educationin developing countriesis lessthanhalfthatorindustrialcoumtries.Buttheoverallgrowthof averageeducationstockin industrialcountriesappearsto haveslowed,owinglargelyto thefactthattheprimary educationstockhas declinedmarginally.Themostrapidlyexpandingcomponentof theeducationcapitalstockin educaton. But,despitesuchgrowth,themeanschoolyearsof industrialcountrieshasbeenin post-secondary educational institutionsstandsat lessthan I.0. Similarly,themeanschoolyearsof educationin post-secondary to a potentialmaximumof 6.0),indicatingsignificantroom educationin secondaryschoolsis below3.0 (compared forfurtherexpansion. -9- 36. In the developingregions,the bulk of the averageeducationstockderivesfrom primaryschooleducation; the meanschoolyears of educationin secondaryand tertiaryschools,when taken together,are less than 1.0. At Table 1: Level and growth of average education stock In Industrial and developingcountries (schoolyears of educationper person betweenthe ages of 15 and 64) {-: ~ "::;______ Industrial:..". - .: . . : . Stockin 1987 Primary 'Secondary Tertiary -6.53 .. 2.60 0.88 .Developing 3.70' 0.72' East Asia' 4.38 0.72 South Asia 2.39 .0.88 Latin America '4.65 ' 0.S6' 'Sub-aliaran 2.33 0.19 ' 'De@. Euope -' :4.39. MENA'.'=':';''i-.'3j.24.", Wo02d'd: .38'. (a] .OLSgrowti rates.--..... . ' .0.88', ''''51.,13;-' 0.06 0.03 0.12 0.31 0.02 4.48 5.13 3.39 S.S2 2.54 3.2 3.9 2.9 1.5 3.9 6.0 9.2 4.3 5.3 9.7 5.3 3.4 6.4 6.7 12.6 4.0 4.2 3.3 2.0 4.2 0.23 5.50 4.79' 1.6 2.2 4.0 1.9 - 6.0 6.3 2.0 2.3 585 1.0 2.9 9 4.4 1;4 ,,0.41 1.17. Growth 1960-.81(%per year) [al Total Primary SecondarY TeTtiary Total .10.0 *0.5 2.2 4.9 0.3 0.29 the same time, the most rapidlyexpandingcomponentsof the averageeducationstockare the tertiaryand secondarycomponents. DevelopingEurope,Latin America,_nd East Asia have the highestaverageeducation stock amongthe developingregions,Sub-SaharanAfrica and SouthAsia the lowest. In general,the regionswhere the averageeducationstock is high, such as Latin Americaand Developing Europe, the growth of the stock has been Table 2: Correlation coefficientsof average education stock using low; and in regionswhere the average alternative estimation procedures stockis low, sucthas in Sub-Saharan ;___;________:___:__________..____:_ Africa,growth hasbeen rapid.Asia, .' EC '';R: ; PA : -. .: .K .B-L however,presents an exception. In JEC . 1.0 ... SouthAsia, growth in the average PA : 0.84 0. 1.0 ' '' ' * educationstockhasbeenrelativelylow ' '-L : :- .0.81 .... 0.92. 1.0 .. K 0.89 ... 86. 0.89 1.0 'Source: IEC InternationalEconomicsDepartment,The World ' K . Psacharapoulosand Arriagada(1991) f5 ' Barro'nd A"'BL: Liee(1993) : -................. .. WK:i Kyriacou(1991) despiteits level also being low, in East Asia, both the level and the growthrate are high. . 37. The EECdata were comparedto otherdaabaseson educationstbcksthat have beenprepared usingdifferent -10- techniques. The correlationcoefficientsbetween IECdata and the databasesof Psacharapoulosand Arriagada (1991),Barro and Lee (1992), and Kyriacou(1992)all exceed0.8 (Table2). 16 The correlationcoefficient between the data preparedby Barro and Lee and that preparedby Psacharopoulosand Arriagadahavea correlation coefficientof 0.92, reflectingthe use of similar data (censussurveys)and methodologies.Both the EECand the Barro-Leedata are correlatedto a similar degreewith the Kyriacoudatabase; this is significant,since the three use altogetherdifferenttechniquesfor estimatingeducationstocl.. 38. It was noted earlier that the educationstock data presentedin this paper werebased on an annual enrollmentseries that went as far back as 1930and in severalcases to 1902. The series that stoppedin 1930had to be extrapolatedbackwardto 1902,and this could have introduceda measurementbias. Prima facie, the countrieswith incompleteseries tendedto be low and middle incomecountriesand the countrieswith complete seriestended to be high income. The measurementbias, if any, could thereforebe expectedto be correlatedwith per capita GNP. 39. The IECdata were, therefore,comparedto the data seriesgeneratedby Psacharopoulosand Arriagada, and the differencewas regressedagainst per capita GNPas well as against time; this was donefor the entire sample of countriesand for the low and middle incomegroup only. In no case did the coefficientof the GNPper capita variableexceed0.0003, indicatingvirtuallyno associationbetweenthe differencesbetweenthe alternative data sets and per capita income(Table3). A similar exercisewas conductedby comparingthe TECand the BarroLee data set, and the findingswere identical. Table 3: Regressions estimates relating differences between average education stock data from alternative sources and per capita income D -ldependentarInbe variable Constant GNPper capita Time Adj. R7 Prob>F IDependent variable DifferencebetweenIEC data Barro-Leedata All countries 0.327** .0.0001*0 0.033 0.000 Allcountuies' -11.516 .0.0001*0.0060 0.036: - 0.002 OnlyLMICs 0.051 0.0002** 0.019 0.Q17 Only LMICs 30.230 0.0003+* -0.1530 0.022' 0.024 Psacharopoulosdata All countries: -0.083 -0.0(101** 0.047 0.055 All countries 113.371 *0.001 -0.0574 0.073 0.007. Only .MICs .0.411 0.0102 0.004 0.038 Oni; LMICs' 109.635 0.0',03* *0.0557 0.038 0.085 * Significantat the 95 percentconfidencelevel. * Significantat the 90 percentconfidencelevel. Source: Psacharopoulosand Arriagada(1986);Psacwaropoulosand Arriagada(1992); Barro and Lee (1992);authors' estimates. 16 The cofrelationbetweenthe Kyriaouand Psacharopoulosdatabasescould not be calculatedbecauseof an insufficientnumber of overlappingyears. -11- Simpletests were also conductedon the IEC educationstock data to check if thterewas any association 40. with a wide range of social, especiallyhealth, indicators. The resultsappear encouraging. The correlation coefficientbetweenaverageeducationstockand a varietyof social indicatorsof development,includingsuch indicatorsas the fertilityrate, birth rate, adult literacyrate, and infant mortalityrate seem to have the right signs and ordersof magnitudethat one wouldexpect (seeTable4 and AppendixTable 1). 17 For example,the overall correlationsacrosscountriesare high - all above0.8 - but there is considerabledifferentiationoncecountriesare dividedinto differentincomegroups.18 For low incomecountries,the correlationcoefficientsare often not as high as in the case of middle incomecountries,suggestingone of three possibilities: educationstandards are not as high in low incomecountries;other factorsassociatedwith low incomelevels tend to preventeducationfrom raising social indicators;or the data on averageeducationstock(as wellas the social indicatorswith which they are being compared)are inferiorin quality in low incomecountries. In the case of high incomecountries,the correlationcoefficientsare smalland, in some cases,of the wrong sign. This is not altogethersurprising; a clear and strong relationshipbetweeneducationand healthwould not be expectedto prevail in economiesat the higher end of the per capita incomespectrum. Table 4: Correlation coefricientsof average education stock and selected social indicators of development Fertilityrate Low.income . MiddlYe '.income'::-.0.63' Highu.com ::-0.07 Alli::''' '.' -"0.80 .'.......... ' Birth rate Adult rate Uteracy Infant mortality -- .-0.4 :-0.65 -.0.15 0.61 0.81 0.17 rate -0.68 -0.65 -0.16 :' -0.82 ..... 0.84 . .. -0.82 .:Source:'EC'andBESD.'. -: Table 5: Correlation coefficientsof fertility rates and average education stock by level of education Iounr. Primary: Secondary Tertiary Total -0.33 -0.33 0.06 *0.48 -0.63 -0.07 .0.65 -0.80 . group 431 0.:' '-0.73 Lowincome 'Middleincio'm'e .-0.63: '' ' High"'inc'o'me'''' ' '" 0.10 " '' : All ':-:: . ::: .:: .--O.40.72 0.2 -034 '-0.28 Source:TECand BESD 17 The correlationcoefficientswere calculatedusing cross sectiondata for one particularyear. For education stock,this year was 1987; for the other social indicators,the latestavailabledata was used from the World Bank's BESDdatabase. -12- Fipre 1: Partial scater of total educationstockand adult literacyrateaftercorrecting for per capita GNP Figure2: Partial scatter of total educationstockand fertilityrates after correcting for per capita GNP 40- 4 *e -50 5 tlota edclnstc .50 totaleducationstock educationstock 41. Whenaverageeducationstockis brokendownintoprimary,secondary, and tertiary(i.e.post-secondary), reinforcetheseinitialimpressions.Forexample,theassociation betweenaverage thecorrelationcoefficients educationstockand fertilityratesappearsto be relativelystrongat thesecondarylevelstageforlowincome appearsstrongestat theprimaryeducation countries;but in thecaseof middleincomecountries,theassociation stage(seeTable5). Thistendsto supporttheviewthatowingto thepoorqualityofeducationin lowincome countries,an additionalnumberofyearsof educationarerequiredto makea significantdentin fertilityrates. betweenaverageeducationstockandfertilityrates Amonghighincomecountries,as notedearlier,the relationship appearsweak. 42. Sincebothaverageeducationstockandothersocialindicatorswouldbe expectedto be associatedstrongly therelativelyhighcorrelationcoeficients withincomelevels(atleastamonglowandmiddleincomeeconomies), describedaboveshouldcomeas littlesurprise.To eliminatetheeffectofper capitaincome,bothaverageeducation stockas wellas adultliteracyratesand fertilityrateswereregressedagainstper capitaincome.19 Theorthogonal components fromtheseregressionswerethenplottedagainsteachother(seeFigures1 and2). Thedatashowsthat therelationshipbetweenaverageeducationstockand theadultliteracyrate,andbetweenaverageeducationstock andfertilityrates,is extremelystrongevenaftercorrectingfortheinfluenceof per capitaincome. Similarresults wereachievedwhenothersocialindicatorswereused. Is Theincomegroupsusedhere- low,middle,and high- usestandardBankdefinitionsas theyappearin World bank(1992b)and WorldBank(1992a). 19Theper capitaincomemeasureusedwasthe latestpercapitaGNPcalculatedaccordingto the WorldBank's Atlasmethod. -13- 43. In assessingthecomparisons describedabove,it is importantto recognizethat themeaures ofsocial development that wereusedto testtheaverageeducationstockestimatesare themselves subjectto considerable error. Forexample,theinfantmortalityrateestimatesaregenerallyofpoorquality(especially for lowincome countries):theyare oftenbasedon observations fromoldcensuses;thesampleusedin thesurveysareoftennot representative of theentirepopulation;and so-calledIndirectestimation techniques oftenmakeassumptions about pasttrendsthat maynotbe accurate.20 Givenconcernsabouttheaccuracyof the educationstockdataas wellas thesocialindicatorsdata,therelativelystrongassociation betweenthemfoundin thispaperprovidessomesource of comfort.Ofcourse,any ull fledgedanalysisof theinterrelationships betweeneducationandothersocial indicatorsof development wouldrequiremoredetailedeconometric workwhichis outsidethe scopeof thispaper. Conclusion 44. Thispaperpresentsa description of thedataand methodology usedin derivingannualeducationstocks for85 industrialand developing econoniesovertheperiod1965-87.Thedatacomparefavorablywithotherdata seriesmeasuringthe samevariablepreparedbyotherauthorsusingdifferentmethodsand datasources.Thepaper alsofindsa strikingassociationbetweenaverageeducationstockandotherindicatorsof socialdevelopment, especiallyoncetheeffectsof percapitaincomearecorrectedfor. Theweakpointsofthe datalie in the estimates of repeaterand drop-outrates,as wellas mortalityrates,whichcanbe improvedconsiderably if moreoriginaldata wereavailable.Nevertheless, the inclusionof suchvariablesin estimatingnetenrollmentlevelswasitselfan advanceoverpreviouseffortsat calculatingeducationstocks.In addition,the estimatesof educationstockare particularly uncertainforten countriesfor whichtheenrollmentseriesbeginafterthe SecondWorldWar. Here, again,theavailabilityofadditionaldatawouldsignificandy improvetheseestimates. 45. Theeducationstockdatapresentedin thispaperwerepreparedas partof a largerresearcheffortto estimatetotalfactorproductivity growthfora widerangeof countries.It wasintendedthat they(theeducation stockdata)wouldbe usedas proxiesfor humancapitalstock.Thehighcorrelationbetweenaverageeducation stockand otherindicatorsof humancapital- especiallya varietyof healthindicators- suggestthatthe useof sucha proxywouldbe a reasonablestep. A similarefforthasbeenlaunchedto estimatestocksofphysicalcapital in industrialanddevelopingcountries.Theresultsof thiseffortwillbe describedin a forthcoming paper. 20 SeeBos,E., M.Vu,andP. Stephens(1992). Appendix Table 1: CoriglationCoefficientsBetween Avera2eEducation Stock Estimats andSocialDevelonmentIndicators Fertility rate 1988 Birth rate 1988 Deathrate 1988 Health expend.as percent of GDP Dailycalone supply % of requirements Urban population ('%,1988) All countries Primary Secondary Post-secondary Total Low birth weight babies %, 1988 Adult literacy rate (%, 1985) -0.72 -0.72 -0.65 -0.80 -0.73 -0.77 -0.67 -0.82 -0.59 -0.30 -0.42 -0.54 -0.56 -0.53 -0.50 -0.61 0.59 0.74 0.59 0.71 0.60 0.71 0.63 0.71 0.68 0.61 0.68 0.73 0.83 0.64 0.66 0.84 0.50 0.73 0.55 0.66 -0.77 -0.69 -0.65 -0.82 Low income countries Primary Secondary Post-secondary Total -0.31 40.73 -0.33 -0.48 -0.35 -0.75 -0.34 -0.51 -0.56 -0.68 -0.43 -0.68 -0.19 0.19 -0.12 -0.12 0.33 -4.15 0.00 0.25 0.24 0.49 0.68 0.37 0.22 0.16 0.40 0.26 0.65 0.19 -0.06 0.61 0.16 0.35 0.86 0.33 -0.59 4.60 -0.28 -0.68 Middle income countries Prinmary Secondary Post-secondary Total -0..:i 40.34 -0.33 -0.63 -. -0.43 -0.35 -0.65 -0.50 .0.30 -0.31 -0.51 -0.10 40.40 0.03 -0.17 0.33 0.24 0.15 0.34 0.11 0.48 0.03 0.21 0.46 0.24 0.49 0.48 0.79 0.41 0.61 0.81 0.47 0.52 0.41 0.59 -0.61 -. 49 -0.34 -0.65 0.10 -0.28 0.06 -0.07 0.07 40.42 0.12 415 0.13 0.33 -0.15 0.23 -0.13 0.09 0.34 0.01 0.19 0.23 -0.03 0.24 0.49 0.08 0.30 0.40 -0.27 -0.28 -0.15 -0.34 0.01 0.31 0.06 0.17 -0.03 0.32 -0.08 0.13 0.00 -0.36 0.16 -0.16 High income countries Prnmary Secondary Post-seaonday Total Source:IEC;BESD Scientistsand Infant technicianper mrtality ra O ofpopulation per '0 live birth ApRendixTable2: Distribution of Countries by Their Rankings Ranking Numberof yearsfor which No.of Countries enrollment data extrapolated I 2 S0 26 6 4 3 4 None 28 (1902-1929) 48 (1902-1949) 58 (1902-1960) Thefollowingis the listof countriesby regionforwhichtheeducationalstockserieshasbeenconstructed. Thenumbersindicatetherankingof thecountries. AFRICA Angola Cameroon Coted'lvoire Ethiopia Ghana Kenya Liberia Madagascar Malawi Mali Mauritius Mozainbique Nigeria Rwanda Senegal SierraLeone Sudan Tanzania Uganda Zaire Zambia Zimbabwe RANK 2 2 1 3 I 2 4 3 2 4 1 2 I 4 4 2 2 2 3 2 2 I LATINAMERICA Argentina Bolivia Brazil Chile Colombia CostaRica Ecuador El Salvador Guatemala Honduras Haiti Jamaica Mexico Panama Peru Paraguay Uruguay Venezuela RANK I I 1 1 I 1 1 1 1 1 1 2 I 2 1 I 1 1 EASTASIA China HongKong Indonesia Japan Korea,Republicof Malaysia Philippines Singapore SOUTHASIA 3 1 1 2 I I 1 Bangladesh India Myanmar Pakistan SriLanka 1 1 1 1 I -16- Taiwan 1 Thailand 2 EUROPE, MIDDLE EAST, NORTHAFRICA RANK Algeria Egypt Groeoe Iran Iraq Israel Jordan morow" Portugal Spain Syrian, Afab Republic Tunisia 1 1 1 2 2 2 3 2 1 2 2 1 HIGH INCOME COUNTRIES Australia Austria Belgium Canada Cyprus Denmark Finland France Germany Ireland Italy Netherlands NewZealand Norway Sweden Switzerland UnitedKingdom United States Turkey RANK I 2 1 1 I I 1 I I 2 1 I 1 I 2 2 2 I 2 -17- AppendixTable3: ReferencePeriodUsedfor Estimationof Growth Ratesof Enrollments COUNTRY Angola Austria Cameroon China Ethiopia HongKong Iran Iraq Ireland Israel Ivorycoast Jamaica Jordan Kenya Korea Liberia Madagascar Malawi Mali Morocco Mozambique Panama Rwanda Senegal SierraLeone Spain Sudan Sweden Switzerland Syria Tanzania Thailand Turkey Uganda UnitedKingdom Zaire Zambia PRIMARY 1929-1950 1917-1950 1910-1950 1930-1950 1930-1950 1930-1950 1920-1950 1927-1950 1924-1950 1920-1950 1936-1950 1932-1950 1930-1950 1926-1950 1910-1950 1950-1970 1930-1950 Notneeded 1948-1970 1913-1950 1926-1950 1930-1950 1950-1970 1948-1970 1936-1950 1920-1950 1930-1950 1920-1950 1920-1950 1927-1950 1921-1950 1913-1950 1923-1950 1950-1970 1920-1950 1930-1950 1927-1950 SECONDARY 1930-1950 1918-1950 1910-1950 1949-1960 1948-1970 1950-1960 1920-1950 1927-1950 1924-1950 1920-1950 Notneeded 1950-1960 1952-1970 1926-1950 1912-1950 1950-1970 1945-1950 19451960 1948-1970 1914-1950 1920-1950 1934-1950 1950-1970 1950-1970 Notnecded 1920-1950 1930-1950 1920-1950 1920-1950 1927-1950 1920-1950 1913-1950 1923-1950 1930-1950 1930-1950 1930-1950 1926-1950 -18- Biblioeranhv Growth in a Cross-Sectionof Countries",in The QuarterlyJournalof Economcs. 1. Barro. R. 1991."Economnic May 1991,pp. 407449. 2. 3. Barro, R. and 3.Lee. 1993. InternationalComparisonsof EducationalAttainment paper presentedat the conferenceon "HowDo NationalPoliciesAffectLong-RunGrowth",WorldBank, WashingtonD.C. Becker,Gary. 1964.Human Capital(NewYork: ColumbiaUniversityPress). 4. Becker,G., K. Murphy,and R. Tamura._. "HumanCapital,Fertility,and EconomicGrowth"in Journal of PoliticalEconomy,Vol. 98, No.S,pt.2, S12-S37. S. Behrman,J.R. and M.R.Rozensweig.1992.The Qpalitvof AggregateInter-CountrvTime-SeriesData on Educationa_Investmentsand Stocks.EconomicallyActivePopulalions.and Emplovment,paper preparedfor the Conferenceon Data Base of DevelopmentAnalysis,Yale University,May 15-16, 1992. 6. Behrman,J. and A. Deolalikar.1987."WillDevelopingCountryNutritionImprovewith Income? A Case Studyfor Rural SouthIndia", in Journal of PoliticalEconomv,Vol. 95, No.3, pp. 492-507. 7. Bos,E., M. Vu, and P. Stephens. 1992. "Sourcesof WorldBank Estimatesof Current Mortality Estimates",Policyand RescarchWorkingPaper SeriesNo. 85I. Populationand HumanResources Department,WorldBank, WashingtonD.C. 8. Cochrane,S., D. O'Hara, and J. Leslie. 1980."Effectsof Educationon Health", WorddBank Staff Workina.No. 405, World Bank, WashingtonD.C. 9. Harley, M.J. and E. V. Swanson.1989. Retentionof Basic Skills AmongDropoutsfrom Envptian Primary Schools in Robert S. Mariano (ed.), Advancesin Stati- .cal Analysisand StatisticalComputing, JAI Press). (Loondon: 10. Kyriacou,G.A. 1991. Leveland GrowthEffectsof HumanCapital: A Cross-CountryStudyofthe ConverEenceHvpothesis,mimeo.,Departmentof Economics,NewYork University 11. Lau, L.J., S. Bhalla, and L.J.Louat. 1991."Humanand Physical CapitalStockin DevelopingCountriesConstructionof Data and Trends",draft mimeo.,WorldDevelopmentReport,The WorldBank. 12. Lau, L.J., D.T. Jamison,and F. Louat. 1991."Educationand Productivityin DevelopingCountries: An AggregateProductionFunctionApproach",Pf" WorkingPaper SeriesNo.612, WorldBank, Washington D.C. 13. Lavy, Victor. 1991. "HumanCapitalAccumulationand iiconomicGrowth: An EmpiricalEvaluation", mimeo.,Populationand HumanResourcesDivision,The WorldBank. 14. Psacharopoulos,G. and A.M. Arriagada.1986. "The EducationalCompositionof the Labor Force: An InternationalComparison"in IntemationalLaborAeview,Vol. 125, No.5, pp.561-574. 15. Psacharopoulos,G. and A.M. Arriagada.1992."The EducationalCompositionof the LaborForce: An Intemational Update",ER_EE BackgroundPaper SeriesNo. PHREE/92/49Educationand Employment Division,Populationand HumanResourcesDepartment,The WorldBank, WashingtonD.C. -19- 16. 17. World Bank. 1992a. GlobalEconomicProspectsand the DevelooingCountries. 1992(WashingtonD.C: World Bank). WorldBank. 1992.WorldDevelopmeptReBort(NewYork: OxfordUniversityPress). Policy Research Working Paper Series Titk* Author Date Contact for paper WPS1102DidtheDebtCrisisor Declining Oil AndrewM.Warner PricesCauseMexico'sInvestment Collapse? February1993 M. Divino 33739 WPS1103 CapitalMobilityin Developing PeterJ. Montiel Countries: SomeMeasurement Issues andEmpiricalEstimates February 1993 R.Vo 31047 WPS1104 TradePolicyReformin LatinAmerica AsadAlam andtheCaribbean in the 1980s SarathRajapatirana February1993 J. Troncoso 37826 WPS1105 Estimating Quasi-Fiscal Deficitsin Philippe Le Houerou a Consistency Framework: TheCase HectorSierra of Madagascar February1993 N. Velasco 34346 WPS1106 ImprovingWomen'sAccessto HalilDundar HigherEducation: A Reviewof World JennierHaworth BankProjectExperience February 1993 S. David 33752 WPS1 107 FinancialReformLessons and February1993 W. Pitayatonakarn 37664 February1993 A. Correa 38549 March1993 M.Raggambi 37664 WPS1110 WhatWouldHappenIfAll DevelopingWillMartin CointriesExpanded TheirManufactured Exports? March1993 D. Gustafson 33714 WPS1111ForeignInvestment Lawin Central andEastemEurope CherylW.Gray WilliamJarosz March1993 M.Berg 31450 WPS1112Privatization, Concentration, and Pressurefor Protection: A Steel SectorStudy YingQian RonaldC.Duncan March1993 .S. WPS1113 TheLuckyFewAmidstEconomic Christiaan Grootaert Decline:Distributional Changein RaviKanbur C6tedilvoireAsSeenThroughPanel DataSets,1985-88 March1993 E. Vitanov 38400 WPS1114 DoesPriceUncertainty Really ReducePrivateInvestment? A SmagModelApplied to Chile March1993 D. Bievenour 37899 GerardCaprio,Jr. kzakAtiyas JamesHanson Strategies WPS1 108 PublicOutputandPrivateDecisions:ThanosCatsambas Conceptual Issuesin the Evaluation of Govemment ActivitiesandTheir Implications for FiscalPolicy WPS1109 RiskManagement andStable FinancialStructures AndrewSheng YoonJe Cho AnitaGeorge JacquesMorisset Lipscomb 33718 Policy Research Working Paper Serles TRIt Author WPS1115Looking at the Facts:WhatWeKnow RossLevine aboutPolicyandGrowthfromCross- SaraZervos CountryAnalysis Date March1993 Contact for paper D. Evans 38526 WPS1116 Implications of Agricultural Trade Liberalization forthe Developing Countrles AntonloSalazarBrandao March1993 WillMartin D. Gustafson 33714 WPSI117 PortfolioInvestment Flowsto Emerging Markets Sudarshan Gooptu March1993 R.Vo 31047 WPS1118 TrendsInRetirement Systems and Lessonsfor Reform OliviaS. Mitchell March1993 ESP 33680 WPS1119TheNorthAmerican FreeTrade RaedSafadi Yeats Agreement: Its Effecton SouthAsia Alexander March1993 J. Jacobson 33710 WPS1120Policiesfor CopingwithPrice Uncertainty for MexicanMaize DonaldF.Larson March1993 D. Gustafson 33714 WPSI121 Measuring CapitalFlight:A Case Studyof Mexico HaraldEggerstedt March1993 ReoeccaBrideauHall *Sweder vanWijnbergen H.Abbey 80512 WPS1122 FiscalDecentralization In TransitionalRichardBird Economies: Towarda Systemic Christine Wallich Analysis March1993 S. Pacheco 37033 WPS1123SocialDevelopment is Economic Development April1993 S. Rothschild 37460 April1993 M.ColeridgeTaylor 33704 NancyBirdsall WPS1124 A NewDatabase on HumanCapital VikramNehru Stock:Sources,Methodology, and EricSwanson Results AshutoshDubey