Growth effects of education and social capital in the OECD countriesGrowth effects of education and

《Growth effects of education and social capital in the OECD countriesGrowth effects of education and》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

Growtheffectsofeducationand*socialcapitalintheOECDcountriesJonathanTempleDepartmentofEconomics,UniversityofBristol8WoodlandRoad,BristolBS81TN,UK19June2001AbstractThispapersurveystheempiricalliteratureonthegrowtheffectsofeducationandsocialcapital.Themainfocusisonthecross-countryevidencefortheOECDcountries,butthepaperalsobrieflyreviewsevidencefromlaboureconomics,toclarifywhereempiricalworkoneducationusingmacrodatamayberelativelyuseful.Itisarguedthatonbalance,therecentcross-countryevidencepointstoproductivitybenefitsofeducationthatareatleastaslargeasthoseidentifiedbylaboureconomists.Thepaperalsodiscussestheimplicationsofthisfinding.Finally,thepaperreviewstheemergingliteratureonthebenefitsofsocialcapital.Sincethisliteratureisstillinitsearlydays,policyconclusionsareaccordinglyhardertofind.*PreparedfortheOECDandforthcominginOECDEconomicStudies.TheviewsexpressedheredonotnecessarilyrepresentthoseoftheOECDoritsmembergovernments.IamgratefultoGavinCameron,DamonClark,MartineDurand,JørgenElmeskov,TomHealy,JohnMartin,MarkPearsonandDirkPilatforveryhelpfulcommentsonanearlierdraft.Anyerrorsaremyresponsibility.1 1.IntroductionPublicandprivateexpenditureoneducationalinstitutionsaccountsforjustover6%ofthecollectiveGDPoftheOECDmembercountries,orroughly$1550billioneach1year.Thisfigureunderstatesthetrueopportunitycostofeducationalinvestments,sinceitdoesnottakeintoaccountforgoneearnings.Overall,itshouldbeclearthattheprovisionofeducationrepresentsamajorcommitmentofresourceswithintheOECD,andsomeasuringtheassociatedwelfarebenefitsisanimportanttask.Oneaimofthissurveyistoexaminetheavailableevidenceonthebenefitsofeducationindevelopedcountries.Themainfocusisrestrictedtotheeffectsofeducationonlabourproductivity,atopicforwhichthereisaconsiderablebodyofevidence,admittedlyindirect.Iwilldrawonresearchfromtwofieldsinparticular:laboureconomics,andcross-countryempiricalworkoneconomicgrowth.Anunderlyingargumentwillbethat,althoughthelaboureconomicsliteraturedoesanimpressivejobofmeasuringtheprivatereturnstoeducation,itremainsthecasethatmacroeconomicstudieshaveacomplementaryroletoplay.Theemphasisthroughoutisverymuchoneducation,ratherthanonanybroaderconceptofhumancapital.Thechiefomissionisanyconsiderationoftraining.Thisdoesnotreflectmyviewofitsrelativesignificance,butratherthefocusofthepresentsurveyoncross-countryevidence.Thenatureoftrainingvariesconsiderablyacrosscountries,andinthemanufacturingsectoristightlyconnectedtoproductionstrategies(BroadberryandWagner1996).Itisdifficulttocapturethesedifferencesinwaysthatlendthemselvestoempiricalmodelling.Thismeansthat,inexplainingproductivitydifferencesacrossOECDcountries,thecross-countryevidencehaslittletosayabout2theroleoftraining,despiteitspotentialimportance.Thisisoneareainwhichanswersshouldbesoughtfromlaboureconomicsanddetailedcomparisonsofpracticesinindividualcountries,ratherthanfromthecross-countryempiricalworkreviewedhere.Asecondthemeofthesurveyistherelationbetweengrowthandwhathascometobeknownas‘socialcapital’.Itisdifficulttoarriveataprecisedefinitionofthisterm,andIwilldiscussthisissueinmoredetaillateron.Fornow,itcanbethoughtofascapturingsuchthingsastheextentoftrustworthiness,socialnorms,andparticipationinnetworksandassociations.Inthelastfewyears,someprominentacademicsandcommentatorshavearguedthatthesequalitiesofsocietiesarepotentiallyvaluablenotonlyinthemselves,butalsobecausetheymakeacontributiontoeconomicsuccess.Thisisanotherareainwhichcross-countryevidencemayhavesomethingworthwhile1TheexpenditureshareistakenfromOECD(2000a)andrelatesto1997.CollectiveGDPisbasedontotalGDPin2000fortwenty-nineofthethirtycurrentOECDmembers,atcurrentpricesandexchangerates,wheretheSlovakRepublicisthecountryexcluded.TheGDPfigureistakenfromthenationalaccountsstatisticsavailableonlineathttp://www.oecd.org/2OneconsequenceofthisomissionisthatIwillhavetoignoretheinteractionbetweeneducationandtraining.Totheextentthateducationisabout‘learninghowtolearn’,itmayhaveconsequencesforthevalueofsubsequenton-the-jobtraining.SomeinternationalcomparisonsoftrainingprogrammescanbefoundinOECD(1998,ch.3).TheworkofvanArkandPilat(1993)includesanexaminationoftheroleofvocationalskillsinexplainingproductivitydifferencesacrossGermany,JapanandtheUnitedStates.2 tocontribute,andlaterinthepaper,Iwillreviewthesmallbutgrowingliteratureonthecorrelationsbetweenmeasuresofsocialcapitalandeconomicperformance.Empiricalworkonsocialcapitalandgrowthisaveryrecentdevelopment,andwiththisinmind,Idevotethemajorityofthesurveytoresearchoneducationandgrowth.Section2providesthetheoreticalbackground,andshowsthatrecentmodelsprovidesomegoodreasonsforseeingeducationasacentraldeterminantofeconomicgrowth.Section3turnstotheempiricalevidence.Itstartswithabriefaccountofresearchinlaboureconomics,anessentialstepinunderstandingwherethecross-countryevidencemayberelativelyuseful.Therestofthesection,perhapstheheartofthesurvey,coversevidencefromgrowthaccountingandgrowthregressions,recentattemptstomeasureexternalitiestoeducation,andsomeworkonwiderbenefits.Thesecondpartofthepaperturnstosocialcapitalandgrowth.Section4discussesthedefinitionofsocialcapital,reviewsthemacroeconomicevidenceonitsgrowtheffects,andbrieflydiscussestheprospectsforfurtherresearchinthisarea.Section5roundsoffwithsomepossibleconclusions.2.Thegrowtheffectsofeducation:theoryTheaimofthissectionistoinvestigatewhetherformalmodelsshedanylightonthe3claimthateducationplaysacentralroleingrowth.Canthepossibleroleofeducationbegivenasecurefoundationintermsofeconomictheory?Howplausiblearethenecessaryassumptions?Dothemodelscapturethegrowtheffectsofeducation,asitisgenerallydefinedandunderstood,orofsomethingelse?OneofthemostprominentandinfluentialcontributionsisthatofLucas(1988),whichisinturnrelatedtopreviousworkbyUzawa(1965).Inthesemodels,thelevelofoutputisafunctionofthestockofhumancapital.Inthelongrun,sustainedgrowthisonlypossibleifhumancapitalcangrowwithoutbound.ThismakesitdifficulttointerprettheUzawa-Lucasconceptionofhumancapitalintermsofthevariablestraditionallyusedtomeasureeducationalattainment,suchasyearsofschooling.Theiruseoftheterm‘humancapital’seemsmorecloselyrelatedtoknowledge,ratherthantoskillsacquiredthrougheducation.OnewaytorelatetheUzawa-Lucasmodeltothedataistosuggestthatthequalityofeducationcouldbeincreasingovertime(BilsandKlenow2000).Inthisview,theknowledgeimpartedtoschoolchildrenintheyear2000issuperiortotheknowledgethatwouldhavebeenimpartedin1950or1900,andwillmakeagreaterdifferencetotheirproductivityinlateremployment.Evenifaverageeducationalattainmentisconstantovertime,thestockofhumancapitalcouldbeincreasinginawaythatdrives4risinglevelsofoutput.Yetthisargumentrunsintodifficulties,evenatthelevelofuniversityeducation.Theremaybesomedegreecoursesinwhichtheknowledgeimpartedcurrentlyhasagreatereffectonproductivitythanbefore(medicine,computerscience,perhapseconomics)but3AmoredetailedandrigoroussummarycanbefoundinAghionandHowitt(1998,ch.10).4Notethatthiseffectispotentiallyindependentofotherbenefitsofincreasedknowledge,suchasincreasesinthequalityofcapitalgoods,ormoregeneralformsoftechnicalprogress.3 thereareother,lessvocationalqualificationsforwhichthisargumentislessconvincing.Atthelevelofprimaryandsecondaryschooling,withtheirfocusonbasicskillssuchasliteracyandnumeracy,theideathatincreasesinthequalityofschoolingdrivesustainedgrowthseemsevenhardertosupport.Finally,notethatthesemodelsaretypicallysilentonexactlyhowtheincreaseinthequalityofschoolingisbroughtabout:individualscanraisethestockofhumancapital,orknowledge,simplybyallocatingsomeoftheirtimetoitsaccumulation.Analternativeclassofmodelsplacesmoreemphasisonmodellingtheincentivesthatfirmshavetogeneratenewideas.Endogenousgrowthmodelsbasedontheanalysisofresearchanddevelopment,notablythelandmarkcontributionofRomer(1990),yieldtheresultthatthesteady-stategrowthratepartlydependsonthelevelofhumancapital.Theunderlyingassumptionisthathumancapitalisakeyinputintheproductionofnewideas.IncontrastwiththeUzawa-Lucasframework,thisopensupthepossibilitythatevenaone-offincreaseinthestockofhumancapitalwillraisethegrowthrateindefinitely.Indeed,inmanyendogenousgrowthmodels,humancapitalmustbeaboveathresholdlevelforanyinnovationtotakeplaceatall.Inpractice,thegeneralityoftheseresults,andthecontrastwiththeUzawa-Lucasmodel,shouldnotbeoverdrawn.TheUzawa-LucasframeworkcanbeseenasamodelofknowledgeaccumulationinasimilarspirittothatofRomer,buteasiertoanalyse;andrestrictiveassumptionsareneededtoyieldtheRomerresultthatthelong-rungrowthratedependsonthelevelofhumancapital(Jones1995).Butevenundermoregeneralassumptions,ariseinthelevelofhumancapitalislikelytobeassociatedwithapotentiallysubstantialriseinthelevelofoutput,broughtaboutthroughatransitionalincreaseingrowthrates.Inmostendogenousgrowthmodelsbasedonresearchanddevelopment,thestockofhumancapitalistakentobeexogenouslydetermined.Morerecentpapers,notablyAcemoglu(1997)andRedding(1996),haverelaxedthisassumption,andconsideredwhathappenswhenindividualscanchoosetomakeinvestmentsineducationortraining,whilefirmsmakeinvestmentsinR&D.Forsomeparametervalues,multipleequilibriaarepossible,sincetheincentivesofworkerstoinvestinhumancapital,andthoseoffirmstoinvestinR&D,areinterdependent.Thisprovidesawayofformalizingearlierideasaboutthepossibleexistenceofa“low-skill,low-qualitytrap”inwhichlowskilllevelsandslowratesofinnovationreflectacoordinationfailure(FinegoldandSoskice,1988).Themodelssuggestthat,attheaggregatelevel,greaterinvestmentsineducationortrainingmightraiseexpenditureonR&D,andviceversa.Anotherinterestingaspectofrecentgrowthmodelsistheirsuggestionthatindividualsmayunder-investineducation.RustichiniandSchmitz(1991)examinethisargumentinsomedetail.Theypresentamodelinwhichindividualsdividetheirtimebetweenproduction,originalresearch,andtheacquisitionofknowledge.Eachindividualknowsthatacquiringknowledge(througheducation)willraisetheirproductivityinsubsequentresearch,butsincetheydonotfullycapturethebenefitsofresearch,theywilltendtospendtoolittletimeacquiringknowledgerelativetothesociallyoptimal4 5outcome.RustichiniandSchmitzcalibrateasimplemodel,andfindthatalthoughpolicyinterventionhasonlysmalleffectsontheallocationoftimetoeducation,itcan6haveasubstantialeffectonthegrowthrate.Morerecently,Romer(2000)haspointedoutthatmodelsofgrowthdrivenbyR&Dshouldpotentiallyinformeducationpolicy.Henotesthat,inthemodelsreviewedabove,growthisdeterminedbythequantityofinputsusedinR&D,notsimplyexpenditureuponit.OnereasonthispointmattersisthatincentivestoencourageR&D,suchastaxcredits,maybeineffectiveunlesstheyencourageagreaternumberofscientistsandengineerstoworktowardsdevelopingnewideas.Toillustratethis,consideraverysimplemodel,inwhichafixedsupplyofscientistsonlyworkinR&Dandaretheonlyinputtotheresearchprocess.ThenanincreaseinR&Dspendingwillsimplyraisethewagesofscientists,withnoeffectonthenumberofresearchersengagedinR&D,orthegrowthrate.Inamoregeneralandrealisticmodel,therewillbesomeeffectofgreaterR&D7spendingontotalresearchinputsandthereforegrowth.Tocreatealargeeffect,higherwagesforscientistsshouldencouragemoreindividualstotrainasscientists.Thisrequiressomeflexibilityonthepartoftheeducationsystem,andintheprovisionofrelevantinformationtopotentialstudents.SotheeffectivenessofdirectsubsidiesortaxcreditsforR&Dmaybeenhancedbycomplementaryeducationpolicies,aimedatimprovingorsubsidizingthesupplyofresearchinputs,ratherthansimplythedemandforthem.Insummary,themodelsofthenewgrowththeoryareimportantforseveralreasons.First,theyseehumancapitalasanimportantinputinthecreationofnewideas,andthismechanismprovidesarelativelyappealingjustificationforviewingeducationasacentraldeterminantofgrowthrates,evenoverlongtimeintervals.Second,theysometimesyieldtheresultthatthelaissez-faireoutcomedeliversslowergrowththanissociallyoptimal.Third,themodelssuggestthatpolicy-makerswishingtoraisethelevelofoutputhaveseveraloptions:notjustdirectsupportforR&D-whichmaybedifficulttoimplementandmonitor-butalsosubsidiestocertainkindsofeducation,perhapsespeciallythosewhichcouldleadtolaterworkinresearchanddevelopment.3.Thegrowtheffectsofeducation:evidenceAswehaveseen,theoreticalmodelsimplythat,insearchingforthedeterminantsofgrowth,policyoneducationisoneofthefirstplacestolook.Inthissection,Iwillturntotheattemptsofeconomiststoquantifyeducation’simportance.Themainfocuswillbeonthemacroeconomicevidence:thebodyofresearchwhichmeasures,orattempts5Theassumptionthatitisdifficulttofullycapturethebenefitsofresearchisuncontroversial.Thepresenceofsubstantialresearchspilloversisintuitivelyplausible,andsupportedbyempiricalevidence.Griliches(1992)providesasurvey.6Acompletewelfareanalysisofpolicyinterventionwouldneedtoconsidertheeffectsontheleveloftheoutputpath,aswellasitsgrowthrate.7Forexample,someindividualsoutsidetheR&Dsector,butoriginallytrainedasscientists,mayswitchintoR&Dcareersinresponsetohigherwages.5 tomeasure,theproductivitybenefitsofeducationusingthevariationineducational8attainmentandgrowthratesacrosscountries.Itwouldbeamistake,however,toreviewthisevidencewithoutfirstdiscussingtheworkoneducationandearningsbylaboureconomists.Ifeducationaffectsproductivitydirectly,thistendstoimplyanobservablerelationshipbetweenanindividual’seducationandtheirearnings.Theevidencefortheseeffectsisthebestestablishedintheliterature,andanunderstandingofitsstrengthsandweaknesseshelpsplacethecross-countryevidenceincontext.Thiswillclarifytheareasinwhichthemacroapproachmayhavesomethingworthwhiletocontribute,andalsopointtotheareasinwhichmicroevidenceismorelikelytobefruitful.Withthisinmind,section3.1reviewsstudiesoftheeffectsofeducationbasedonearningssurveys.Latersectionscovergrowthaccounting(3.2),theevidencefromcross-countryregressions(3.3),recentworkonexternalitiestohumancapital(3.4)andthewiderbenefitsofeducation(3.5).Finally,section3.6attemptstotietogetherthevariouspiecesofevidence.Thereviewpointsoutthateachapproachtomeasuringtheproductivityeffectsofeducationhasitsownimportantweaknessesandareasofuncertainty.Yettakentogether,thevariousmethodstendtoagreeinpointingtoquitesubstantialeffects.Asaresult,itwouldbedifficulttousetheavailableevidencetoconstructacasethateducationiscurrentlyover-providedintheOECDasawhole,andperhapsevenharderifoneacknowledgedthewiderbenefitsdiscussedinsection3.5below.Broadlyspeaking,thisworkmightalsojustifyanexpansionofeducationalprovisioninsomecountries,especiallythosewherecurrentpoliciesimplyrelativelylowlevelsofattainmentinfutureyears.Afullanalysisofpolicyquestions,however,wouldneedbothtoacknowledgethepotentialimportanceoftraining,andtoinvestigatehowagivenquantityofeducationalspendingisbestallocated;thesetopicsarebeyondthe9scopeofthepresentreview.Beforeturningtothevariousstrandsofevidenceinmoredetail,itmaybehelpfultoclarifytheconceptsofproductivitythatthedifferentapproacheshaveinmind.Atthelevelofindividuals,outputperworkerhourseemsthemostrelevantmeasureofproductivity,notleastbecauseonebenefitofanincreaseinhourlyproductivitymaybethatindividualschoosetoworkfewerhours.Inexaminingproductivitydifferencesacrosscountries,however,therearesometimesdisadvantagesinusingoutputperworkerhourasthebasisforcomparison.Thismeasureofproductivityisaffectedbylabourforceparticipationrates,andotheraspectsoflabourmarketinstitutions.FurtherdiscussionandsomerecentevidencecanbefoundinScarpettaetal.(2000).Itisalsoworthpointingoutthat,forsomepurposes,policy-makersareinterestedinoutputperworkerandoutputperhead,aswellasoutputperworkerhour.Education8SianesiandVanReenen(2000)alsoprovideareviewofthemacroeconomicliteratureoneducationandgrowth,withextradetailonindividualpapers.Scarpettaetal.(2000)analysetherecentgrowthperformanceofOECDmembercountriesinmoregeneralterms.9AninnovativepaperbyJudson(1998)investigateswhethereducationalspendingisallocatedefficiently.Itseemslikelythatfutureresearchwillgiveincreasingemphasistothistopic.6 mayalsohaveindirecteffectsonthesevariables,notsimplythroughhourlyproductivity.Forexample,educationisoftenthoughttoaffectlabourforce10participation,particularlythatofwomen.Itmayalsoaffectthenon-monetarybenefitsassociatedwithworkandleisure,andsoaffectworkinghours.Sincecross-countryempiricalworkistypicallybasedonoutputpercapitaoroutputperworker,itwilltendtoconflatetheseeffectswiththedirectimpactofeducationonlabourproductivitythatlaboureconomistshavesoughttoquantify.3.1EvidencefromlaboureconomicsThissectionreviewsevidencefromlaboureconomics.Ratherthanattempttoprovideasummaryofavastempiricalliterature,theemphasiswillbeonhowtheconventionalfindingsshouldbeinterpreted,andtowhatextentwecaninfergenuineeffectsof11educationonproductivity.Researchersinthisfieldtypicallystudythelinkbetweeneducationandproductivityusingsurveydataontheearningsandcharacteristicsoflargenumbersofindividuals.Thetechniquesusedtoanalysethesedatahavebecomeincreasinglysophisticated,andwewillseethatevidencefrom‘naturalexperiments’providesmeasuresoftheprivatereturntoeducationthatareprobablyquiteaccurate.Thereismuchgreaterdisagreementontheextenttowhichlaboureconomistshaveidentifiedthesocialreturn12toeducation.Forexample,educationalqualificationsmaybevaluedinthelabourmarketbecausetheyactasasignalofability.Asaresult,privatereturnstoschoolingmaybehighevenifeducationhasnoeffectonproductivity.Thisargumentwillbediscussedfurtherbelow.Inanalysingtheprivatereturn,thestandardempiricalapproachistoexplainthevariationinearningsacrossindividualsusingregressions,wheretheexplanatoryvariablesincludeyearsofschooling,eitherageorasimpleproxyforlabourmarketexperience,andothercharacteristics.ThemostpopularspecificationdrawsheavilyontheworkofMincer(1974),andearliercontributionson‘humancapitalearningsfunctions’.Thestartingpointistypicallyaspecificationthatlookssomethinglikethis:2lnw=a+bS+bE+bE(1)012whichrelatesthenaturallogarithmofwages(w)toyearsofschooling(S)andaproxyforlabourmarketexperience(E).Undersomeassumptions,andgiventhesemi-logarithmicformulation,thecoefficientonschoolingcanbeinterpretedastheprivatereturntoeducation.Empiricalestimatesoftheprivatereturntypicallyhavearelativelysmallstandarderrorandliesomewherebetween5%and15%,dependingonthetime10Theinteractionbetweengrowth,humancapitalandfemalelabourforceparticipationisdiscussedinmoredetailbyMincer(1996).ForevidenceonfemalelabourforceparticipationintheOECD,seeOECD(1998,ch.4).11Card(1999)andHarmonetal.(2000)provideexcellentanddetailedreviewsofthevariousissues.AnotherusefulsurveyisthatofAshenfelteretal.(1999),whichincludesadetailedinvestigationofproblemsassociatedwithpublicationbias.12Differentauthorsusetheterm‘socialreturn’indifferentways.HereIwilluseitdenotetheoverallreturnonanindividual’seducationfromsociety’spointofview,ratherthan(say)theprivatereturnsimplyadjustedfortaxationanddirectcostsofeducation.7 andcountry.Ifworkersarepaidtheirmarginalproduct,theseeducationalwagedifferentialsmayalsotellussomethingusefulabouttheeffectofeducationonproductivity.Theevidencethatearningsarepositivelyassociatedwithschoolingisrobustanduncontroversial;theobviousdifficultyliesingivingthisassociationacausalinterpretation.Oneofthemosteasilyunderstoodproblemsisthat,throughlackofsuitabledata,theregressionsinevitablyomitsomeimportantvariablesthatarelikelytobecorrelatedwithbothschoolingandearnings.Familybackgroundandtraitssuchasinnateabilityordeterminationarenotableexamples.Thebasicproblem,fromtheeconometrician’spointofview,isthatthegroupofpeoplewitharelativelyadvancedlevelofeducationalattainmentisnotarandomselectionfromthepopulationasawhole.Forexample,ifmoreableindividualshaverelativelyhighearningsregardlessofextraeducation,andalsochoosetospendmoretimeinschool,thentheestimatedreturntoschoolingoverstatestheeffectofeducationonproductivity.Ifabilityisnotobservedbyemployers,thentheregressionestimatemaystillcapturetheprivatereturntoschooling,butitwillnotcapturethesocialreturnthatisultimatelyourmaininterest.Unfortunatelytheproblemsdonotstopthere.Itseemsprobablethatthecostsandbenefitsofeducationvaryacrossindividuals,perhapssubstantially.Indeed,thisislikelytobetheprincipalcauseofthevariationincompletedschoolingthattheeconometricianusestoidentifytheeffectsofeducation.Theheterogeneitywilltypicallymeanthattheprivatereturnstoeducationvaryacrossindividuals.Intheunlikelycasewherethereturnsvaryindependentlyoftheexplanatoryvariables,theregressionsshouldstillrecoveranunbiasedestimateoftheaveragereturn.Moregenerally,however,theheterogeneityproblemwillleadtobiasedestimates.Therecentfocusoftheliteratureoneducationhasbeenonidentifyingnaturalexperiments,inthehopethatthesewillallowstrongerclaimsaboutcausalitytobemade.Researcherslookforsituationsinwhichthelevelofschoolingvariesacrossindividualsforreasonsthatarelikelytobeindependentoftheunobservedcharacteristicsofthoseindividuals(ability,determination,andsoon).Theideaisbestexplainedbymeansofanexample.Agoodstartingpointisoneofthemostinfluentialpapers,byAngristandKrueger(1991).Thepaperstartsfromtheobservationthat,whenitiscompulsorytostayinschooluntilacertainage,individualsbornearlierinthecalendaryearwillreachthelegalminimumageforschool-leaversatanearlierstageintheireducation.Asadirectresult,thereislikelytobeacorrelationbetweenanindividual’squarterofbirthandtheirlengthofschooling.Thecorrelationmeansthatquarterofbirthcanpotentiallybeusedtoidentifyexogenousvariationinschooling-thatis,variationindependentofunobservedcharacteristicslikeabilityordetermination.Ineconometricterms,quarterofbirthcanbeusedasaninstrumentforschooling,underthemaintainedassumptionthatpersonalcharacteristicsotherthanschoolingareindependentofquarterofbirth.Somewhatsurprisingly,AngristandKruegerfindthattheinstrumentalvariableestimatesofthereturntoschoolingare8 similartotheleastsquaresestimates,supportingtheideathatconventionalestimates13arereasonablyaccurate.Anothermuch-discussednaturalexperimentisprovidedbyidenticaltwinswhohavedifferentlevelsofschooling.Giventhatsuchtwinshavethesamegenes,andwillusuallysharethesamefamilybackground,thewagedifferentialbetweentwinswithdifferentyearsofschoolingmayprovideusefulinformationontheproductivityeffectofeducation.Finally,othernaturalexperimentsareprovidedbythepossibleconnectionbetweenthegeographicalproximityofcollegestoindividuals,andtheirchoiceofschooling(seeCard1999).Researchofthiskindhasconsiderablystrengthenedthecaseforproductivityeffectsofeducation,buteventhesestudiesretainanimportantweakness.Ithaslongbeenunderstoodthattheprivatereturntoeducationmaybeapoorguidetothesocialreturn.ThetheoreticalworkofSpence(1973)indicatedthateducationalattainmentmaybevaluedbyemployersmainlybecauseitactsasasignalofinnateability,andnotbecauseithasaneffectonproductivity.Modelsofsignallingstartfromtheobservationthatindividualshavetraitswhichemployersvaluebutdonotfullyobserveatthetimeofhiring(ability,determination,andsoon).Ifthereisasystematicassociationbetweenthesetraitsandthecostsandbenefitsofeducation,thismayleadtoanequilibriuminwhichhigh-abilityindividualsstayinschoolforlongerbecausethisdecisionsignalstheirabilitytoemployers.Thisargumentprovidesaplausiblereasonforacorrelationbetweenabilityandyearsofschooling,andsuggeststhatearningsmaybecorrelatedwithschoolingevenifschoolinghasnoeffectonproductivity.Fewdoubtthatsignallingplayssomeroleinexplainingeducationalwagedifferentials,butitsoverallimportanceremainscontroversial.Weiss(1995)andQuiggin(1999)provideverydifferentperspectivesonthetheoreticalgeneralityandempiricalvalidityofsignallingmodels.Therearetwomainargumentsagainstsuchmodels,whichnotetheimplicationsoftheassumptionthateducationhasnoeffectonproductivity.First,giventhewagepremiumearnedbythosewithmoreyearsofschooling,employerswouldprobablyhavestrongincentivestoconducttheirowntestsofabilityandothercharacteristics,andusethisdirectinformationratherthanthesomewhatindirectsignalprovidedbytheschoolingdecision.Thisviewissupportedbyevidencethatmeasuredperformanceinschoolanduniversitiesiscorrelatedquitestronglywiththeoutcomesoftestscarriedoutatanearlierstage(seeQuiggin1999forreferences).Yettheargumentisnotconclusive,mainlybecauseemployersmaynotbeabletoappropriatethereturnstoacquiringmoreinformationabouttheiremployees;otherfirmscouldbidawaythoseworkersfoundtohavehigherability(Stiglitz1975).Thesecondargumentisthat,ifeducationdoesnotaffectproductivity,onewouldexpecttoseetheeducationalwagedifferentialdeclinewithjobtenure,asemployersacquiredirectknowledgeofthecharacteristicsoftheiremployees.Thisdoesnotseem13Spaceprecludesafulldiscussionoftheinterpretationoftheseresults,andtheirrobustness.Foramoredetailedreview,seeHarmonetal.(2000).9 tobeobservedinthedata,althoughthisquestionhasnotreceivedthesustainedattentionitprobablydeserves.Moregenerally,thereisclearlyroomtodevelopandtestsignallingargumentsinmoredetail.Thisisimportantnotleastbecause,asWeiss(1995)haspointedout,eventheresultsofnaturalexperimentsarenotnecessarilyinconsistentwiththesignallingviewofeducation.Toseethis,recallthatemployersmayuseyearsofschoolingtogaininformationaboutcharacteristicsthatarenotobservedatthetimeofhiring.TheresultsfromtheAngristandKruegerquarter-of-birthstudyandtheworkontwinscaneasilybeinterpretedintermsofthesesignallingeffects,andsoonecouldstilldefendeventheextremeviewthatproductivityisentirelyindependentofeducation.Fornow,letusassumethatemployersfullyobserveallrelevantcharacteristics,andhencedonotinferanyinformationaboutthemfromschoolingdecisions.Eveninthiscase,asCardnotes,notmuchispresentlyknownaboutthemechanismsbywhicheducationmightcontributetohigherwages.Thesimplestinterpretationoftheevidencefromearningsfunctionsisthatmoreeducatedindividualsaremoreproductive,whatevertheirchosenoccupation.Inpractice,acollegedegreeisunlikelytomakeoneanoticeablybetterpostmanorroadsweeper.Education’srolemaybetoequipworkersforthetaskofworkingwithmoreadvancedtechnologies,forprovidingahigherqualityofservice,orfor‘learningbydoing’inthecourseofemployment.Understandingthemechanismscouldbeimportant,andwillhaveimplicationsfortheinterpretationofearningsfunctions.Forinstance,moreeducatedworkersmayhavebetteraccesstothosejobsinwhichworkerssharesomeoftherentsearnedbyimperfectlycompetitivefirms.Ifmechanismslikethisareatwork,therewouldagainbelessreasontobelievethattheobservedcorrelationbetweenschoolingandearningsrepresentssolelyadirectproductivityeffect.Thereareotherwaysinwhichprivateandsocialreturnscoulddiffer.Insomecountries,especiallypoorerones,thepublicsectorisamajoremployerofthewell-educated.AsPritchett(1996)emphasises,theassumptionthatwagedifferentialsreflectdifferencesinmarginalproductsismuchhardertosustaininthiscontext.Ifeducationalcredentialsareusedasameansofdeterminingaccesstorationedhigh-payingjobsinthepublicsector,estimatedearningsfunctionsmaydetectaneffectofeducationevenwhenithaslittleornoeffectonproductivity.Thegeneralproblemisthatestimatesofearningsfunctionscapture,atbest,theprivatereturntoeducation,yetitisthesocialreturnwhichisofmostinteresttopolicy-makers.Thetwomaydivergeforanumberofreasons,includingthepossibilitythateducationactsmainlyasasignallingdevice.Theargumentsdiscussedaboveimplythatthesocialreturntoeducationislessthantheprivatereturn,andaswehaveseen,evenjustalowerboundonthesocialreturnsisdifficulttoestablish.Therearealsosomereasonstobelievethatthesocialreturntoeducationcouldexceedtheprivatereturn.Itisplausiblethatindividualsdonotfullycapturesomeofthebenefitstosocietyoftheirschooling,andIwillreviewsomeoftheempiricalevidenceonexternalitiesandwiderbenefitsinsections3.4and3.5below.Anotherimportantargumentisthateducationalprovisionmayplayavaluableroleinallowingamore10 efficientmatchingbetweenworkersandjobsthanwouldotherwisebepossible(Arrow1973,Stiglitz1975).Inotherwords,evenifeducationdoesactmainlyasasignal,thereshouldnotbeapresumptionthateducationisthereforesociallywasteful.Insummary,thereisaningeniousandpersuasivebodyofworkwhichsupportstheviewthatprivatereturnstoschoolingarequitehigh.Card(1999)concludesthattheaveragemarginalreturntoeducationisunlikelytobefarbelowthestandardregressionestimates.Theviewthatthisprivatereturnoriginatesinagenuineproductivityeffectisfarfromuniversallyaccepted,however.AsWeiss(1995)hasargued,eventhemostrecentresultscanbeinterpretedastheoutcomeofsignallingeffects.Thissuggeststwolinesofenquirythatmightbeparticularlyfruitful.Thefirstisfurthertheoreticalexamination(andperhapscalibration)ofsignallingmodels,withaparticularfocusontheextenttowhichtheycanincorporatethedirectproductivityeffectsenvisagedinthetraditionaltheoryofhumancapital.Second,moreevidenceontheextenttowhicheducationalwagedifferentialsevolvewithjobtenurecouldbeofgreatinterestinadvancingthedebate.3.2GrowthaccountingAswehaveseen,thelaboureconomicsliteratureprovidesawealthofevidenceontheprivatereturnstoschooling.Itisnecessarilysilent,however,onthecontributionofeducationrelativetoothersourcesofaggregategrowth.Makingassumptionssimilartothoseoflaboureconomists,researchersinthegrowthaccountingtraditionhavesetaboutthecomplextaskofevaluatingtheoverallgrowthcontributionofchangingeducationalattainment.Thissectionwilldescribethemethodandreviewtheavailableevidence.Growthaccountingessentiallydividesoutputgrowthintoacomponentthatcanbeexplainedbyinputgrowth,anda‘residual’whichcapturesefficiencychange,partly14reflectingchangesintechnology.Inexplainingthechangeinoutput,thechangeinthequantityofeachinputisweightedbyitsmarginalproduct,proxiedbyitsmarketreward.Thisprinciplecanbeextendedtoanynumberofinputs,andwheresufficientlydetaileddataareavailable,itispossibletodisaggregatethelabourforceintovariouscategories,whereeachtypeofworkerisweightedbytheaveragewageofthattype.Forinstance,inanalysingthecontributionofchangesineducationalattainment,theresearcherdisaggregatesthelabourforcebylevelofschooling,andoftenbyotheravailablecharacteristicssuchasageandgender.Changesinthenumberofemployeesateachlevelofschoolingarethenweightedbytheirmarginalproducts,proxiedbythemeanincomeassociatedwitheachschoolinglevel,togivetheoverallchangeinanindexof‘effective’orquality-adjustedlabour.Thisultimatelyallowstheresearchertoquantifytheproportionofoutputgrowththatcanbedirectlyattributedtoincreasesineducationalattainment.14Notethat,dependingontheapproachadopted,someformsoftechnicalchangewillbetreatedaschangesinthequalityofcapitalinputs,andwillnotappearintheresidual.11 Griliches(1997)providesabriefbutusefulsurveyofthisliterature,andpointsoutthetwomajorassumptions,bothofwhichwillhaveafamiliarringtoreadersoftheprevioussection.First,itisassumedthatdifferencesinobservedmarketrewardscorrespondreasonablycloselytodifferencesinmarginalproducts.Secondly,thecalculationsassumethatdifferencesinmarketrewardsacrossschoolinglevelsoriginateinschooling,andnotinotherfactorssuchasnativeabilityorfamilybackgroundthatmaybecorrelatedwithschooling.Theadvantageofthefirstassumption,thatmarketrewardscorrespondtomarginalproducts,isthatitallowsthegrowthaccountanttoobtaintheory-consistentweightsusingtheavailabledata,atleastundertheassumptionsofconstantreturnstoscaleandperfectcompetition.Lessrestrictiveframeworksarepossible,butwillgenerallytendtorequireadditionalandperhapscontroversialassumptionsaboutparameters.Itshouldalsobeclearthatconventionalgrowthaccountingwillnotshedanylightonthepossiblecontributionofexternalities.Thisisamajorlimitation,andanimportantmotivationforthecross-countryempiricalstudiesthatwillbeconsideredfurtherbelow.Whatofthesecondassumption,thatdifferencesinwagesoriginateinschooling?Thedangerherecanbeseenfromconsideringanextremescenario,inwhicheducationhasabsolutelynoeffectonanindividual’sproductivity,butmoreableindividualsbothspendlongerinschoolandearnmorewhileinemployment.Thisscenarioclearlyimpliesthateducationalattainmentandearningsarepositivelycorrelated.Nowconsideranexogenousincreaseintheproportionofindividualswiththehighestlevelofeducation:sincetheindexoflabourqualityweightsthenumbersineacheducationclassbythemeanincomeofthatclass,theindexmustincrease.Asaresult,thegrowthaccountantwillattributesomeportionofgrowthtoeducationalimprovement,eventhougheducationplaysnoroleinproductivitygains.Thisimpliesthat,asinthelaboureconomicsliterature,growthaccountingcangiveussomeinsightintotheproductivitycontributionofeducation,buttheanswersarebynomeanscompleteorconclusive.Beforedescribingtheresultsofgrowthaccountingexercisesinmoredetail,itisimportanttoclarifytheconnectionbetweenchangesineducationalprovisionandthemeasuredeffects.Anexpansioninprovisiontypicallyaffectsonlyyoungercohorts,andsohaslong-livedeffectsoneducationalattainmentinthelabourforceasawhole.Averageattainmentwillcontinuetoincreaseforsometimeasolder,lesseducatedcohortsretirefromemploymentandarereplacedbythemorehighlyqualified.Whenusinggrowthaccountingmethods,itistheselong-livedeffectsthatarequantified,andoneshouldbearthisinmindwheninterpretingspecificfindings.Thepracticalimplicationisthatresultsforrecentyearsaredrivenbychangesineducationalprovisionmuchfurtherbackintime.Growthaccountingexercisesvarywidelyintheextenttowhichtheydisaggregatelabourinput.NearlyallthestudieswhichcarryoutadetaileddisaggregationbylevelofschoolingarerestrictedtotheUnitedStates;theclassicstudyisJorgenson,GollopandFraumeni(1987).Fortheperiod1948-79,theyfindthatgrowthinlabourinputhascontributedaboutathirdofgrowthinaggregatevalueadded,wherethemeasureoflabourinputtakesintoaccountbothhoursworkedandthequalityoflabour.Changesintheiraggregateindexoflabourqualityarebasedonchangesinthecompositionoftotal12 hoursworkedbyage,sex,education,employmentclassandoccupation.Theyfindthatafavourableshiftinlabourqualityisresponsibleforaboutatenthofthegrowthinvalueadded,oraboutafifthoftheproductivityresidualthatremainsafteraccountingforthecontributionofgrowthinphysicalcapital(seetheirTable9.5).IninterpretingtheresultsofJorgenson,GollopandFraumeni,itisimportanttonotethatsomeofthecompositionalshiftswithinthelabourforcehaveanegativeeffectontheindexoflabourqualityoverthe1948-79period,whichpartlyoffsetthebenefitsofimprovementsineducationalattainment.Aspreviouslynoted,thecalculationofthelabourqualityindexassumesthatdifferencesinmarketrewardsreflectgenuinedifferencesinmarginalproducts.Oneconsequenceisthattheincreasingentryofwomenandyoungworkersintothelabourmarket,mainlyintolow-payingjobs,hasanegativeeffectontheaggregateindexoflabourquality.Overthe1948-79period,thenegativeeffectontheindexoflabourqualityismorethanoffsetbypositivechangesinthecompositionofthelabourforcebyeducationalattainmentandoccupation.Oneimplicationisthatthelattereffectsarelikelytoberesponsibleformorethanafifthoftheproductivityresidual,sincethefavourableshiftinlabourqualitywouldhavebeenlargerintheabsenceofthechangeincompositionbyageandsex.Inreviewingtheevidenceasawhole,Griliches(1997)writesthatincreasesineducationalattainmentseemtohaveaccountedforperhapsathirdoftheproductivityresidualintheUSoverthepost-warperiod.Inthe1950sand1960s,thiswouldcorrespondtoaneffectontheannualgrowthrateofaggregateoutputofaround0.5percentagepoints;duringthe1970sproductivityslowdowntheeffectofeducationalimprovementwillhavebeenlower,perhapsraisingthegrowthrateby0.2or0.3percentagepoints.Asdiscussedabove,theseeffectsareinherentlytransitionalones,drivenbylong-standingchangesineducationpolicythatshifttheeducationalcompositionofthelabourforcetowardsanewsteadystate.OtherOECDmembershavealsoseenimportantchangesineducationalattainmentinthelastfiftyyears.EnglanderandGurney(1994a)notethattertiaryeducationinparticularhasexpandedrapidlyinmanyOECDcountriessince1960.MoredetailonthegeneraltrendscanbefoundinOECD(1998,ch.2)andOECD(2000a,2000b).Asyet,however,therearefewstudiesthatcoverrecentexperienceofotherOECD15membercountriesinthesamedegreeofdetailasJorgenson,GollopandFraumeni.ThebestknownstudiescoveringanumberofdevelopedcountriesforrecentyearsareththoseofMaddison(1987,1991).Maddison(1991,p.138)arguesthatthe20centurysawafairlysteadyimprovementineducationalattainmentforthesixcountriesheconsiders(France,WestGermany,Japan,theNetherlands,theUKandtheUS).OneimplicationisthatchangingtrendsineducationalattainmentareunlikelytoprovideasatisfactoryexplanationforthetransitionfromEurope’s‘GoldenAge’ofrapidgrowth(1950-73)totheproductivityslowdownafter1973.15TheKLEMSprojectisseekingtoextendthisformofanalysistoothermajoreconomies.Seehttp://www.conference-board.org/economics/klems/index.htm13 Forthesesixcountries,Maddisonestimatesthegrowthimpactofchangesineducationalattainmentbydisaggregatingthelabourforceintothosewithprimary,secondaryandhigherqualifications.Hethencombinesthesethreedifferenttypesoflabourusingweightsthatarethesameacrosscountriesandovertime.Inselectingtheweights,hefollowsDenison(1967)inassumingthatobservededucationalwagedifferentialsoverstatethecontributionofeducationtoproductivity,forthereasonsdiscussedinsection3.1above.Inevitably,theadjustmentsmadearesomewhatarbitrary,buttheydoservetohighlighttheuncertaintyinherentinthegeneralapproach.Theotherpointtonoteisthat,becauseoftheseadjustments,theestimatesofDenisonandMaddisonarenotdirectlycomparablewiththoseofotherstudies.Withallthisinmind,wecanturntoMaddison’sresultsonthecontributionofincreasesinlabourqualitytooutputgrowthinFrance,WestGermany,Japan,theNetherlands,theUKandtheUS.Hisfiguressuggestthatchangesinthequalityofthelabourforcetypicallyaddedbetween0.1and0.5percentagepointstoannualgrowthratesbetween1950and1984(hisTable20).TheMaddisonindexoflabourqualitytakesintoaccountchangesinthemale/femalecomposition(thoughnotagecomposition)ofthelabourforce,aswellaschangesineducationalattainment.Incountrieswheretheproportionofwomeninthelabourforcehasnoticeablyrisen,suchastheUKandtheUS,thecontributionofeducationtogrowthwillbeslightlyhigherthanthereportedcontributionofgrowthinlabourquality.MorerecentstudiesincludethatofJorgensonandYip(1999),whohaverecentlycarriedoutadetailedgrowthaccountingexercisefortheG7,andpresentestimatesofgrowthinlabourqualityfor1960-95(theirTable7).TheseestimatessuggestthatlabourqualityhasgrownparticularlyquicklyinJapan,andtoalesserextent,relativelyquicklyinFranceandtheUS.TheJorgenson-YipdisaggregationofthelabourforceisslightlyfinerthanthatadoptedbyMaddison,andthismakesithardertoassesstheroleofeducationwithinchangesintheoverallindexoflabourquality.AusefulsurveybyEnglanderandGurney(1994b)drawstogethertheresultsofanumberofstudiesfortheG7,althoughsomeofthisevidenceisbasedonregressionsratherthangrowthaccounting.Theirsummarysuggeststhatforthe1960sto1980sthegrowthofhumancapital(sometimesincludingdemographiceffects,ofthekinddiscussedabove)typicallyaccountsforatenthtoafifthofgrowthintotaloutput.Forthosecountries,liketheUS,wheretherehasbeenarapidincreaseinemployment,thesefiguresprobablyslightlyunderstatetheproportionofgrowthinoutputperworkerthatcanbeattributedtorisingattainment.AnotherOECDcountryforwhichrecentanddetailedgrowthaccountingresultsareavailableisKorea.ThemostinfluentialcontributionisthatofYoung(1995),whoexaminesandcomparesthegrowthperformanceoffourEastAsianeconomies.Forthepurposeofthepresentsurvey,thecaseofKoreaisparticularlyinterestinginthatthecountryhasseenadramaticincreaseintheeducationalattainmentofthelabourforce.Between1966and1990,theproportionoftheworkingpopulationwithsecondaryleveleducationorhigherroughlytrebled,from27%to75%.Yetthisdramaticexpansiondoesnottranslateintoanequallydramaticeffectonthegrowthrate,atleastundertheassumptionsofgrowthaccounting.Foreachofthefoureconomiesheconsiders,Youngfindsthattheimprovingeducationalattainmentoftheworkforceraisedtheannual14 growthrateofeffectivelabourinputbyaboutonepercentagepoint(Young1995,p.645).Iendthissectionbynotinganessentialqualificationtotheresultsabove,andapossibleextensiontotheconventionalapproach.Allgrowthaccountingresultsrequirecarefulinterpretation,becausetheapproachdoesnottelluseverythingweneedto16knowabouttherelevantcounterfactual.Asanexample,consideraclaimthatXpercentagepointsofgrowthinagivencountryisduetoachangeinthequalityofthelabourforce.Thisdoesnotimplythat,intheabsenceofthechangeinlabourforcequality,thegrowthrateofoutputwouldhavebeenpreciselyXpercentagepointslower.Theproblemisthateducationalattainmentmayhaveother,indirecteffectsonoutputthroughlabourforceparticipation,investment,andevenR&Dandthegrowthoftotalfactorproductivity.Growthaccountingdoesnotcapturetheseindirecteffects,andsogivesonlyapartialpictureoftheoverallimportancetogrowthofvariableslikeeducation.Finally,althoughaccountingdecompositionsareusuallyappliedtogrowthrates,thesameideascanbeappliedtodecompositionsofoutputlevels.Wecanthenaskquestionssuchas:towhatextentdodifferencesineducationalattainmentexplainthevariationinGDPpercapitaacrossOECDmembercountries?Researchapplyingsuchideasisjuststartingtoemerge,andWoessmann(2000)discussestheapproachinmoredetail.Workingontheassumptionthatmeasuredprivatereturnstoschoolingarecapturingagenuineproductivityeffectofeducation,hisanalysissuggeststhatdifferencesineducationalattainmentaccountformostoftheoutputvariationacrossOECDmembers.3.3EvidencefromgrowthregressionsAlthoughgrowthaccountingexercisesareinformativeandoftenuseful,itisclearthattheyarenotacompletesubstituteforotherformsofinvestigation,giventhenecessaryassumptions.Griliches(1997,p.S333)writesthat“themain,andpossiblyonly,approachtotestingtheproductivityofschoolingdirectlyistoincludeitasaseparatevariableinanestimatedproductionfunction”.Suchestimatescouldbeattheleveloffirmsorregions,butmuchoftheevidenceusesthevariationineducationacrosscountries,anditistosuchestimatesthatIturnnext.Thekeyattractionofgrowthregressionsisthattheyprovideawayoftestingdirectlyforproductivityeffectsofeducation.Thishassometimesbeennotedinthetheoreticalliterature:Arrow(1973,p.215)pointedoutthattheuseofmacroeconomicevidencewouldbeonewayoftestingthesignallingarguments,althoughhealsoexpresseddoubtsaboutthelikelyreliabilityofsuchanapproach.Recentworkhasledtoabetterunderstandingofpreciselywhenandwherescepticismmightbejustified.Inwhatfollows,Iwillreviewthemostimportantproblemsassociatedwithmeasuringgrowtheffectsofeducationatthemacroeconomiclevel.An16BarroandSala-i-Martin(1995,p.352)makethispointingreaterdetail.15 underlyingthemeisthat,despitetheseproblems,therearesomegroundsforoptimismthatthisresearchcanyetmakeaworthwhilecontribution.Thismayseemsurprising,giventhatseveralwell-knownpapersinthisfieldtakeverydifferentviewsontheimportanceofeducation.Theargumentbelowisthatamorecoherentstoryisgraduallystartingtoappear,inwhichtheresultsofcross-countrystudiesincreasinglylookconsistentwiththeeffectsidentifiedbylaboureconomists,andwhichcanalsoexplainwhysomeearlierstudiesfailedtodetectanysignificanteffectofeducationusingaggregatedata.Intheearlyworkinthisfield,someoftheestimatedeffectslookedtoolargetobecredible,aswillbediscussedfurtherbelow.OneofthebestknownandmostinfluentialcontributionstotheempiricalgrowthliteratureisthatofMankiw,RomerandWeil(1992).Iftakenatfacevalue,theirparameterestimatesforanOECDsampleimplythatifhumancapitalinvestment(asashareofGDP)isincreasedbyatenth,outputperworkerwillriseby6%;ifinvestmentinhumancapitalisdoubled,outputperworker17willeventuallyrisebyabout50%.Resultsofthiskindareoftenperceivedasratherdubious,sinceallgrowthregressionsshareanumberofimportantstatisticalproblems(Temple1999a).Inthepresentcontext,onedrawbackofmostregressionstudiesistheirfocusonalargesamplethatincludeslessdevelopedcountriesaswellasOECDmembers.OneshouldclearlyberatherwaryaboutdrawingconclusionsforOECDpolicybasedonsamplesthatareoftendominatedbydevelopingcountries.IwillusuallyconcentrateonthefewstudiesthatincludeseparateestimatesofregressionsrestrictedtoeitherOECDmembersorrichcountries.Researchershavegenerallyusedoneoftwospecificationsinmodellinggrowthandeducation.Inthefirstandmostcommonspecification,theresearcherchoosestoregressgrowthoncontrolvariablesandtheinitiallevelofaneducationmeasure,suchasthesecondaryschoolenrollmentrateor(preferably)averageyearsofschooling.Theunderlyingideaisthatthestockofhumancapitalcouldaffectsubsequentgrowthinavarietyofways,notablybyinfluencingacountry’sabilitytoadopttechnologyfrom18abroad.Thesecondspecificationusesthechangeineducationalattainment,notitslevel,toexplainoutputgrowth;thisapproachwillbediscussedfurtherbelow.Ithassometimesbeenarguedthatinpractice,onemightexpectanegativeeffecttoemergefromregressionsbasedonthelevelofeducation,andthispotentialambiguitycouldmaketheresultshardtointerpret(Topel1999).Forexample,countrieswithalowlevelofeducationmayalsoberelativelyfarbehindtechnologicalleadersliketheUS,andthereforehavemoreopportunitiestocatch-upandgrowquickly.Argumentsofthiskindarenotyetaltogetherconvincing.Inthisspecificcase,oneshouldnotethatgrowthregressionsusuallycontrolforinitialoutputperworker,andthiswill17OtherpaperswhichextendthesefindingsfortheOECDsample,andatleastimplicitlyexaminetheirrobustness,includeNonnemanandVanhoudt(1996),Temple(1998)andVasudevaMurthyandChien(1997).18Thereisalsoimportantworkonhumancapitalasadeterminantoftechnologicalcatch-upusingdataatthesectorallevel.Forexample,Cameron,ProudmanandRedding(1998)investigatetheroleofhumancapitalandopennesstotradeinexplainingcatch-upbyUKmanufacturingsectors.16 incorporatealargepartofthecatch-upeffectsassociatedwithtechnologicalbackwardness.Whenresearchersrelategrowthtotheinitiallevelofeducation,theytypicallyfindaneffectofschoolingthatisbothlargeandpreciselyestimated,atleastwheninitialoutputperworkerisalsoincludedasanexplanatoryvariable(asinBarro1991).YetitisnotclearthattheseresultsareapplicabletoOECDmembers.Inaninterestingexercise,EnglanderandGurney(1994a)re-estimategrowthregressionsbasedonfourinfluentialpapers,includingBarro(1991),butrestrictingthesampletotheOECD.Threeofthefoursetsofregressionsincludehumancapitalvariables,typicallyprimaryand19secondaryschoolenrollmentrates.Thesevariablesturnouttoperformrelativelywell,butarestillfarfromrobust.Infurtherwork,itmaybevaluabletorepeatthisexercise,drawingonmorerecentdatasetsthatallowonetouseaverageyearsofschoolingratherthanenrollmentrates.AmorerecentpaperthatincludesresultsspecifictoOECDsamplesisGemmell(1996).Heemphasisestheproblemsofusingenrollmentrates,andconstructsalternativemeasuresofhumancapitalbasedonattainmentattheprimary,secondaryandtertiarylevels.Forasampleof21OECDcountries,hefindsacorrelationbetweenthenumberofpeoplewithtertiaryqualificationsandsubsequentgrowth.HealsofindssomeevidencethatinvestmentinOECDcountriesispositivelycorrelatedwiththeextentofsecondaryschoolinginthelabourforce.Onedrawbackofmostcross-countryworkisthelikelihoodofimportantdifferencesinthenatureandqualityofschoolingacrosscountries,whichcouldunderminetheusefulnessofinternationalcomparisons.Evensuchthingsasthelengthoftheschoolyearcanshowasurprisingdegreeofvariationacrosscountries.Analternativedataset,whichmayovercometheseproblemstosomeextent,hasbeenintroducedbyHanushekandKimko(2000).Theymeasureeducationalattainmentusingscoresininternationaltestsofcognitiveskillsinmathsandscience.Theirresultssupporttheideathateducationhasasubstantialeffectongrowthrates,althoughtheapplicabilitytoOECDcountriesisnotclear.ThelackofstudieswithdirectrelevancetotheOECDisnottheonlydilemmaforthosewhowishtodrawpolicyconclusionsfordevelopedcountries.Theratheratheoreticapproachofthemacroeconomicliteratureoneducationandgrowthhasattractedacertainamountofcriticism.Oneargument,usedbyTopel(1999),isthatthemeasuredeffectoftheinitiallevelofhumancapitalisoftentoolargetobecredible.Theunderlyingassumptionhereisthateducation’seffectsaremostlyaccountedforbyexaminingthecorrelationbetweeneducationandearningsattheindividuallevel.Themodelsofthenewgrowththeory,reviewedinsection2above,indicatethatthisviewofeducation’sroleisperhapstoonarrow.19Ofthetwomeasures,onlythesecondaryschoolenrollmentrateislikelytoberelevantinexplaininggrowthwithintheOECD.EnglanderandGurney(1994a)reportthataveragesecondaryenrollmentintheOECDwasabout70%in1960,sotheremaybeenoughvariationacrosscountriesforregressionevidencetobeinformative.17 Nevertheless,theperspectiveoflaboureconomicsremainsofinterest.StartingwithPritchett(1996),researchershavenotedtheimplicationsoftraditionalearningsfunctionsforanalysesatthecross-countrylevel.Ifanindividual’seducationcontributesdirectlytotheirproductivity,inthemannerenvisagedbylaboureconomists,weshouldexpecttoobserveacorrelationbetweenthechangeinoutputperworkerandthechangeinaverageeducationalattainment,atleastaftercontrollingforothervariables.Furthermore,itshouldbepossibletodetectthiseffectregardlessof20whetherornottheinitiallevelofeducationalattainmentdeterminesgrowth.Thisargumenthasshiftedthefocusofresearchtowardsregressionsthatrelategrowthtothechangeineducationalattainment,ratherthanitslevel.Severalwell-knownstudieshavefoundthecorrelationtobesurprisinglyweak;BenhabibandSpiegel(1994)andPritchett(1996)bothcometothisconclusionforalargesampleof21countries.BenhabibandSpiegeldofindastatisticallysignificantcorrelationbetweenthelevelofeducationalattainmentandgrowthforthewealthiestthirdofthesample(theirTable5,model2)butnoconnectionbetweenthechangeinattainmentandgrowthinalargersample.Onereasonforthismaybetheeffectofoutliers,asdiscussedinTemple(1999b,2001).Thereareanumberofotherproblemsthatdictatecautioninreadingthesepapers.Oneisthespecificationchosenfortherelationbetweenyearsofschoolingandoutput.ThespecificationadoptedbyBenhabibandSpiegelimplicitlyassumesthatthereturnstoanextrayearofschoolingaremuchhigheratlowlevelsofschoolingthanhighlevels.AsTopel(1999)pointsout,thisrunscontrarytothestandardsemi-logarithmicformulationforearningsfunctions,whichinitssimplestformassumesthatthereturnstoanextrayearofschoolingareindependentofthelevelofschooling.Whengrowthregressionsarespecifiedinawaythatismoreconsistentwiththisidea,theevidenceforaneffectofeducationisratherstronger.KruegerandLindahl(1999)havearguedconvincinglythatanotherimportantproblemislikelytobemeasurementerror.Thedifficultyisthataspecificationbasedonanaggregateproductionfunction(asinBenhabibandSpiegel)typicallyseekstoexplaingrowthusingthechangeineducationalattainment,butfirst-differencingtheeducationvariableinthiswaywillusuallyexacerbatetheeffectofanymeasurementerrorsinthedata.Tosupportthisargument,KruegerandLindahlexaminethecorrelationbetweentwodifferentmeasuresofthechangeinaverageyearsofschoolingthathavebeenusedintheliterature.Thecorrelationislowenoughtosuggestthatasubstantialcomponentofthemeasuredchangeineducationalattainmentisuninformativenoise.Asa20ProblemsindiscriminatingbetweenthetwoeffectsarediscussedinCannon(2000).21Thisfindingisalsoassociatedwithanumberofpaneldatastudiesusingfixedeffects,buttheseresultsshouldalmostcertainlybediscounted.Researchersusingpanelstypicallydonotallowforlagsintheeffectofvariableslikeenrollmentrates.Inanycase,giventhewaytheeducationdataareconstructed,thetimeseriesvariationwillsometimesbetoonoisytodrawsensibleconclusions.18 consequence,regressionsthatusethechangeineducationtoexplaingrowthwilltend22tounderstateitsimportance.ThecaseforseeingmeasurementerrorasanimportantpartofthestoryhasbeenconsiderablystrengthenedbytheimpressivelycarefulanddetailedworkofdelaFuenteandDomenech(2000).Unusually,theyrestrictattentionthroughouttoOECDmembers.Theircloseexaminationofstandarddatasetsrevealsthatschoolinglevelsforsomecountriesappearimplausible;someofthefiguresforaverageyearsofschoolingdisplaysurprisingshort-runvolatility;andothersappeartogiveamisleadingviewoftrends.Otherwriters,notablySteedman(1996),havealsonotedinconsistenciesinthewaydataonhumancapitalarecollectedandcompared.BydrawingonnationalsourcesandmorerecentfigurescompiledbytheOECD,delaFuenteandDomenechcompileanewandmorereliabledatasetforyearsofschoolinginOECDmembercountries.Intheirempiricalwork,theyfindthatchangesinoutputandeducationalattainmentarepositivelycorrelated,eveninpanelestimatesthatincludecountryandtimefixedeffects.Thissupportstheideathat,wherepreviousresearchershavefailedtodetectaneffect,thismaybeduetomeasurementerror.Morerecently,BassaniniandScarpetta(2001)haveextendedthedelaFuenteandDomenechdatabaseforwardintime,andestimatedtheeffectofeducationover1971-98for21OECDmembersusingthePooledMeanGroup(PMG)estimator.Thekeyadvantageofthisapproachisthat,comparedtotraditionalmethodsofestimatingpaneldatamodels,itallowsgreaterflexibilityintheshort-rundynamics.UsingthePMGestimator,BassaniniandScarpetta’spreferredestimateisanelasticityof0.6foroutputpercapitainresponsetoadditionalyearsofschooling.Thisimpliesthat,atthesamplemeanofaverageschoolingofabouttenyears,anextrayearofaverageschoolingwouldraiseoutputpercapitabysixpercent.Thiseffectissimilarinmagnitudetothatfoundinmicroeconomicestimatesbasedonsurveydata,ofthetypereviewedearlier.Engelbrecht(1997)alsofindssignificanteffectsofeducationonOECDgrowth.HisempiricalmodelcontrolsfortheeffectsofR&Dspending,andisestimatedusingtheeducationdataofBarroandLee(1993)forthepopulationaged25andover.Again,theseresultssuggestthatthegrowthofproductivityisrelatedtothechangeinaverageyearsofschooling,asonewouldexpectifmicroeconomicestimatesofthereturntoschoolingarepickingupagenuineproductivityeffect.Inaseparatesetofestimates,Engelbrechtalsofindssupportfortheideathatthelevelofeducationplaysaroleintechnologicalcatch-up;hefindsproductivitygrowthismorerapidwherecountrieshaveahigherlevelofaverageschooling.Overall,thisliteratureisbeginningtosuggestthatthereisacorrelationbetweenchangesineducationandgrowth,ofthekindthatmostlaboureconomistswouldexpecttoobserve.Thisisreassuring,butthereareanumberofinterestingopenquestions.Oneobviousquestionmarksurroundstheinterpretationoftheearlierresultsthatrelatedgrowthtotheinitiallevelofattainment,ratherthanthechangeinattainment.GrowthstudiesfortheOECDthatallowaroleforbothpossibilitiessimultaneouslyareyetto22Note,though,thatmeasurementerrorinotherexplanatoryvariables(notablyphysicalcapital)couldbiasthecoefficientoneducationintheoppositedirection.19 appear.Thisomissionmaybeinevitablegiventhesmallsamplesize,butitshouldnotleadonetounderestimatethepossibleroleforhumancapitalintechnologicalcatch-uporthecreationofnewideas,eitherofwhichcouldyieldarelationshipbetweenthelevelofeducationandsubsequentgrowth.Thereisanotherreasonwhytheeffectoftheinitiallevelofeducationremainsofsomeinterest.Studyingtherelationbetweenthechangeinoutputandthechangeineducationremainssomewhatvulnerabletothechargethatcausalityrunsfromoutput23(oranticipatedoutput)toeducation,andnotsimplyviceversa.Toalargeextent,long-runchangesinaverageeducationalattainmentaredrivenbygovernmentpolicy.Itseemsplausiblethatasoutputandtaxrevenuesincrease,governmentswilloftenallocatemoreresourcestoeducation,andattainmentwillriseforatransitionalperiod.Yettheargumentthatpaneldataresults,suchasthoseofdelaFuenteandDomenech(2000),aredrivenbyreversecausationisratherlessstrongthanitmayappearatfirst.Thisisakeyadvantageoftheiruseofdataonaverageyearsofschoolinginthepopulation,ratherthanenrollmentrates.Giventhatnewentrantsaretypicallyasmallfractionofthelabourforce,averageattainmentwillchangeonlyveryslowlyinresponsetoanychangeineducationalprovision.Itthereforeseemsratherunlikelythatreversecausationexplainsthepaneldatafindings.Wheredoesthisleaveus?Earlierinthesurvey,wesawtheimportantqualificationsthatsurroundmicroeconomicestimatesofthesocialreturnstoschooling.Ultimatelywewouldlikethecross-countryevidencetoshedlightontheaccuracyoftheseestimates.Inpractice,wearelikelytoremainsomewayshortofthisgoal,atleastintheabsenceofbetterdata.Theaggregateevidenceiscurrentlytoofragiletodrawanystrongconclusionsaboutthepossibleextentofsocialreturns.Evenso,theresultswehaveprovidesomegroundsforoptimism,anditisreassuringthatseveralrecentstudiesfindeducationtobeimportant,despitethelikelypresenceofmeasurementerror.Thissuggeststhatbetterdata,andmoresophisticatedmethods,mayyetleadtoasteadyimprovementintheprecisionofourestimatesofthegrowtheffectsofeducation.Theprospectsforthisshouldnotbeexaggerated,butthereiscertainlymorereasontobehopefulnowthanintheearlydaysoftheliterature,whenthevarioussetsofestimateswerehardtoreconcileintoanykindofcoherentstory.Anotheradvantageretainedbythemacroeconomicapproach,comparedtomicroestimates,isthatwecanexploreindirecteffectsofeducation,notablythoseworkingthroughinvestment.TheseeffectsarepresentinthemodelintroducedbyMankiw,RomerandWeil(1992)andmayhavewiderrelevance.Two-sectormodelsofendogenousgrowth,suchasthosereviewedinBarroandSala-i-Martin(1995,ch.5),typicallyyieldasteady-stateinwhichthereisanequilibriumratioofhumancapitaltophysicalcapital.Animmediateconsequenceisthatariseineducationalattainmentwilleventuallybemetwithacorrespondingriseinthestockofphysicalcapital.23Thetwo-wayinteractionbetweengrowthandeducationisdiscussedinmoredetailbyMincer(1996)andBilsandKlenow(2000).BilsandKlenowarguethatthedirectionofcausalitymaybeuncertainevenwhenattentionisrestrictedtothegrowtheffectoftheinitiallevelofeducation.20 Analysingtheconsequencesforwelfareisnotwhollystraightforward.Growtheconomistshavenotyetdevelopedandcalibratedamodelwhichderivesoveralloutput24andwelfareeffectsofeducationbasedonsensiblemicrofoundationsforinvestment.Thismayexplainwhytheeffectisignoredbymostinterpretationsoftheempiricalliteratureoneducationandgrowth.Fornow,itisimportanttobeawarethatgrowthregressionsandgrowthaccounting,byusingcapitalgrowthasoneoftheconditioningvariables,mayunderstatethetotalimpactofanincreaseineducationalattainmentonoutputperworker.Theprobablemagnitudeofthiseffect,anditssignificanceforwelfare,remainuncertain.3.4HumancapitalexternalitiesOneimportantmotivationforlookingatthecross-countrydataisthepossiblepresenceofexternalitiestohumancapital.Aswehaveseen,however,theempiricalgrowthliteraturegivesratherimpreciseanswersaboutthesocialreturnstoeducation.Inthissection,Iwillbrieflyreviewtheoreticalworkonthistopic,andthendiscusssomeinnovativerecentevidencebasedonmicroeconomicdatasets.InterestinhumancapitalexternalitieswasrevivedbyLucas(1988,1990).Oneofhisargumentswasthat,intheabsenceofsuchexternalities,itisdifficulttoreconcileobservedpressuresformigrationfrompoortorichcountrieswiththeabsenceofmassivecapitalflowsintheotherdirection.HealsodrewontheworkofJacobs(1969)toarguethatsuchexternalitiesareanaturalexplanationfortheexistenceofcities.Inmorerecentwork,Acemoglu(1996)hasprovidedaningeniousjustificationforthepresenceofexternalities.Histheoryisbasedonmicroeconomicfoundations,andsoisparticularlyworthyofattention.Inhismodel,firmsandworkersmakeinvestmentsinphysicalcapitalandhumancapitalrespectively,beforeproductionbegins.Productionrequiresapartnershipbetweenafirmandaworker,butwhenfirmsorworkersmaketheirrespectiveinvestments,theydonotknowtheidentityoftheirfuturepartner.Akeyassumptionofthemodelisthatfirmsandworkersarethenbroughttogetherviaamatchingprocessthatisimperfect,perhapsbecausesearchingforpartnersiscostly.Acemoglushowshowthestructureofthemodelyieldsanimportantresult:anincreaseintheaveragelevelofhumancapitalcanhaveapositiveeffectontheprivatereturntohumancapital,atleastoversomeregion.Theintuitionisasfollows:saythatasubsetofworkersdecidestoacquiremorehumancapital.Thiswillraiseaveragehumancapital,andanticipationofthisencouragesfirmstomakegreaterinvestmentsinphysicalcapital.Sincethematchingprocessisinefficient,thefirmswhohaveinvestedmorearenotnecessarilymatchedwiththeworkerswhohaveinvestedmoreinhumancapital.Asaresult,someoftheotherworkerswillgainfromtheincreaseinaveragehumancapital,sincetheyarematchedwithfirmsusingmorephysicalcapitalthanbefore;andinthissensetheaveragelevelofhumancapitalhasanexternalbenefit.24Althoughsometheoreticalworkhasstartedtoappear:Masters(1998)analysestheefficiencyofinvestmentsinhumanandphysicalcapitalinabilateralsearchcontext.21 Workofthiskindhashelpedtomotivatetherecentsearchforexternalities,usingsurveydatasetsthatincludeindividualswholiveindifferentcitiesorregions.Theideaistoestimatehumancapitalearningsfunctionsinthenormalway,butincludinganewvariable,theaveragelevelofschoolingineachindividual’scityorregion.Thecentralideaisthat,iftherearesignificantexternalitiestohumancapital,individualsshouldearnmorewhentheyworkinthosecitieswithahigheraveragelevelofschooling.Theexercisewillmissexternalitiesthatworkatthenationallevel,perhapsthroughsocialstructuresorinstitutions,butitremainsofconsiderableinterest.SeveralstudiesbasedonthisideahavebeencarriedoutfortheUS.TheinitialresultsofRauch(1993)appearedpromising.Considertwootherwisesimilarindividualslivingintwodifferentcities,thesecondcitywithapopulationthathasanextrayearofaverageschooling.Hisestimatessuggestedthatanindividuallivinginthesecondcitycouldexpecttogainawagepremiumofaround3%,aneffectlargeenoughtobeworthyoffurtherinvestigation.Unfortunately,asCicconeetal.(1999)pointout,thereisanimportantargumentagainstinterpretingtheobservedwagepremiumassolelydrivenbyexternalities.Differencesinaverageyearsofschoolingacrosscitiesarelikelytobeassociatedwithdifferencesintherelativesuppliesofskilledandunskilledlabour.Theserelativesupplyeffectsmaygiverisetoanapparentwagepremiumforaverageschoolingevenintheabsenceofexternalities.TheempiricalworkofCicconeetal.(1999)supportsthisproposition.WhentheyfollowRauchanddonotallowforrelativesupplyeffects,theyareabletoobtainahighandpreciseestimateofthesocialreturntoeducation.Inamoregeneralapproach,whichbuildsinaroleforsupplyeffects,themeasuredexternalitiesaregreatlyreduced;indeeditisnotpossibletorejectthehypothesisthatexternalitiesareabsentaltogether.RelatedworkbyAcemogluandAngrist(1999)alsoindicatesthattheoverallsocialreturnstoeducationmaybeclosetotheprivatereturns,thistimeusingthevariationinaverageschoolingacrossUSstatestocapturetheeffectsofexternalities.3.5WiderbenefitsofeducationSofar,thesurveyhasonlyconsideredtheeffectsofeducationonproductivity,yetitisclearthatthebenefitsofeducationarelikelytobemorefar-reaching.Thetraditionalcaseforeducationisthatitmakesafundamentalcontributiontopersonaldevelopment,andprobablytothehealthofsocietiesmoregenerally.Inthinkingaboutpublicprovision,itiscrucialtorememberthateducationmayhavesignificantwelfarebenefitsthatarenotcapturedinthemodelsanddatatypicallyanalysedbyeconomistsandgovernments.Theseincludeeventhebenefitsofeducationthataccruedirectlytoindividuals.Itisplausiblethateducationhasbothanimmediateconsumptionbenefitandalong-termeffectonlifesatisfaction,otherthingsequal.Thedifficultyhereisthatitismuchhardertomeasurewell-beinginameaningfulwaythanitistomeasureoutputofgoodsandservices,andeconomistsareonlyjuststartingtoinvestigatewell-beinganditsdeterminants.22 Inaninnovativepaper,BlanchflowerandOswald(2000)reportestimatesof‘happinessequations’,regressionsthatrelatesurveymeasuresofwell-beingtoindividualcharacteristics.Theyfindthateducationalattainmentisassociatedwithgreaterhappiness,evenwhencontrollingforfamilyincome.Suchfindingscouldhaveimportantimplicationsforeducationpolicy.Forexample,itisquitepossiblethattheextentofanindividual’seducationhasapositiveeffectonthewell-beingofothers,inwhichcaseself-interestedindividualsmaytendtounder-investineducationfromsociety’spointofview.Alternatively,educationmayaffecthappinessbecauseitinfluencesperceptionsofstatusrelativetoothers,inwhichcasetheoverallwelfarebenefitsofeducationmaybelessthantheresultsofBlanchflowerandOswaldseemtoimply.Educationpolicyalsohasimplicationsforsocietyasawhole.Someeconomistsmayfeelthatthesewiderbenefitslieoutsidetheremitofthesubject,butthisargumentwouldmeandepartingfromtheorthodoxdefinitionofeconomics-namelythestudyoftherelationbetweentheallocationofscarceresourcesandhumanwelfare.Educationalprovisionmayaffectpublichealth,crime,theenvironment,parenting,andpoliticalandcommunityparticipation.SomeoftheseeffectsarediscussedinmoredetailinOECD(1998,ch.4),BehrmanandStacey(1997)andWolfeandHaveman(2000).Allofthesewiderbenefitscouldfeedbackintoeconomicperformance,whichreinforcesthecaseforamuchbroaderviewofeducation’srole.3.5AtentativesummaryoftheevidenceAtthispoint,onemaybeleftwonderingwhattheevidenceultimatelyachievesintermsoflessonsforpolicy.Themostusefulperspectiveisprobablytocombinethevariousstrandsofevidenceandseewhethertheyformacoherentwhole,despitetheproblemsinherentineach.Laboureconomistsseemtobeagreedthattheprivaterateofreturntoayear’sextraschoolingistypicallybetween5%and15%.Workingundersimilarassumptions,growthaccountantsfindthatincreasesineducationalattainmentaccountforperhapsafifthofgrowthinoutputperworker.Laboureconomicsandgrowthaccountinghavearelativelylonghistory,andthestrengthsandweaknessesoftheavailableevidencearewellunderstood.Itispossiblethatbothapproachesoverstatethesocialbenefitsofeducation,perhapsbecauseofsignallingeffects.Actingintheotherdirection,theestimatesprovidedbythisresearchmayunderstatetheroleofeducation,becausetheyrarelyallowmeasurementofexternalities,orquantifytheimportanceforproductivityofanimprovedmatchingbetweenworkersandjobs,orincorporatethemoregeneralmechanismsconnectingeducationandgrowththatarefoundintheoreticalmodels.Thegreatstrengthoftheemergingmacroeconomicliteratureisthat,atleastinprinciple,itcouldprovideadirecttestoftheproductivitybenefits.Aswehaveseen,however,thisfieldhassignificantweaknessesofitsown.Estimatesthataresufficientlyaccurateandrobusttoallowconfidentconclusionsaresomewayoff.Thatmayhaveto23 waituntilgrowtheconomistshavelongerspansofdatatoworkwith,andgreaterskillatmatchingavarietyofpossiblestatisticaltechniquestothequestionathand.Withthesecaveatsinmind,abriefsummaryofthemacroeconomicevidencemaybeuseful.Althoughinsomewayssuchanexerciseisratherpremature,itshouldatleastpreventtheunwaryfromjumpingtoanover-hastyconclusionbasedonthereadingofoneortwopapersalone.Thatwouldbeaneasymistaketomake:overthelasttenyears,growthresearchershavebouncedfromidentifyingquitedramaticeffectsofeducation,tocallingintoquestiontheexistenceofanyeffectatall.Morerecentresearchisplacedsomewherebetweenthesetwoextremes,butperhapsleaningclosertotheoriginalfindingsthateducationhasamajorimpact.Inexaminingthestudiesthathavenotdetectedaneffect,wehavesomeconvincingreasons(measurementerror,outliers,incorrectspecification)todoubtsuchresults.Thebalanceofrecentevidencepointstoproductivityeffectsofeducationwhichareatleastaslargeasthoseidentifiedbylaboureconomists.Thisshouldreassureusthatmostcountriesarenotover-providingeducation,butafullerdiscussionofpolicyimplicationswillbedeferreduntilsection5.4.SocialcapitalandgrowthInthissection,Iprovidesomediscussionoftheemergingideaof‘socialcapital’,anditspotentialroleinthegrowthprocess.Theliteratureonsocialcapitalisstillrelativelyundeveloped,andinreviewingtheempiricalworkonthistopic,Iwilldrawheavilyonafewkeypapers.Thenatureofthesurroundingdiscussionisnecessarilybroaderandmorespeculativethanelsewhereinthesurvey,andthisreflectssomeoftheuncertaintiescurrentlysurroundingthefield,whichshouldbeborneinmindthroughout.Beforedescribingtheunderlyingideasinmoredetail,itmaybehelpfultoplacetheminthewidercontextofempiricalgrowthresearch.Theaimistoindicatewhysocialcapitalmightyetbeausefulconcept,giventhatviewsonitsimportancecurrentlydiffergreatly-certainlycomparedtoviewsoneducation,wherethereisfairlygeneralagreementthateducationmatters,evenifourmeasurementsofitseffectareimprecise.Ideally,researchersstudyingdevelopmentandgrowthwouldliketofindasetofpolicyinterventionssufficienttoraiselivingstandardsandwelfare.Itissometimesarguedthatthisisanimpossiblegoal,partlybecausethecircumstancesofeachcountryareunique.Alessextremepositionisthatgrowthresearchcangiveussomeinsightintopossiblegeneralisations,bytellingusabouttheaveragepattern;atthesametime,itshouldberecognisedthatanyproposedsetof‘sufficient’conditionswillneverbeuniversal.Onewayofmakingourgeneralisationsmorewidelyapplicableistodiscriminatemorefinelybetweensocieties,byintroducingextradimensionsintoouranalysisofthegrowthprocess.Thiscannotbepushedtoofar,sinceweonlyhavealimitedsetofcountries,andalimitedtimespan,fromwhichtodrawevidence.Thecentralchallengeforgrowthresearchersistoidentifythedimensionswhicharemostrelevantforgrowth,withoutendlesslymultiplyingthepossibilitiesinsuchawaythatweultimatelyasktoo24 muchofthedata.Atthemoment,thehopeappearstobethatacoherentpicturewillultimatelyemergethroughagradualaccumulationofevidence,asempiricalresearchersbothintroducenewvariablesandindicatethatsomeearlierproposalsshouldbediscarded.Thefundamentalproblemhereisthatthemostgeneralmodel,whichinprinciplewouldallowustodiscriminateeasilybetweenthecompetinghypotheses,hasalreadybecometoolargetobeinformative(LevineandRenelt1992).Inthiscontext,inexplaininggrowth,itmakessensetoconcentrateonthosedimensionsofsocietieswhichhaveastrongpriorclaimonourattention.Amongthedimensionsrecentlyproposedforfurtherinvestigation,onestandsoutasbothpromisingand-intermsofitspriorclaim-relativelycontroversial.Theconceptof‘socialcapital’appearstobeapotentiallyformidablewayofdiscriminatingbetweencountriesandtheirgrowthprospects.Itprovidesausefulwaytothinkaboutaspectsofsocietieswhich,thoughdifficulttomeasureandincorporateintoformalmodels,maybeimportantdeterminantsoflong-runeconomicsuccess.Forsomeeconomists(notall)theintuitionthat‘societymatters’isstrongenoughtooutweighthecurrentabsenceofmuchinthewayofatheoreticalunderpinning.Thereisalongacademictraditionthatsomethingisnotfullyunderstooduntilitcanbemeasured,andtheconceptofsocialcapitalpresentsseriousproblemsofdefinition,letalonemeasurement.Butinthisrespect,itisinterestingtonotethecommentofLucas(1988,p.35)abouttheearlydaysofhumancapitaltheory.Hewrotethat“theideaofhumancapitalmayhaveseemedetherealwhenitwasfirstintroduced-atleast,itdidtome-butaftertwodecadesofresearchapplicationsofhumancapitaltheorywehavelearnedto‘see’itinawidevarietyofphenomena”.Thepossibleanalogywiththepresentandfuturestatusofsocialcapitalshouldbeclear.Overall,itiseasytoseewhygrowtheconomistsandothershavestartedtoemphasisesocialcapitalonlyveryrecently,eventhoughthebasicideashavealongintellectualhistory.Inthispartofthesurvey,Iwilldiscusssomeofthemostrecentwork,startingwithadiscussionofthenatureofsocialcapital(section4.1).Thisprovidesanecessarybackdropforsection4.2,whichcoversthelimitedcross-countryevidencesofaravailable,mostofitbasedonsurveyevidenceonwillingnesstotrust.Theimplicationsforpolicymayseemrathermeagre,butitshouldberememberedthatthisliteratureisstillinitsearlystages.Section4.3willdiscusssomeofthequestionsthatremaintobeanswered.4.1Whatissocialcapital?Itiswidelyacknowledgedthatsocialcapitalneedstobecarefullydefined,ifitistoproveanythingmorethansuggestiveinthinkingaboutgrowth.OneofthebestknownandmostrepresentativedefinitionscanbefoundinthehighlyinfluentialworkofPutnam(1993):“socialcapital...referstofeaturesofsocialorganization,suchastrust,norms,andnetworks,thatcanimprovetheefficiencyofsocietybyfacilitatingcoordinatedactions”(p.167).25 AsWoolcock(1998)andothershavenoted,thisisusefulbutcomesclosetodefiningsocialcapitalintermsofitsfunction,sothatitbecomesdifficulttoseparateanalyticallythesourcesofsocialcapitalfromitsconsequences.Asanexample,socialcapitalintheformoftrustmaybecreatedbyparticipationincivicassociations,buttheseassociationscouldthemselvesberegardedasanimportantformofsocialcapital.Theimportanceofthispointisreinforcedwhenoneconsidersthatsocialcapitalmayalsohavecosts:oneperson’svaluablenetworkmaybeanother’srestrictiveinterestgroup.Manydiscussionsofsocialcapital,includingthoseofPutnam(1993),Schuller(2000)andWoolcock(2000),associateitwitharesourcethatisusefulinachievingcommonobjectives.Forexample,thesuggesteddefinitionofWoolcock(2000,p.5)isthat“socialcapitalreferstothenormsandnetworksthatfacilitatecollectiveaction”.Thisemphasisoncollectiveactionmaybeproblematicforeconomistswhowishtomakewideruseoftheidea.AsIwilldiscusslater,anunderstandingoftheformationofsocialcapitalislikelytorequireanunderstandingofitsvalueasaresourcetoindividuals(Glaeser2000).Thiscaneasilyconflictwithadefinitionofsocialcapitalthatemphasisesitsroleincollectiveaction,intheusualsenseofthelatterterm.Forexample,anentrepreneurwhogainsknowledgefromparticipatinginvariousnetworksisarguablybenefitingfromsocialcapital,andthisbenefitoccurs,andmaybeworthyofanalysis,eveniftheentrepreneurdoesnotsharegoals,objectivesoroutcomeswithothers.AbroaderexplorationofthetermcanbefoundinWoolcock(1998).Heproposesaschemeinwhichithasfourdimensions,roughlycorrespondingto(i)theextentofhorizontalassociations;(ii)thenatureofsocialtieswithincommunities;(iii)thenatureoftherelationbetweencivilsocietyandthestate;and(iv)thequalityofgoverninginstitutions.Independentlyofthesocialcapitalliterature,economistshavemadesomeprogressundercategory(iv),inanalysingthegrowthimpactofthequalityofinstitutions(forinstanceKnackandKeefer1995).Atleastforpresentpurposes,itisnotclearthatbringingthisworkundertheumbrellaofsocialcapitalwillyieldextra25insight.Inanycase,measuringthebenefitsofgoodinstitutionsisarguablyalessurgenttaskthanformulatingpracticaladviceonhowtoimprovebadones,andthegrowthliteraturedoesnothavemuchtoofferhere.Withthesepointsinmind,thissurveywillmainlyrestrictitselftorecentempiricalworkthatusestheextentoftrustinasocietyasanindicatorofitsunderlyingsocialcapital.ItshouldalreadybeclearthatthisisanimperfectandsimplisticwayofcapturingtheideasofPutnamandothers.Trustmaybedeterminedbysocialcapital,butalsobyotheraspectsofsocieties;andtheextentoftrustmaybeinfluenced,inverydifferentways,byallfourofthedimensionsofsocialcapitalidentifiedbyWoolcock.Yetafocusontrusthasonekeyadvantage:itcanpotentiallybemeasuredinawaythatiscomparableacrosscountries,aswewillseebelow.25FollowingAbramovitz(1986),TempleandJohnson(1998)argueinfavouroftheuseoftheterm“socialcapability”whenreferringtosocialarrangementsandinstitutionsdefinedmorebroadly.Therearelikelytobeimportantbenefits,intermsofclarityandrigour,fromkeepingtheterm“socialcapital”narrowlydefined,asPutnamhasadvocated.26 4.2EmpiricalevidenceThemostimportantmacroeconomicevidenceonsocialcapitaltakestheWorldValuesSurveyasitsstartingpoint.The1981surveyisbasedonresponsesfromthousandsofindividualsacross21marketeconomies,whilethe1990-91surveycovers28marketeconomies.Overall,29marketeconomiesarecoveredatleastonce.Theselectionofrespondentsisnotcompletelyrandom,butadjustmentstotakethisintoaccountare26available.Amongtheissuesaddressedinthesurveys,economistshavemainlyfocusedonaquestiondesignedtocapturewillingnesstotrust.Respondentswereasked“Generallyspeaking,wouldyousaythatmostpeoplecanbetrusted,orthatyoucan’tbetoocarefulindealingwithpeople?”.Thepercentageofrespondentsineachnation27replying“mostpeoplecanbetrusted”formsapotentiallyusefulindexoftrust.Table1showsvaluesforthisindex,TRUST,forthoseOECDcountriescoveredinthesurvey,andalsoforasmallselectionoflessdevelopedcountries.Table1AmeasureoftrustDataforsomeOECDmembersNorway61.2Ireland40.2Finland57.2Korea38.0Sweden57.1Spain34.5Denmark56.0Austria31.8Canada49.6Belgium30.2Australia47.8Germany29.8Netherlands46.2Italy26.3UnitedStates45.4France24.8UnitedKingdom44.4Portugal21.4Switzerland43.2Mexico17.7Iceland41.6Turkey10.0Japan40.8Non-OECDmembersIndia34.3Nigeria22.9SouthAfrica30.5Chile22.7Argentina27.0Brazil6.7Source:KnackandKeefer(1997).Clearly,measurementerrorispotentiallyamajorprobleminusingsuchdata.ForthetwentycountrieswithTRUSTvaluesforboth1981and1990,thecorrelationbetweenthetwois0.91,whichsuggestsalimitedrolefortransitorymeasurementerrors.Thisleavesopenthepossibility,however,thatthevariableisanaccuratemeasureof26Foramoredetaileddiscussion,seeKnackandKeefer(1997).27Notethatsomeone’sresponsetothesurveyquestionmaytellusmoreabouttheirowntrustworthiness,ratherthanaviewoftrustintheircountryasawhole.Eventhen,thepatternofresponsesmayformausefulguidetotheprevalenceoftrustinaparticularcountry(seeforinstanceGlaeser2000).27 somethingotherthantheextentoftrust.KnackandKeefer(1997)reportonaninterestingexperiment,thatprovidesindependentevidenceonthepossiblevalidityoftheTRUSTmeasure.Intheexperiment,alargenumberofwalletscontaining$50weredeliberately‘lost’inanumberofcities.Thepercentageof‘lost’walletsthatarereturnedtotheirownersineachcountryhasacorrelationwithTRUSTof0.67,providingatentativeindicationthatpeoplearegenuinelymoretrustworthyincountrieswithhighvaluesoftheTRUSTindex.KnackandKeefer(1997)alsoconstructasecondindex,CIVIC,designedtocapturethestrengthofnormsofciviccooperation.Theindexisconstructedbyaveragingacrossfivequestions,addressingtheattitudesoftherespondentstosuchthingsasfraudulentbenefitclaimsandavoidanceoffaresonpublictransport.Perhapssurprisingly,thisindexshowsrelativelylittlevariationacrossOECDcountries,althoughitispositivelycorrelatedwithTRUST.InwhatfollowsIwillconcentrateontheempiricalevidencerelatingtotheTRUSTvariable;KnackandKeefernotethatresultsarebroadlysimilarwhenCIVICisusedinitsplace.Beforeturningtotheevaluationoftheresults,itisworthexaminingthedatainTable1moreclosely.Mostoftheempiricalworkisbasedonsamplesthatcontainasmallnumberoflessdevelopedcountries,aswellasOECDmembers;asaresult,onemightbeconcernedthatpoorercountriesareresponsibleformostoftheidentifyingvariationintheTRUSTvariable.TheevidenceofTable1,however,suggeststhatthereis28substantialvariationinTRUSTacrossOECDmembers.TheindexoftrustisusedbybothKnackandKeefer(1997)andLaPortaetal.(1997).Boththesestudiesreportcross-countryregressionsthatrelateawidevarietyofdependentvariablestotrustandanumberofcontrols.Inmanycases,theresultsshouldberegardedasindicatingtheexistenceofassociations,ratherthanestablishingacausalrelationship.Theevidenceforcausalityisarguablystrongestintheregressionsseekingtoexplaingrowthinoutputperhead.LaPortaetal.(1997)reportthatthetrustindexisweaklyassociatedwithgrowthover1970-93,althoughtheexplanatorypoweroftheirgrowthregressionislowandthesampleincludessomecountriesthatwerecentrallyplannedduringthisperiod.ThissuggeststhatoneshouldbequitecarefulaboutdrawingconclusionsforOECDmembers.KnackandKeeferexcludesocialistcountriesandfocusonashorterperiod,1980-92.Theyfindstrongerresults.Controllingforinitialincomeperhead,ahumancapitalvariable,andtherelativepriceofinvestmentgoods,theyfindthataonestandarddeviationchangeinthetrustindexisassociatedwithachangeinthegrowthrateof0.56ofonestandarddeviation.Inalternativeterms,alevelofTRUSTthatistenpercentagepointshigher(slightlylessthanonestandarddeviation)isassociatedwithanannualgrowthratethatishigherby0.8percentagepoints.28ThescatterplotspresentedinKnackandKeeferarealsoreassuringinthisrespect,astheysuggestthatthepartialcorrelationsbetweengrowth,investmentandTRUSTthatwillbediscussedlaterarenotsimplydrivenbytheinclusionofafewlessdevelopedcountries.28 Asalwaysintheempiricalgrowthfield,oneshouldbecarefulnottoregardthesegrowtheffectsasonesthatwillpersistindefinitely.Itwouldperhapsberatherimplausibletoassertthatcountrieswillgrowatpermanentlydifferentrates,simplybecauseofdifferinglevelsoftrust.ThecorrelationshighlightedbyLaPortaetal.(1997)andKnackandKeefer(1997)arebetterseenasindicationsofapossibleroleforsocialcapitalindeterminingthesteady-statelevelofincome.Inotherwords,changesinsocialcapitalmightaffectgrowthrates,butonlyforatransitionalperiod.Thisqualificationisalsotrueofalmostanyothervariablethatonemightusetoexplaingrowth,anditshouldberememberedthattransitionaleffectsmayeasilybelargeenoughtobeworthconsiderableattention.KnackandKeefercarryoutanumberofrobustnesstests.Wheninfluentialoutliersaredeleted,orgrowthanalysedoverlongerperiods(1960-92and1970-92),thepointestimateofthegrowtheffectisroughlyhalved,butremainsstatisticallysignificant(seetheirTableII).Theydonotethat,overthelongertimespan,theeffectofTRUSTisnotalwaysrobusttotheinclusionofotherexplanatoryvariablesinthegrowthequation.TheevidencesuggeststhattheeffectofTRUSTislargeenoughtobeworthyoffurtherinvestigation.Itisimportanttonote,however,thatresultsaretypicallylessstrongwhenattentionisrestrictedtoasampleofOECDcountries.AlsousingWorldValuesSurveydata,Helliwell(1996)foundanegativeeffectoftrustongrowthinasampleof17OECDmembers.Knack(2000)reportsthatinasamplerestrictedto25OECDmembers,theeffectoftrustisimpreciselymeasured,andthehypothesisthatithasnoeffectcannotberejectedatconventionalsignificancelevels.Thesearequitesmallsamples,soinasenseitisnotsurprisingthattrustisinsignificantwhenattentionisrestrictedtotheOECD.Knack(2000)makestwoadditionalpointsinrelationtotheOECDresults.First,asinKnackandKeefer(1997),thereisevidencethattheeffectoftrustisgreaterinlowincomecountries,basedonaninteractionterminthegrowthregressions.EvenifoneisscepticalthattrustmattersforthehighincomemembersoftheOECD,itmaystillplayanimportantroleinpoorercountrieslikeMexicoandTurkey.Secondly,Knack(2000)reportsastatisticallysignificantandpositivecorrelationbetweeninvestmentandTRUSTwithinanOECDsample,supportingtheideathattrustplayssomeroleevenforrichernations.BothLaPortaetal.(1997)andKnackandKeefer(1997)reportevidenceonotherinterestingassociationsbetweenTRUSTandindicatorsofperformance.LaPortaetal.findstrongpositiveassociationsbetweenTRUSTandanumberofmeasuresofgovernmentperformance,includingtheeffectivenessofthejudiciaryandthequalityofthebureaucracy(theirTable2).KnackandKeeferpresentverysimilarresults(theirTableV).TheyalsoprovidesomeevidencethattheeffectoftrustworksthroughraisingtheshareofinvestmentinGDP.Theseresultsareintriguing,butoneshouldbecarefultoavoidjumpingtostrongconclusionsabouttheimportanceoftrust,orotheraspectsofsocialcapital.Afundamentalproblemisthattheextentoftrustmaywellbedeterminedby,orcorrelatedwith,otheraspectsofsocietiesthatareomittedfromthegrowthregressions.Forinstance,itmaybethatcorruptionorweaklegalenforcementlowerstrustand,forquiteindependentreasons,thegrowthrate.AsKnackandKeefernote,onecouldeven29 tellastoryinwhichtrustisaproductofoptimisminsocietiesthatareperformingwellineconomicterms.Oneobviousvariablethatmightbecorrelatedwithsocialcapitaliseducationalattainment,andthisisparticularlyinterestingfromthepointofviewofthepresentsurvey.LaPortaetal.(1997,p.336)arguethattrusthasapositiveeffectoneducationalachievement,butitshouldbeclearthatcausalitymayrunintheoppositedirection.KnackandKeeferreportastrongcorrelation(r=0.83)betweenTRUSTandanestimateofaverageyearsofschoolingfor1980,andnotethat“educationmaystrengthentrustandcivicnorms,forexample,ifignorancebreedsdistrust,oriflearningreducesuncertaintyaboutthebehaviorofothers,orifstudentsaretaughttobehavecooperatively”(p.1270).Ifweseetrustasendogenoustotheextentandqualityofeducation,wehavethebeginningsofastoryaboutexternalitiestoeducation,ofthekindbrieflydiscussedearlier.4.3ThefutureforsocialcapitalresearchGiventhatinterestingandsuggestiveevidencefortheimportanceofsocialcapitalhasbeencompiledinsoshortatime,furtherresearchonsocialcapitalappearstohaveabrightfuture.Toliveuptothispromise,however,thereareatleasttwopotentiallydifficultquestionsthatwillneedtobeaddressed.Thefirstquestionconcernstheoriginsandformationofsocialcapital;thesecond,theprecisemechanismsbywhichsocialcapital,onceformed,givesrisetoparticularmicroeconomicandmacroeconomicoutcomes.Itshouldbeclearthat,toincorporatetheideasofthesocialcapitalliteratureinpolicyadvice,wewilloftenneedtounderstandhowsocialcapitaliscreated,andhowitmightsometimesbeundermined.Inlinewiththeusualmethodsofeconomists,Glaeser(2000)hasconvincinglyarguedthatweneedtogivemoreattentiontothevalueofsocialcapitalasaresourceforindividuals,aswellasforcommunitiesasawhole.Itseemsunlikelythatsocialcapitalisbestunderstoodassimplyanunintendedby-productofotherdecisions.Withthisinmind,weneedamodelthatcapturestheincentivesofindividualstoformorunderminesocialcapital,andwhichalsoshowshowtheseincentivesareaffectedbypolicy.Withoutsuchamodel,ourknowledgeofpolicyimplicationswillremainincomplete,howeverstrongourintuitionandevidencethatsocialcapitalmatters.Itcanseemthatsocialcapitalresiststheusualmethodsofanalysisofeconomists,giventhatitisusuallyunderstoodtobeapropertyofgroupsratherthanindividuals.TheGlaeserargumentworkswellforthe‘networks’aspectofsocialcapital,sinceparticipationinnetworkscanbemodelledastheoutcomeofindividualinvestmentdecisions;theargumentislessclearlyapplicabletootheraspectsofsocialcapital,suchassocialnorms.Yetevenforsocialnorms,suchasthevalueoftrustworthiness,itispossibletoanalysetheircreationandevolutionintermsofindividualdecisionstoobserve(ornottoobserve)theprevailingnorm.Economistshaverecentlystartedto30 givegreaterattentiontoconstructingmodelsinwhichsocialnormsareendogenous,29anditseemsprobablethatthisworkwillyieldsomevaluableinsights.Asecond,andrelated,questionconcernstheprecisemechanismsbywhichsocialcapital,onceinplace,affectseconomicoutcomes.Again,formalmodellingmaybeuseful.Forexample,ZakandKnack(1999)presentamodelinwhichagentsdividetheirtimebetweenproductionandverifyingtheactionsofthosetheytransactwith.Theirmodelcapturesthesimpleideathatinlowtrustsocieties,someresourcesandtimearedivertedtoverification,andthisresultsinloweroutput.Itwillbeverydifficulttodiscriminatebetweenalternativetheoreticalmodelsusingmacroeconomicdata,andtheprospectsforfurthercross-countryempiricalresearchappearlimited.Studiesbasedattheleveloffirmsorregionsmayultimatelybemoreinformative,andsomeinterestingworkhasalreadystartedtoappear.Guisoetal.(2000)arguethatoneofthebesttestinggroundsfortheimportanceofsocialcapitalmaylieinthefinancialsector,sinceitisherethattrustmaybeespeciallyrelevanttoeconomicactivity.TheystudythiseffectwithinItaly,usingameasureofcivicengagement(essentiallyvoterturnoutincertainelections)asaproxyforsocialcapital,asinPutnam(1993).Usinglargesamplesofhouseholdsandfirms,theyfindthattheirmeasureofcivicengagementhelpsexplainvariationinfinancialpracticesacrossItalianregions,evenwhencontrollingfordifferentlevelsofdevelopment.Suchstudiesarelikelytoplayanincreasinglyimportantroleinthewiderdebateontheimportanceofsocialcapital.Scepticswillremainunconvincedbytheeconomicimportanceoftrustandotheraspectsofsocieties(networks,norms,participation)untilwehaveamorecompleteanddetailedstorydescribingtheirconnectiontoeconomicoutcomes,supportedbyreliableevidence.5.SummaryandconclusionsThissectionroundsoffthepaperwithadiscussionofhowtheseareasofresearchmightinformfuturepolicy.Iwilllookateducationfirst,butcareisneededhere.Griliches(1997,p.S339)notesthatforacademiceconomists,anemphasisontheimportanceofeducationforeconomicgrowth“maybesomewhatself-serving”andoccasionallyintheliteratureonedoescomeacrossapaperwhichechoestothesoundofgrindingaxes.Thisisparticularlytrueinreadingopposingassessmentsofthesignallingargument,wherethelackofreliableevidenceseemstoencourage,ratherthandiscourage,strongviews.Inassessingtheempiricalevidenceforproductivitybenefitsofeducation,itisquitepossiblethatanoveralljudgementisfrequentlycontaminatedbyakeenawarenessofwiderbenefitsofeducationthatarenotcapturedineconomicdata.Afterall,onecouldprobablyconstructaviablecaseformucheducationalexpenditureentirelybasedonitsimplicationsforpersonaldevelopment,independentofanyproductivityeffects.ItisworthquotingWeiss(1995,p.151):29ManyreferencestoresearchinthisareacanbefoundinZakandKnack(1999).31 Educationdoesnothavetobejustifiedsolelyonthebasisofitseffectonlabourproductivity.ThiswascertainlynottheargumentgivenbyPlatoordeTocquevilleandneednotbeours.Studentsarenottaughtcivics,orart,ormusicsolelyinordertoimprovetheirlabourproductivity,butrathertoenrichtheirlivesandmakethembettercitizens.Mosteconomists,appropriatelyenoughforpractitionersofthe‘dismalscience’,haveconcentratedonexaminingarathermorenarrowcaseforeducation,intermsofitscontributiontoproductivitygrowth.Aswehaveseen,theweightofevidencepointstosignificantproductivityeffects,butthedegreeofuncertaintyislarge,andevenalowerboundissurprisinglydifficulttoestablish.Theevidencefromlaboureconomicshasthegreatestweightofexperience,timeandacademicfirepowerbehindit,andthissuggeststhatitwouldbeamistaketosummarisethemacroeconomicresultsinisolation.Althoughareconciliationofthesetwoliteraturesisinitsearlystages,thecorrelationacrosscountriesbetweenmeasuresofhumancapitalandgrowthisarguablyrobustenoughtosupportthebeliefthatearningsfunctionspickupgenuineproductivityeffects,andnotsimplytheeffectsofsignallingoromittedcharacteristics.Thatisreassuring,butitleavesmanyquestionsopenforpolicy-makers.Thereislikelytobepervasiveheterogeneityinratesofreturnacrossindividuals,letaloneacrosscountries.Agreaterunderstandingofthepatternofheterogeneitywillleadtobetterpolicydecisions,butonthissubjectthemacroeconomicliteraturesurveyedhereisnecessarilysilent.Evidencecompiledbylaboureconomistswillbefarmoreusefulin30thisrespect.Otherlimitationsofthemacroeconomicevidenceareworthnoting.Growthregressionsarebestthoughtofaspickingupanaverageeffectofschooling,andshouldcertainlynotbeusedtoconcludethateveryOECDmemberiscurrentlyunder-providingeducation.Indeed,theresultsfromgrowthaccountingexercisessuggestthat,althoughincreasesineducationalprovisioncanyieldaworthwhileincreaseinthegrowthrate,oneshouldnotnecessarilyexpectaneffectthatislargerelativetocurrentratesofgrowth.Forpolicy-makerswhowishtoraisethegrowthrate,policyoneducationremainsanaturalplacetolook,butitisbynomeansapanacea.Notonlythat,onecannotaltogetherdismissthepossibilityof‘overeducation’insomecountries,atsomelevelsofeducation.Forexample,itisplausiblethattherearesomepeopleforwhomconventionalacademiceducationbeyondacertainpointisultimatelylessusefulthanvocationaltraining.Thisdrawsattention,onceagain,tothewayinwhichtheunbalancednatureoftheexistingevidencemaymislead.Weneedtolearnmoreabouttherelativemeritsofschoolingandtrainingfordifferentindividuals.Thiswillrequireestimatesofhowreturnstoschoolingvarywithpersonalcharacteristics,andshouldalsoinvolvesomeconsiderationofwiderbenefits.30AspecialissueofthejournalLabourEconomics(November1999)includesstudiesthatmeasurethereturnstoschoolingforavarietyofOECDmembercountries,andthusshedslightonthepossibleheterogeneityacrosscountries.32 Inthinkingbroadlyabouttheovereducationquestion,itisinterestingtoconsidertheevolutionofeducationalwagedifferentialssincethelate1970s.Eventhoughtherelativesupplyofskilledlabourhasincreased,therehasbeenasubstantialandwell-documentedincreaseineducationalwagedifferentialsintheUKandtheUSA,with31lesspronouncedchangesinotherOECDcountries.ItseemsdifficulttoexplaintheevidencefortheUKandtheUSwithoutadominantroleforashiftintherelativedemandsfordifferenttypesoflabour,favouringthemoreeducated.Muchresearchhasfocusedontheoriginsofthechangeinrelativelabourdemand,butforpolicy-makersanequallyimportantquestioniswhetherthischangeisgeneratingarisingreturnmainlytoeducation,ortoothercharacteristicssuchasinnateabilityorinitiative.Clearlythepolicyimplicationsareverydifferentunderthealternativescenarios,yetdisentanglingthetwoeffectsisdifficult.Existingresearchoftenfindsthatitisthereturntoabilitywhichisrising,buttheworkofCawleyetal.(1998)suggeststhat,duetosomeimportantidentificationproblems,theseresultsarenotrobusttosmallchangesinassumptions.Itisalsothecasethatovereducationmaytaketimetoappearinthedata.Oneproblemhereisthataverageattainmenttypicallyevolvesonlyslowly,andsomeasuredreturnstoeducationwillalsochangeonlyslowly,andarenotnecessarilyinformativeonthedesirabilityofcurrentprovision.Thismeansthatevidenceoncurrentwagedifferentialsneedstobesupplementedbyotherapproaches,includingthosediscussedinHarmonetal.(2000).Therecentshiftsindifferentialsalsoremindusthatpolicyoneducationhasdistributionalconsequences(Topel1997).Giventhattradedoesnotseemtoequalisefactorpricesacrosscountries,anyincreaseintherelativesupplyofskilledlabourislikelytolowerthewagepremiumforthepossessionofskills.Inturnthiscouldmake32animportantcontributiontoreducingincomeinequality.Tosummarise,canwejustifythemassiveamountofresourcesallocatedtoeducationbyOECDmembercountries,around$1550billionintotaleachyear?Ontheavailableevidence,includingrecentchangesinwagedispersion,theargumentsforcuttingbackonthisprovisionseemratherweak.Indecidingifprovisionshouldbeexpanded,perhapsthekeyopenquestionisthevalidityofthesignallingarguments.Moreevidenceonthesignallingdebatewouldbeextremelyhelpfulinjudgingthebenefitsofexpandinghighereducation,oneofthemainchangesinprovisionwithintheOECDsincethe1960s.Inexploringsomeofthedetailsofsucharguments,empiricalevidenceisnottheonlywayforward.Theoryandcalibrationexercisesmayalsoshedlightontheseissues.Anexampleistheinterestingimplicationofnewgrowththeorythatindividualsmayunder-investineducation,becausethosewholatergointoresearchcareersdonotcaptureallthebenefitsofthenewideasthattheyhelptocreate.Thisprovidesthe31SeeforinstanceKatzandAutor(1999),p.1501-1503.EvidenceonrecenttrendsinwagedispersionmoregenerallycanbefoundinOECD(1996,ch.3).32Onehastobecarefulinmakingthisargument,eveninasimplemodelwithjusttwotypesoflabour.Inequalitydependsnotonlyontheskillpremium,butalsoontherelativesuppliesofskilledandunskilledlabour.33 beginningsofanargumentforsubsidisingeducationinengineeringandscience,atleastatthoselevels(perhapsPhDs,orpost-docs)whereahighproportionofindividualssubsequentlygointoresearchanddevelopmentactivity.Romer(2000)hasrecentlypresentedspecificpolicyproposalsalongtheselines.Theliteratureonsocialcapitalandgrowthisatanearlierstagethanthemacroeconomicevidenceoneducation,andthepolicyimplicationsarelessclear.Indeed,oneweaknessofthesocialcapitalliterature,atleastinrelationtorichercountries,isthatitiscurrentlydifficulttoseewhatpolicyconclusionscouldeverbedrawn.Whatcanapolicy-makerinMexicoorTurkeyactuallydo,confrontedwiththeevidencefromtheWorldValuesSurveythattheygovernalow-trustsociety?Standardrecommendations,suchasattemptingtoeliminatecorruptionandimprovethelegalsystem,arenothingnew,andmakegoodsensequiteindependentlyofanyemphasisonsocialcapital.Perhapsthebestanswerliesindrawingananalogywiththeintroductionofhumancapitaltheoryintoeconomics.Initsearlystages,asLucas(1988)makesclear,humancapitalseemedaratheretherealconcept,andpresumablyonewithlittleimmediatemessageforeducationpolicies.Workonsocialcapitalisstillinitsearlystages,andaswelearnmoreaboutwhatitis,whereitcomesfrom,andwhatitdoes,theremayultimatelybeimplicationsandconclusionsthatleaveoursuccessorswiserinwaysthatwecanonlyguessat.ReferencesAbramovitz,M.(1986).Catchingup,forgingahead,andfallingbehind.JournalofEconomicHistory,46,385-406.Acemoglu,D.(1996).Amicrofoundationforsocialincreasingreturnsinhumancapitalaccumulation.QuarterlyJournalofEconomics,111,779-804.Acemoglu,D.(1997).Trainingandinnovationinanimperfectlabourmarket.ReviewofEconomicStudies,64,445-464.Acemoglu,D.andAngrist,J.(1999).Howlargearethesocialreturnstoeducation?Evidencefromcompulsoryschoolinglaws.NBERworkingpaperno.7444.Aghion,P.andHowitt,P.(1998).Endogenousgrowththeory.MITPress,Cambridge.Angrist,J.D.andKrueger,A.B.(1991).Doescompulsoryschoolattendanceaffectschoolingandearnings?QuarterlyJournalofEconomics,106,979-1014.Ark,B.vanandPilat,D.(1993).ProductivitylevelsinGermany,Japan,andtheUnitedStates:differencesandcauses.BrookingsPapersonEconomicActivity:Microeconomics2,WashingtonD.C.,December.Arrow,K.J.(1973).Highereducationasafilter.JournalofPublicEconomics,2,193-216.34 Ashenfelter,O.,Harmon,C.andOosterbeek,H.(1999).Areviewofestimatesoftheschoolings/earningsrelationship,withtestsforpublicationbias.LabourEconomics,6(4),453-470.Barro,R.J.(1991).Economicgrowthinacrosssectionofcountries.QuarterlyJournalofEconomics,106(2),407-443.Barro,R.J.andLee,J.-W.(1993).Internationalcomparisonsofeducationalattainment.JournalofMonetaryEconomics,32,363-394.Barro,R.J.andSala-i-Martin,X.(1995).Economicgrowth.McGraw-Hill,NewYork.Bassanini,A.andScarpetta,S.(2001).DoeshumancapitalmatterforgrowthinOECDcountries?Evidencefrompooledmean-groupestimates.OECDEconomicsDepartmentworkingpaperno.282.Behrman,J.R.andStacey,N.(1997).Thesocialbenefitsofeducation.UniversityofMichiganPress,AnnArbor.Benhabib,J.andSpiegel,M.M.(1994).Theroleofhumancapitalineconomicdevelopment:evidencefromaggregatecross-countrydata.JournalofMonetaryEconomics,34(2),143-173.Bils,M.andKlenow,P.J.(2000).Doesschoolingcausegrowth?AmericanEconomicReview,90(5),1160-1183.Blanchflower,D.G.andOswald,A.J.(2000).Well-beingovertimeinBritainandtheUSA.NBERworkingpaperno.7487.Broadberry,S.N.andWagner,K.(1996).Humancapitalandproductivityinmanufacturingduringthetwentiethcentury:Britain,GermanyandtheUnitedStates.InVanArk,B.andCrafts,N.(eds.)Quantitativeaspectsofpost-warEuropeaneconomicgrowth.CambridgeUniversityPress,Cambridge.Cameron,G.,Proudman,J.andRedding,S.(1998).Productivityconvergenceandinternationalopenness.InJ.ProudmanandS.Redding(eds.)Opennessandgrowth.BankofEngland,London.Cannon,E.(2000).Humancapital:levelversusgrowtheffects.OxfordEconomicPapers,52,670-676.Card,D.(1999).Thecausaleffectofeducationonearnings.InO.C.AshenfelterandD.Card(eds.)HandbookofLaborEconomics,Vol.3A,North-Holland,Amsterdam.35 Cawley,J.,Heckman,J.andVytlacil,E.(1998).Cognitiveabilityandtherisingreturntoeducation.NBERworkingpaperno.6388.Ciccone,A.,Peri,G.andAlmond,D.(1999).Capital,wagesandgrowth:theoryandevidence.CEPRDiscussionPaperno.2199.delaFuente,A.andDomenech,R.(2000).Humancapitalingrowthregressions:howmuchdifferencedoesdataqualitymake?Manuscript,CSIC,CampusdelaUniversidadAutonomadeBarcelona.Denison,E.F.(1967).Whygrowthratesdiffer.BrookingsInstitution,WashingtonDC.Engelbrecht,H.-J.(1997).InternationalR&Dspillovers,humancapitalandproductivityinOECDeconomies:anempiricalinvestigation.EuropeanEconomicReview,41,1479-1488.Englander,A.S.andGurney,A.(1994a).Medium-termdeterminantsofOECDproductivity.OECDEconomicStudies,22,49-109.Englander,A.S.andGurney,A.(1994b).OECDproductivitygrowth:medium-termtrends.OECDEconomicStudies,22,111-129.Finegold,D.andSoskice,D.(1988).ThefailureoftraininginBritain:analysisandprescription.OxfordReviewofEconomicPolicy,4,21-53.Gemmell,N.(1996).Evaluatingtheimpactsofhumancapitalstocksandaccumulationoneconomicgrowth:somenewevidence.OxfordBulletinofEconomicsandStatistics,58(1),9-28.Glaeser,E.(2000).Theformationofsocialcapital.Manuscript,HarvardUniversity,March.Griliches,Z.(1992).ThesearchforR&Dspillovers.ScandinavianJournalofEconomics,94,S29-S47.Griliches,Z.(1997).Education,humancapital,andgrowth:apersonalperspective.JournalofLaborEconomics,15(1),S330-S344.Guiso,L.,Sapienza,P.andZingales,L.(2000).Theroleofsocialcapitalinfinancialdevelopment.NBERworkingpaperno.7563.Hanushek,E.A.andKimko,D.D.(2000).Schooling,laborforcequality,andthegrowthofnations.AmericanEconomicReview,90(5),1184-1208.Harmon,C.,Oosterbeek,H.andWalker,I.(2000).Thereturnstoeducation:areviewofevidence,issuesanddeficienciesintheliterature.CentrefortheEconomicsofEducation,LondonSchoolofEconomics,December.36 Helliwell,J.(1996).EconomicgrowthandsocialcapitalinAsia.NBERworkingpaperno.5470.Jacobs,J.(1969).Theeconomyofcities.RandomHouse,NewYork.Jones,C.I.(1995).R&D-basedmodelsofeconomicgrowth.JournalofPoliticalEconomy,103,759-784.Jorgenson,D.W.,Gollop,F.M.andFraumeni,BarbaraM.(1987).ProductivityandUSeconomicgrowth.HarvardUniversityPress,Cambridge.Jorgenson,D.andYip,E.(1999).Whateverhappenedtoproductivitygrowth?Manuscript,HarvardUniversity,June.Judson,R.(1998).Economicgrowthandinvestmentineducation:howallocationmatters.JournalofEconomicGrowth,3,337-359.Katz,L.F.andAutor,D.H.(1999).Changesinthewagestructureandearningsinequality.InO.C.AshenfelterandD.Card(eds.)HandbookofLaborEconomics,Vol.3A,North-Holland,Amsterdam.Knack,S.(2000).Trust,associationallifeandeconomicperformanceintheOECD.Manuscript,TheWorldBank.Knack,S.andKeefer,P.(1995).Institutionsandeconomicperformance:cross-countrytestsusingalternativeinstitutionalmeasures.EconomicsandPolitics,7(3),207-227.Knack,S.andKeefer,P.(1997).Doessocialcapitalhaveaneconomicpayoff?Across-countryinvestigation.QuarterlyJournalofEconomics,112(4),1251-1288.Krueger,A.B.andLindahl,M.(1999).EducationforgrowthinSwedenandtheworld.NBERworkingpaperno.7190.LaPorta,R.,Lopez-de-Silanes,F.,Shleifer,A.andVishny,R.(1997).Trustinlargeorganizations.AmericanEconomicReview,87(2),333-338.Levine,R.andRenelt,D.(1992).Asensitivityanalysisofcross-countrygrowthregressions.AmericanEconomicReview,82(4),942-963.Lucas,R.E.(1988).Onthemechanicsofeconomicdevelopment.JournalofMonetaryEconomics,22,3-42.Lucas,R.E.(1990).Whydoesn’tcapitalflowfromrichtopoorcountries?AmericanEconomicReview,80(2),92-96.Maddison,A.(1987).Growthandslowdowninadvancedcapitalisteconomies:techniquesofquantitativeassessment.JournalofEconomicLiterature,25,649-698.37 Maddison,A.(1991).Dynamicforcesincapitalistdevelopment.OxfordUniversityPress,Oxford.Mankiw,N.G.,Romer,D.andWeil,D.(1992).Acontributiontotheempiricsofeconomicgrowth.QuarterlyJournalofEconomics,107,407-437.Masters,A.M.(1998).Efficiencyofinvestmentinhumanandphysicalcapitalinamodelofbilateralsearchandbargaining.InternationalEconomicReview,39,477-494.Mincer,J.(1974).Schooling,experienceandearnings.ColumbiaUniversityPress,NY.Mincer,J.(1996).Economicdevelopment,growthofhumancapital,andthedynamicsofthewagestructure.JournalofEconomicGrowth,1(1),29-48.Nonneman,W.andVanhoudt,P.(1996).AfurtheraugmentationoftheSolowmodelandtheempiricsofeconomicgrowthforOECDcountries.QuarterlyJournalofEconomics,111,943-953.OECD(1996).Employmentoutlook.OECD,Paris.OECD(1998).Humancapitalinvestment:aninternationalcomparison.CentreforEducationalResearchandInnovation,OECD,Paris.OECD(2000a).Educationataglance.OECD,Paris.OECD(2000b).OECDEconomicOutlook68.OECD,Paris.Pritchett,L.(1996).Wherehasalltheeducationgone?WorldBankPolicyResearchDepartmentworkingpaperno.1581.Putnam,R.(1993).Makingdemocracywork.PrincetonUniversityPress,Princeton.Quiggin,J.(1999).HumancapitaltheoryandeducationpolicyinAustralia.AustralianEconomicReview,32(2),130-144.Rauch,J.(1993).Productivitygainsfromgeographicconcentrationofhumancapital:evidencefromthecities.JournalofUrbanEconomics,34(3),380-400.Redding,S.(1996).Thelow-skill,low-qualitytrap:strategiccomplementaritiesbetweenhumancapitalandR&D.EconomicJournal,106,458-470.Romer,P.M.(1990).Endogenoustechnologicalchange.JournalofPoliticalEconomy,98(5),S71-S102.Romer,P.M.(2000).Shouldthegovernmentsubsidizesupplyordemandinthemarketforscientistsandengineers?NBERworkingpaperno.7723.38 Rustichini,A.andSchmitz,J.A.(1991).Researchandimitationinlong-rungrowth.JournalofMonetaryEconomics,27(2),271-292.Scarpetta,S.,Bassanini,A.,Pilat,D.andSchreyer,P.(2000).EconomicgrowthintheOECDarea:recenttrendsattheaggregateandsectorallevel.OECDEconomicsDepartmentworkingpaperno.248.Schuller,T.(2000).Thecomplementaryrolesofhumanandsocialcapital.Manuscript,BirkbeckCollege,UniversityofLondon,March.Sianesi,B.andVanReenen,J.(2000).Thereturnstoeducation:areviewofthemacroeconomicliterature.CentrefortheEconomicsofEducation,LondonSchoolofEconomics,DiscussionPaperNo.6.Spence,A.M.(1973).Jobmarketsignaling.QuarterlyJournalofEconomics,87,355-374.Steedman,H.(1996).Measuringthequalityofeducationaloutputs:anote.CEPDiscussionPaperno.302,LondonSchoolofEconomics.Stiglitz,J.E.(1975).Thetheoryof‘screening’,education,andthedistributionofincome.AmericanEconomicReview,65,283-300.Temple,J.R.W.(1998).EquipmentinvestmentandtheSolowmodel.OxfordEconomicPapers,50(1),39-62.Temple,J.R.W.(1999a).Thenewgrowthevidence.JournalofEconomicLiterature,37(1),112-156.Temple,J.R.W.(1999b).Apositiveeffectofhumancapitalongrowth.EconomicsLetters,65(1),131-134.Temple,J.R.W.(2001).Generalizationsthataren’t?Evidenceoneducationandgrowth.EuropeanEconomicReview,45(4-6),905-918.Temple,J.R.W.andJohnson,P.A.(1998).Socialcapabilityandeconomicgrowth.QuarterlyJournalofEconomics,113,965-990.Topel,R.(1997).Factorproportionsandrelativewages:thesupply-sidedeterminantsofwageinequality.JournalofEconomicPerspectives,11(2),55-74.Topel,R.(1999).Labormarketsandeconomicgrowth.InO.C.AshenfelterandD.Card(eds.)HandbookofLaborEconomics,Vol.3C,North-Holland,Amsterdam.Uzawa,H.(1965).Optimaltechnicalchangeinanaggregativemodelofeconomicgrowth.InternationalEconomicReview,6,18-31.39 VasudevaMurthy,N.R.andChien,I.S.(1997).TheempiricsofeconomicgrowthforOECDcountries:somenewfindings.EconomicsLetters,55,425-429.Weiss,A.(1995).Humancapitalvs.signallingexplanationsofwages.JournalofEconomicPerspectives,9(4),133-154.Woessmann,L.(2000).Specifyinghumancapital:areview,someextensions,anddevelopmenteffects.TheKielInstituteofWorldEconomicsworkingpaperno.1007.Wolfe,B.andHaveman,R.(2000).Accountingforthesocialandnon-marketbenefitsofeducation.Manuscript,UniversityofWisconsin-Madison.Woolcock,M.(1998).Socialcapitalandeconomicdevelopment:towardatheoreticalsynthesisandpolicyframework.TheoryandSociety,27,151-208.Woolcock,M.(2000).Theplaceofsocialcapitalinunderstandingsocialandeconomicoutcomes.Manuscript,TheWorldBank.Young,A.(1995).Thetyrannyofnumbers:confrontingthestatisticalrealitiesoftheEastAsiangrowthexperience.QuarterlyJournalofEconomics,110(3),641-680.Zak,P.J.andKnack,S.(1999).Trustandgrowth.Manuscript,ClaremontGraduateUniversity,August.40