DOESSOCIALCAPITALHAVEANECONOMICPAYOFF?ACROSS-COUNTRYINVESTIGATION*NACK - PDF document

DOESSOCIALCAPITALHAVEANECONOMICPAYOFF?ACROSS-COUNTRYINVESTIGATION*NACKANDThispaperpresentsevidencethatªsocialcapitalºmattersformeasurableeco-nomicperformance,usingindicatorsoftrustandcivicnormsfromtheWorldVal-uesSurveysforasampleof29marketeconomies.MembershipsinformalgroupsÐPutnam'smeasureofsocialcapitalÐisnotassociatedwithtrustorwithimprovedeconomicperformance.We®ndtrustandcivicnormsarestrongerinnationswithhigherandmoreequalincomes,withinstitutionsthatrestrainpredatoryactionsofchiefexecutives,andwithbetter-educatedandethnicallyho-mogeneouspopulations.I.IThenotionofsocialcapitalhasattractedgreatacademicandjournalisticattention,particularlywiththepublicationofPut-nam'sMakingDemocracyWork[1993],inwhichtheconceptisusedtoexplaindifferencesintheeconomicandgovernmentper-formanceofnorthernandsouthernItaly.Weexploreinthispaperthefollowingissuesrelatedtosocialcapitalandeconomici)therelationshipbetweeninterpersonaltrust,normsofciviccooperation,andeconomicperformance,andsomeofthepolicyandotherlinksthroughwhichthesedimen-sionsofsocialcapitalmayhaveeconomiceffects;ii)thecon¯ictinghypothesesofPutnam[1993]andOlson[1982],ontherelationshipbetweenassociationalactivityandgrowth;andiii)thedeterminantsoftrustandnormsofciviccooperation,includinglevelsofassociationalactivityandformalTrust,cooperativenorms,andassociationswithingroupseachfallwithintheelasticde®nitionsthatmostscholarshave1997bythePresidentandFellowsofHarvardCollegeandtheMassachusettsInstituteofTechnology.TheQuarterlyJournalofEconomics,November1997.*ValuablecomomentswereprovidedbyOlivierBlanchard,ChristopherClague,LantPritchett,AndreiShleifer,andparticipantsattheMay1997Clare-montConferenceonthePoliticalEconomyofGrowth.ResearchassistancewasablyprovidedbyGaryAndersonandAstaZviniene.WorldValuesSurveydataareavailablethroughtheInter-UniversityConsortiumforPoliticalandSocialResearch.AnalysesandInterpretationofthedata,aswellasanyremaininger-rors,arethesoleresponsibilityoftheauthors.ComputertimewasprovidedbytheUniversityofMarylandComputerScienceCenter.Knack'scontributiontothisresearchwasgenerouslysupportedbytheIRISCenter,UniversityofMaryland,underagrantfromtheU.S.AgencyforInternationalDevelopment. appliedtothetermsocialcapital.Coleman[1990,pp.300±301]writesthatªauthorityrelations,relationsoftrust,andconsen-sualallocationsofrightswhichestablishnormsºcanbeviewedasresourcesforindividuals,notingthatLoury[1977]introducedthetermªsocialcapitalºtodescribetheseresources.FollowingGranovetter[1973],Putnampointstothepotentialimportanceofweaktiesacrosskinshipgroups.BothColemanandPutnamrefertotrustandnormsofcivic-mindedbehaviorasothermanifesta-tionsofsocialcapital.Ouranalysisarrivesatthreemajorconclusions.First,trustandciviccooperationareassociatedwithstrongereconomicper-formance.Second,associationalactivityisnotcorrelatedwitheconomicperformanceÐcontrarytoPutnam's[1993]®ndingsacrossItalianregions.Third,we®ndthattrustandnormsofciviccooperationarestrongerincountrieswithformalinstitutionsthateffectivelyprotectpropertyandcontractrights,andincoun-triesthatarelesspolarizedalonglinesofclassorethnicity.II.HEconomicactivitiesthatrequiresomeagentstorelyonthefutureactionsofothersareaccomplishedatlowercostinhigher-trustenvironments.AccordingtoArrow[1972,p.357],ªVirtuallyeverycommercialtransactionhaswithinitselfanelementoftrust,certainlyanytransactionconductedoveraperiodoftime.Itcanbeplausiblyarguedthatmuchoftheeconomicbackward-nessintheworldcanbeexplainedbythelackofmutualcon®-dence.ºTrust-sensitivetransactionsincludethoseinwhichgoodsandservicesareprovidedinexchangeforfuturepayment,em-ploymentcontractsinwhichmanagersrelyonemployeestoac-complishtasksthataredif®culttomonitor,andinvestmentsandsavingsdecisionsthatrelyonassurancesbygovernmentsorbanksthattheywillnotexpropriatetheseassets.Individualsinhigher-trustsocietiesspendlesstoprotectthemselvesfrombeingexploitedineconomictransactions.Writtencontractsarelesslikelytobeneeded,andtheydonothavetospecifyeverypossiblecontingency.Litigationmaybelessfrequent.Individualsinhigh-trustsocietiesarealsolikelytodivertfewerresourcestopro-tectingthemselvesÐthroughtaxpayments,bribes,orprivatesecurityservicesandequipmentÐfromunlawful(criminal)violationsoftheirpropertyrights.Lowtrustcanalsodiscourageinnovation.Ifentrepreneursmustdevotemoretimetomonitor-QUARTERLYJOURNALOFECONOMICS ingpossiblemalfeasancebypartners,employees,andsuppliers,theyhavelesstimetodevotetoinnovationinnewproductsorSocietiescharacterizedbyhighlevelsoftrustarealsolessdependentonformalinstitutionstoenforceagreements.Informalcreditmarketsdependentonstronginterpersonaltrustcanfacili-tateinvestmentwherethereisnowell-developedformalsystemof®nancialintermediation,orwherelackofassetslimitsaccesstobankcredit.Interpersonaltrustcanalsoprovideanimperfectsubstituteforgovernment-backedpropertyrightsorcontracten-forcementwheregovernmentsareunableorunwillingtopro-videthem.Governmentof®cialsinsocietieswithhighertrustmaybeperceivedasmoretrustworthy,andtheirpolicypronouncementsasthusbeingmorecredible.Totheextentthatthisistrue,trustalsotriggersgreaterinvestmentandothereconomicactivity.Promisesbycentralbankersthattheywillnotraiseinterestrates,assurancesbyministersof®nancethatanominalex-changerateanchoris®xedinstone,andguaranteesthattaxleg-islationwillnotberapidlyamendedarealllikelytobemorecredibleinsocietieswherepeopletrusteachothermore.Asacon-sequence,insuchsocietiespeopleadoptmoreappropriatehori-zonsinmakinginvestmentdecisions,andchooseproductiontechnologiesthatareoptimaloverthelong,ratherthanshort,Trustingsocietiesnotonlyhavestrongerincentivestoinno-vateandtoaccumulatephysicalcapital,butarealsolikelytohavehigherreturnstoaccumulationofhumancapital.Wheretrustimprovesaccesstocreditforthepoor,enrollmentinsecond-aryeducationÐwhich,unlikeprimaryeducation,hasahighcostinforgoneincomeÐmaybehigher[GalorandZeira1993].AsshowninSectionIV,trustandcivicinvolvementarelinkedtobetterperformanceofgovernmentinstitutions,includingpubliclyprovidededucation[Putnam1993;LaPortaetal.1997;Coleman1988].Higher-qualityschoolsincreasethereturntoeducation.Wheretrustfacilitatestheenforcementofcontracts,thereturntospecializededucationwillincrease.Finally,inlow-trustsocieties,hiringdecisionswillbein¯uencedmorebytrustworthypersonalattributesofapplicants,suchasbloodtiesorpersonalknowledge,DOESSOCIALCAPITALHAVEAPAYOFF?1.ThisassumptioniscontainedinClague's[1993]modelofªruleobedienceºandorganizationalperformance. andlessbyeducationalcredentials,thaninhigh-trustsocieties,reducingthereturnstoacquisitionofeducationalcredentialsinlow-trustsocieties.Normsofciviccooperationcanbelinkedwitheconomicout-comesinsomeofthesamewaysastrust.Cooperativenormsactasconstraintsonnarrowself-interest,leadingindividualstocon-tributetotheprovisionofpublicgoodsofvariouskinds.Internal(e.g.,guilt)andexternal(e.g.,shameandostracism)sanctionsassociatedwithnormsalterthecostsandbene®tsofcooperatinganddefectinginprisoner'sdilemmas[Coleman1990].Formanycollectiveactionproblems,normsleadingtocoop-erativesolutionsimposeseriousnegativeexternalitiesonnon-players.Forexample,intheclassicprisoner'sdilemmagame,thepayoffstotwocooperatingcriminalsarehigherthaniftheybothdefect.ªCivicnorms,ºsuchasthenormagainstlittering,arede-®nedhereasthosethatresolveprisoner'sdilemmaswithoutim-posingsubstantialexternalcostsonotherparties(unlikecartelarrangements,forexample).Theyimproveallocativeef®ciencyfromasocietalstandpoint:thetotalbene®tstosocietyfromat-tainingcooperativeoutcomesfarexceedthetotalcosts.Totheextentthatcivicnormseffectivelyconstrainopportunism,thecostsofmonitoringandenforcingcontractsarelikelytobelower,raisingthepayoffstomanyinvestmentsandothereconomicInadditiontothemoredirecteffectsoneconomicactivityoutlinedabove,trustandcivicnormsmayimproveeconomicout-comesindirectly,throughpoliticalchannels.Theymayimprovegovernmentalperformanceandthequalityofeconomicpolicies,byaffectingthelevelandcharacterofpoliticalparticipation.Knowledgeofpoliticsandpublicaffairsbylargenumbersofciti-zens,andtheirparticipation,areimportantpotentialchecksontheabilityofpoliticiansandbureaucratstoenrichthemselvesornarrowintereststhattheyarealliedwith.Butself-interestedcitizenswillrationallydeclinetovoteortoacquireinformationabouttheperformanceofof®cials.Civicnormshelpvoters(prin-cipals)overcomethecollectiveactionprobleminmonitoringof®cials(agents).Putnam[1993]hasshownthatregionalgovern-mentsinthemore-trusting,morecivic-mindednorthernandcen-tralpartsofItalyprovidepublicservicesmoreeffectivelythandothoseintheless-trusting,lesscivicsouth.Moreover,citizen-initiatedcontactswithgovernmentof®cialsinthesouthtendtoinvolveissuesofnarrowlypersonalconcerns,whilecontactsinQUARTERLYJOURNALOFECONOMICS themoretrustingregionstendtoinvolvelargerissueswithimpli-cationsforthewelfareoftheregionasawhole.SurveyevidencefromtheUnitedStatesisconsistentwiththese®ndings.Amongrespondentsinthe1992AmericanNationalElectionsStudy(NES),interpersonaltrustisasigni®cantpredictorofvariousparticipatoryattitudesandbehaviors.Controllingforincomeandeducation,trustisassociatedwithan8.6percentage-pointin-creaseintheprobabilityofvoting,andwithsimilarincreasesininterestinpoliticalcampaignsandinpublicaffairsgenerally,andwithagreementthatvotingisacivicduty.III.MRUSTANDInacritiqueofFukuyama[1995],Solow[1995]arguesthatifsocialcapitalistobemorethanaªbuzzwordºitsstockªshouldsomehowbemeasurable,eveninexactly,ºbutªmeasurementseemsveryfaraway.ºInthisstudy,weusesurveyindicatorsthatarenodoubtinexactÐduetotranslationdif®culties,samplinger-ror,andresponsebiasÐbutwhichproducevaluesthatareconsis-tentwithdatafromindependentsources(asdescribedbelow).TheWorldValuesSurveyscontainsurveydataonthousandsofrespondentsfrom29marketeconomies:21inthe1981sur-veys,and28inthe1990±1991surveys,with29representedinatleastoneofthesetwosurveywaves.SomegroupsÐforexample,city-dwellersandthebetter-educatedÐareoversampledinsomecountries[Inglehart1994].Asacorrection,weusetheweightvariableprovidedinthedataincomputingcountry-levelmeans.Higher-statusgroupsstilltendtobeoverrepresented,particu-larlyinthelessdevelopedcountries,evenwithuseoftheweightvariable[Inglehart1994].ThisproblemshouldhavetheeffectDOESSOCIALCAPITALHAVEAPAYOFF?2.Politicalparticipationcanbemotivatedbygroupsbasedonclass,ethnic,orothertiesseekingtousegovernmentfortheirownendsattheexpenseofthelargersociety.Putnam[1993]implicitlyassumesthattheseinef®cienteffectsofenhancedparticipationareoutweighedbytheef®cienteffects.Ourmeasuresoftrustandcivicnorms,asexplainedbelow,appeartore¯ectattitudestowardcoop-erationmorebroadly,notonlywiththoseoneassociatesoridenti®eswith.3.SeeKnack[1992]forrelatedevidence.4.BothªBritainºandNorthernIrelandareincludedinthesurveys.WetreatBritain'svaluesasrepresentingtheUnitedKingdom(NorthernIrelandaccountsforlessthan3percentofthelatter'spopulation).Manynonmarketeconomieswereincludedinthe1990±1991surveys.FollowingBarro[1991]andothers,wehavenotincludedtheminouranalyses,becauseofthelackofavailabledataoneducationandothervariables,andbecauseeconomicperformanceisdrivenbydifferentprocessesinnonmarketeconomies.TheEasternEuropeannationsandex-Sovietrepublicstendtohavelowvaluesforourtrustmeasure;China's60.3percentranksamongthehighest,however. ofattenuatingthevariationinourmeasuresoftrustandciviccooperationÐwhichtendtobepositivelycorrelatedwithincomeandeducationlevelsÐmakingitmoredif®culttorejectnullhypothesesinvolvingthesevariables.Thequestionusedtoassesstheleveloftrustinasocietyis:ªGenerallyspeaking,wouldyousaythatmostpeoplecanbetrusted,orthatyoucan'tbetoocarefulindealingwithpeople?ºOurtrustindicator(TRUST)isthepercentageofrespondentsineachnationreplyingªmostpeoplecanbetrustedº(afterdeletingtheªdon'tknowºresponses).Themeanvalueis35.8percent,withastandarddeviationof14percent(seeDataAppendixforcoun-tryvalues).Thistrustitemissomewhatambiguouswithrespecttowhichªpeopleºrespondentshaveinmind.Thetermªpeopleºisgeneralenoughthatresponsesshouldnotmerelyre¯ectexpecta-tionsaboutthebehavioroffriendsandfamily.Responses,how-ever,couldeasilyre¯ectavaryingmixoftwoconceptsacrossindividuals:howmuchtrustoneplacesinpeoplewhoarenotclosefriendsorrelatives,andthefrequencyofencounterswithsuchpersons.Peopleinlow-trustenvironmentswilltransactmorewithclosefriendsandrelativesthanwithstrangers,com-paredwithpeopleinhigh-trustenvironments.Ifbyªmostpeopleºrespondentsconsidermostpeoplethattheytransactwith,thevariationinourtrustmeasurewillbereduced,makingitmoredif®culttorejectnullhypothesesregardingtheeffectsoftrust.Thestrengthofnormsofciviccooperationisassessedfromresponsestoquestionaboutwhethereachofthefollowingbehav-iorsªcanalwaysbejusti®ed,neverbejusti®edorsomethingina)ªclaiminggovernmentbene®tswhichyouarenotenti-tledtoºb)ªavoidingafareonpublictransportºc)ªcheatingontaxesifyouhavethechanceºd)ªkeepingmoneythatyouhavefoundºe)ªfailingtoreportdamageyou'vedoneaccidentallytoaparkedvehicle.ºQUARTERLYJOURNALOFECONOMICS5.In26countriesinoursample,respondentsinthe1990surveyswereaskedhowmuchtheytrustedªyourfamily.ºResponseswereona®ve-pointscalerangingfromªtrustthemcompletelyºtoªdonottrustthematall.ºMeansofthisscalewerecorrelatedwithTRUSTacrosscountriesatonly.24.TrustinthefamilyÐunlikeTRUST,asshownbelowÐisuncorrelatedwitheconomicandgovernmentalperformance,formalconstraintsonchiefexecutives,incomeequality,andethnichomogeneity. Respondentschoseanumberfrom1(neverjusti®able)to10(al-waysjusti®able).Wereversedthesescales,sothatlargervaluesindicategreatercooperation,andsummedvaluesoverthe®veitemstocreateascale(CIVIC)witha50-pointmaximum.Eachofthese®veitemsre¯ectsthestrengthofcivicnormsasthatconceptisde®nedabove;cooperativesolutionstotheseprisoner'sdilemmasimposefewornocostsonnonplayers.ThemeanvalueforCIVICis39.4,withastandarddeviationofonly2.Asitisbasedonmultiplesurveyitems,eachwithnumerousresponsecategories,CIVICmaybemorediscriminatingthanTRUST,whichisbasedonanitemwithonlytworesponsecatego-ries.Ontheotherhand,respondentsarelikelytobefarmorereluctanttoadmittocheatingthegovernment,taxpayers,orotherpeoplethantoagreethatotherscheat.ThisproblemmayintroducesubstantialmeasurementerrorintoCIVIC,likelyac-countinginpartforitslowvariationacrosscountries.WeuseTRUSTasourprimarysocialcapitalindicatorinourempiricaltests,becauseitismoredirectlyrelevanttoeconomicactivityÐasindicatedbythegreaterattentiontheconcepthasreceivedintheliteratureÐandbecauseCIVICexhibitssolittlevariationacrosscountries.However,resultsforCIVICareinmostcasesverysimilar,andweoftenreportresultsusingbothmeasures.DatafromexperimentsconductedbytheReader'sDigestreportedinTheEconomist,June22,1996)providereassuringbe-havioralevidenceforthevalidityofthesesurveymeasures.Twentywalletscontaining$50worthofcashandtheaddressesandphonenumbersoftheirputativeownerswereªaccidentallyºdroppedineachoftwentycities,selectedfromfourteendifferentwesternEuropeancountries.TenwalletsweresimilarlyªlostºineachoftwelveU.S.cities.Thenumberofwalletsreturnedwiththeircontentsintactwasrecordedforeachcity.ThepercentageofwalletsreturnedineachcountrycloselytrackstheWVSmea-sures:itiscorrelatedwithTRUSTat.67,andwithitem(d)oftheCIVICindex,ontheacceptabilityofªkeepingmoneythatyouhavefoundºat.52(partialcorrelationscontrollingforpercapitaincomeareevenhigher).CorrelationswiththeotherfourCIVICitemsareallsomewhatlower.Thisevidenceindicatesthatnon-randomsamples,translationproblems,anddiscrepanciesbe-tweenprofessedattitudesandactualbehaviorsdonotintroduceseverenoiseintooursurvey-basedmeasuresofsocialcapital.ThehighcorrelationofTRUSTwiththepercentageofwal-letsreturned(bystrangers),anditsrelativelylowcorrelationDOESSOCIALCAPITALHAVEAPAYOFF? withtrustinfamilymembers,indicatethatTRUSTisprimarilycapturingªgeneralizedºtrustasopposedtoªspeci®cºtrustplacedinpeopleonehasrepeatedinteractionswith.Similarly,CIVICisde®nedbyattitudestowardcooperatingwithanonymousothersinprisoner'sdilemmasettings.Thesecharacteristicsofoursocialcapitalmeasuresreducethechancesthattheyaremeasuringtrustandcooperationlargelyattheleveloflimitedgroupsbasedonkinship,ethnic,orspecialinterestties,whichhavepotentiallylargenegativeeffectsoneconomicperformance.Cooperationandtrustamongtheselimitedgroupsmayfacilitatetheirorganiza-tionforrent-seekingpurposesorevenforviolentcon¯ict.ThevariablesTRUSTandCIVICareinsomesensemirrorimagesofeachother.Thesurveyitemontrustmeasuresexpecta-tionsofwhetherotherswillactopportunisticallyatone'sex-pense;TRUSTthereforere¯ectsthepercentageofpeopleinasocietywhoexpectthatmostotherswillactcooperativelyinpris-oner'sdilemmacontexts.Ourmeasureofciviccooperationre-¯ectsrespondents'ownstatedwillingnesstocooperatewhenfacedwithacollectiveactionproblem;itthuscanbethoughtofasªtrustworthiness.ºOnewouldnaturallyexpecttrustandtrustworthinesstobepositivelycorrelatedacrosssocieties:wherefewerpeopleprovetobetrustworthy,fewerpeoplewillbetrusting[Hardin1992,p.161].Causationlikelyrunstheotherway,also,asmanypeopleareªconditionalcooperatorsºwhoactcooperativelyonlywhentheyhavehighexpectationsthatotherswillreciprocate[Hardin1982],incontrasttoªKantiansºwhofollowmoralrulesprescrib-ingcooperationregardlessofwhatothersdo.Thus,ªnotonlydoexpectationsaffecthonestbehavior,butovertimehonestbehav-ioraffectsexpectationsº[Platteau1994,p.760].FigureIdepictsthepositiverelationshipbetweenTRUSTandCIVICinourIV.TOOPERATIONLittleevidencedirectlyconnectstrustandciviccooperationtoeconomicperformance.NarayanandPritchett[1996]®ndforasampleofTanzanianvillagesthathigherlevelsofassociationalmembershipsarerelatedtohigherincomes.EconomichistoriansQUARTERLYJOURNALOFECONOMICS6.Thesimplecorrelationis.39;thepartialcorrelation(controllingforpercapitaGDP)is.33. havedocumentedcaseswheretrustresultingfromrepeatedin-teractionbetweenparties,withtheexpectationthatthepresentvalueofrewardsfromfutureinteractionsoutweighsthebene®tsfromrenegingoncurrentdeals,wasassociatedwithexpandedtradeandeconomicactivity.Forexample,Greif[1989]showsthatthedevelopmentofformalinstitutionsthatpromotetrust(self-enforcingagreements)hadadramaticimpactonthespreadoflongdistancetradeintheMiddleAges.HelliwellandPutnam[1995]showthat,holdinginitialin-comeconstant,regionsofItalywithamoredevelopedªciviccom-munityºhadhighergrowthratesoverthe1950±1990period.Theymeasureªciviccommunityºbyacompositeindexofnewspa-perreadership,thedensityofsportsandculturalassociations,turnoutinreferenda,andtheincidenceofpreferencevoting.Us-ingWVSdata,Helliwell[1996a]®ndstrust(andgroupmember-ships)tobesigni®cantlyandnegativelyrelatedtoproductivitygrowthinasampleofseventeenOECDmembers.WetesttheimpactoftrustandcivicnormsonbothgrowthandinvestmentratesusingtheWVSindicatorsdescribedabove.Tominimizeendogeneityproblems,wemeasureperformancesubsequenttothemeasurementoftrustandciviccooperationDOESSOCIALCAPITALHAVEAPAYOFF? CivicCooperationandTrust7.Hissampleomitsthepoorandmiddle-incomenationsforwhichwereportbelowthattrusthasthelargesteffects.Healsouses1990dataontrust,ratherthanfromtheearly1980s,andcontrolsonlyforinitialincome. whereverpossible.Asotherexplanatoryvariablesweinclude,fol-lowingmuchoftheliteraturesinceBarro[1991],theproportionofeligiblestudentsenrolledinsecondaryandprimaryschoolsin1960,percapitaincomeatthebeginningoftheperiod,andthepricelevelofinvestmentgoods,relativetotheUnitedStates.Thedependentvariableinequations1±3ofTableIisaverageannualgrowthinpercapitaincomeoverthe1980±1992period.Our29-nationsampleoverthisperiodbehavessimilarlytothelargersampleandlongertimeperiodsusedbyBarro[1991].In-comesconverge,conditionalontheothervariables.Schoolenroll-mentispositivelyrelatedtogrowth,andinvestmentgoodspricesarenegativelyrelatedtogrowth.Thesocialcapitalvariablesexhibitastrongandsigni®cantrelationshiptogrowth.Thecoef®cientforTRUSTinequation1indicatesthataten-percentage-pointriseinthatvariableisasso-ciatedwithanincreaseingrowthoffour-®fthsofapercentagepoint.Aone-standard-deviationchangeintrust(fourteenper-centagepoints)isassociatedwithachangeingrowthofmorethanone-half(.56)ofastandarddeviation,nearlyaslargeasthestandardizedcoef®cientforprimaryeducation(.64).Eachfour-pointriseinthe50-pointCIVICscaleinequation2isassociatedwithanincreaseingrowthofmorethanonepercentagepoint.Whenbothsocialcapitalvariablesareenteredtogetherinequa-tion3,theircoef®cientsdropslightlybutremainsigni®cant.Equation4introducesaninteractionterm,equaltoTRUST*GDP80.TheimpactofTRUSTongrowthshouldbehigherinpoorercountries,iftrustismoreessentialwherecon-tractsarenotreliablyenforcedbythelegalsystem,andwhereaccesstoformalsourcesofcreditismorelimitedduetoanunder-developed®nancialsector.Ontheotherhand,ifgreaterspecial-izationincreasesthenumberoftransactionsbetweenstrangers,andovertimeandacrossspace[Platteau1994,p.541],trustshouldreducetransactionscostsmoreinricherthaninpoorercountries[Putnam1993,p.178].Thenegativeandsigni®cantco-ef®cientonTRUST*GDP80inequation4providessupportfortheformerargument.ForacountrywithapercapitaGDPin1980QUARTERLYJOURNALOFECONOMICS8.Growth,investment,andinvestmentgoodspricesareallfromthePennWorldTablesversion5.6[SummersandHeston1991].Educationenrollmentdata,collectedbyUNESCO,arefromtheBarro-WolfdatasetusedinBarroBarro9.Rotatingcreditassociationsandsimilarinformalinstitutionsªseemingeneraltodisappearascapitalmarketsdevelopº[Besley1995,p.121]. TABLEIRUST,OOPERATION,AND Equation1234567 MethodOLSOLSOLSOLS2SLSOLSOLS variableGrowth1980±19921980±1992 1.0379.617(1.280)(4.730)(4.520)(1.895)(1.898)(3.820)(11.998)0.3750.1520.3660.1620.273(0.131)(0.126)(0.127)(0.274)(0.127)(0.403)(0.364)PRIM606.1925.9307.0614.8186.27011.65513.030(1.051)(1.164)(1.224)(1.709)(1.759)(3.558)(3.274)SEC602.1943.4571.6481.2562.0850.4310.495(1.632)(1.543)(1.485)(1.930)(2.133)(8.286)(7.067)3.117(0.867)(1.100)(0.935)(0.755)(0.809)(1.993)(2.154)TRUST0.0820.0760.1920.0860.146(0.030)(0.030)(0.060)(0.039)(0.078)CIVIC0.2720.2070.872(0.098)(0.092)(0.301) .55.44.56.60.52.37.381.371.521.351.291.374.434.38Mean,D.V.1.451.451.451.451.4522.422.4 White[1980]-correctedstandarderrorsareinparentheses.InstrumentsforTRUSTinequation5includelawstudents/totalpostsecondarystudents,Sullivan'shomogeneityindicator,GDP80,SEC60,PRIM60,andPI80.Notethatdonothavetheirusualinterpretationsin2SLS.'swithoutsocialcapitalvariablesare.41(growth)and.33(inv./GDP).Allequationshave29observations. ofonly$1000,TRUST'scoef®cientis.179,morethandoubletheaverageeffectoverthewholesample.Thesocialcapitalvariablesaremeasurednearthebeginningofoursampleperiodfor21ofour29countries.Fortheothereight,measuringtrustin1990couldintroducereversecausationproblems.Deletingtheseeightreducesthesigni®canceofalloftheregressors,particularlyGDP80,SEC60,andPI80.Coef®-cientsforthesocialcapitalvariablesriseinmostcaseswhentheseeightcountriesaredeleted,suggestingthatreversecausal-ityisnotdrivingourresults.ThisresultisnotsurprisinginlightofthefactthatforthetwentycountrieswithTRUSTvaluesforboth1981and1990,thecorrelationbetweenthetwois.91.Inequation5ofTableIweinstrumentforTRUSTtocorrectforendogeneityproblems,orpossiblemeasurementerror.Sulli-van[1991]measuresthepercentageofacountry'spopulationbe-longingtothelargestªethnolinguisticºgroup,wheregroupsareidenti®edbyrace,language,orreligiondependingonwhichoftheseappearstobethemostimportantcleavageinasociety.Asasecondinstrument,weusethenumberoflawstudentsin1963asapercentageofallpostsecondarystudents.forTRUSTwiththesetwovariablesandalloftheotherright-hand-sidevariables,TRUSTremainsasigni®cantpredictorofgrowth(equation5).WewouldexpectTRUSTandCIVICtoaffectgrowththroughinnovationsthatincreasetotalfactorproductivity,andthroughfactoraccumulation.Wheninvestment'sshareofGDPisincludedQUARTERLYJOURNALOFECONOMICS10.Thesourceisthe1965UNESCOStatisticalYearbook.Dataarefrom1962±1964.Murphy,Shleifer,andVishny[1991]usedsimilarmeasuresoflawandengineeringstudents.Lesstrusting,morelitigious,andmorecrime-riddensocietieswillhaveahigherdemandforlawyers,aswillmorepolarizedsocietiesinwhichspecialinterestslobbygovernmentsforrent.Butlawyersmayalsobeinhigherdemandwhereformalinstitutionssuchascourtsworkbetter.Forexample,Putnam[1993,pp.126±127]attributestheproliferationofªnotaries,lawyers,andjudgesºintheItaliancity-statestotheªunusualcon®denceinwrittenagreements,innegotiation,andinthelawºratherthantoªcontentiousness.ºInoursample,thereisastrong,negativerelationshipbetweenTRUSTandlawyers.AHausmantestoftheoveridentifyingrestrictionsindicatesthatthehomogeneityandlawstudentsvariablesdonotbelonginthegrowthregressiondirectly.11.InstrumentingforTRUSTwithdistancefromtheequator,asmeasuredbyHallandJones[1996],producesa2SLScoef®cientestimateforTRUSTof.158.086).HallandJonesusethisvariableasaclimateindicator,arguingthattemperateregionshaveanadvantageovertropicalorpolarregions;they®ndempirically,however,thatincomeperworkerincreases(roughly)linearlywithlatitudeovertheobservablerange.ThisvariableiscorrelatedwithTRUSTinoursampleat.63.AHausmantestindicatesthatlatitudedoesnotin¯uencegrowthindependentlyofitseffectsontrustinoursample. asaregressoringrowthequations,coef®cientsforTRUSTandCIVICremainpositivebutarenolongerstatisticallysigni®cant,suggestingthataccumulationisthemoreimportantchannel.Similareffectsareobservedwithmeasuresofhumancapitalaccumulation.Whensecondaryenrollmentisomittedfromtheregressions,TRUSTandCIVICcoef®cientsrise,asexpected,giventheargumentsfromSectionIIthatsocialcapitalin¯uenceshumancapitalaccumulation.SubstitutingattainmentmeasuresfortheenrollmentvariablesreducestheTRUSTandCIVICcoef®cientssomewhat.ThegrowtheffectsofTRUSTandCIVICremainsigni®cantusingliteracyrates,orKyriacou's[1991]esti-matesofaverageyearsofcompletededucationfor1980.Theso-cialcapitalcoef®cientsdiminishmuchmoreusingaverageyearsofcompletededucationfor1980asestimatedbyBarroandLeeLee13buteventhenthelogofTRUSTremainsasigni®cantpredictorofgrowth.Equations6and7ofTableIexaminetheimpactofsocialcapitaloninvestment/GDP,averagedoverthe1980±1992period,controllingforotherdeterminantsofinvestment.Trustisposi-tivelycorrelatedwithinvestmentinequation6,butissigni®cantatthe.05levelonlyforaone-tailedtest.Eachseven-percentage-pointriseintrustisassociatedwithaone-pointriseininvest-ment'sshareofGDP.Civicnormsarehighlysigni®cantinequa-tion7,witheachone-pointriseintheindexassociatedwithanincreaseininvestmentofnearlyonepercentagepoint.TheresultsinTableIarefairlyinsensitivetochangesinspeci®cation,theexclusionofin¯uentialobservations,andthein-clusionofadditionalregressors.These®ndingsareespeciallyre-markablegiventhesmallsamplesize.AsaboutonehalfofoursampleconsistsofWesternEuro-peancountries,autocorrelatederrorsduetospatialcorrelation,ortocommonmembershipintheEC,isapotentiallyseriousproblem.ResidualsamongthesubsetofECmembers(orWesternDOESSOCIALCAPITALHAVEAPAYOFF?12.Thesocialcapitalmeasurescouldneverthelessin¯uenceinnovation,asproductiveknowledgemaybeembodiedinmachineryandotherinvestments,orinhumancapital.13.ThecorrelationbetweenTRUSTandBarro-Leeattainmentfor1980is.83,thehighestcorrelationwithTRUSTfoundforanyvariable.14.SincethemarginalimpactofsocialcapitalonperformanceappearstobegreatestatlowerlevelsofTRUSTandCIVIC,weconsideredlogspeci®cations.TheloggedvaluesofTRUSTandCIVICgenerallyproducestrongerresultsthanthosereportedinTableI. Europeancountries)werenotfoundtodiffersigni®cantlyfromthosefornonmembers,however.Themostin¯uentialindividualcaseinTableIregressionsisKorea.The®rstrowofTableII(labeledªNoneº)repeatstheregressioncoef®cientsandstandarderrorsforTRUSTandCIVICfromTableI.SucceedingrowsreporttheeffectsofTRUSTandCIVICongrowthandinvestmentwhentheindicatedchangeismadetotherelevantequationfromTableI.ThesecondrowshowsresultswhentheobservationineachequationwiththeQUARTERLYJOURNALOFECONOMICS15.DeLongandSummers[1991]foundnosigni®cantspatialcorrelationintheirsample,usingmorerigorousteststhattookintoaccountthedistanceofeachpairofnationalcapitalsfromeachother.TheyreportthatthethreeSouthernConecountries(onlytwoofwhichareinoursample)hadsimilarresiduals,butthattheEuropeancountriesexhibitednogeographicpattern.TABLEIIOBUSTNESSTOLTERNATIVEPECIFICATIONS DependentvariableGrowthInvestment/GDP Equation(fromTableI)1267 Speci®cationchangeTRUSTCIVICTRUSTCIVIC None(fromTableI)0.082(0.030)0.272(0.098)0.146(0.078)0.872(0.301)In¯uentialobs.0.054(0.020)0.223(0.081)0.081(0.061)0.657(0.270)2in¯uentialobs.0.038(0.016)0.202(0.082)0.051(0.060)0.575(0.233)Laborforcegrowth0.082(0.027)0.312(0.105)0.146(0.075)0.928(0.311)0.071(0.025)0.334(0.116)0.141(0.072)0.930(0.310)/GDP0.076(0.029)0.185(0.106)0.130(0.079)0.652(0.289)Blackmarket0.070(0.029)0.225(0.094)0.128(0.080)0.806(0.300)Propertyrights0.083(0.039)0.300(0.084)0.155(0.096)0.917(0.311)Currency0.047(0.023)0.221(0.103)0.108(0.078)0.808(0.303)Inst.investorcredit0.065(0.024)0.229(0.086)0.115(0.079)0.799(0.290)Gini(income)0.059(0.028)0.099(0.101)0.143(0.073)0.814(0.418)1970±1992gr,inv0.039(0.022)0.126(0.088)0.160(0.059)0.773(0.328)1960±1992gr,inv0.029(0.017)0.121(0.096)0.160(0.058)0.807(0.321) Standarderrors(inparentheses)areWhite-corrected.IcelandismissingdataforGini.Forallotherequationswherein¯uentialobservationsarenotdeleted,N29.IndependentvariablesincludePRIM60,SEC60,initialincome,andinitialinvestmentgoodsprices. largestpositivein¯uenceontheslopeofthesocialcapitalcoef®-cientisdeleted.ForTRUST'seffectongrowthandinvestment,andtheeffectofCIVICongrowth,thatobservationisKorea.Intwoofthesethreecases,thesocialcapitalvariableremainssta-tisticallysigni®cantwhenKoreaisdeleted.Switzerlandisthemostin¯uentialobservationincreasingCIVIC'srelationshiptoinvestment;thisrelationshipremainssigni®cantwhenthiscaseisdeleted.ThethirdrowofTableIIdeletesthetwocasesmostfavorableforsocialcapital'simpactoneconomicperformance.Thisproce-duredeletesKoreaandBrazil(TRUSTandgrowth;seeFigureII),KoreaandSwitzerland(CIVICandgrowth),KoreaandtheUnitedStates(TRUSTandinvestment),andSwitzerlandandJa-pan(CIVICandinvestment).Allthreecoef®cientsthatwerestillsigni®cantwhenthesinglemostin¯uentialobservationwasomittedremainsigni®cantafterdeletingthesesecondobserva-tionsfromour29-nationsample.NeitherNigerianorMexicoÐtwooil-exportinglow-trustnationswithlowornegativegrowthinthe1980sfollowingrapidpetroleum-ledgrowthinthe1970sÐareamongthesein¯uentialobservations.TheeffectsofTRUSTandCIVICprovetoberobusttotheinclusionofothervariablesoftenusedingrowthregressions,in-DOESSOCIALCAPITALHAVEAPAYOFF? PartialRegressionPlot:Growth(1980±1992)andTrust dicatingthattheyarenotmerelycapturingtheeffectsofimpor-tantomittedvariables.TableII(beginningwiththefourthrow)indicateshowcoef®cientsandstandarderrorsforTRUSTandCIVICchangewhentheindicatedadditionalregressorisin-cludedintherelevantgrowthorinvestmentequation.Mostoftheaddedregressorsaresigni®cantpredictorsofgrowth,butnotofinvestment,inoursample.EstimatesforTRUSTandCIVICarechangedlittlebyadd-inglaborforcegrowth,tradeopenness(asmeasuredbyexportsplusimportsdividedbyGDP),M/GDP(theprimarymeasureof®nancialdevelopmentmeasureusedbyKingandLevine[1993],theblackmarketpremium,orthepropertyrightsindicatorfromICRG(InternationalCountryRiskGuide)introducedbyKnackandKeefer[1995].InresultsnotshowninTableII,TRUSTandCIVICalsoproveinsensitivetotheinclusionofthemineralsec-tor'sshareofGDP,orthegovernmentsizeandpoliticalinstabil-ityindicatorsusedinBarro[1991].Otherpolicyvariablesaffectthesocialcapitalestimatestoasomewhatgreaterextent,butinwaysthatareconsistentwithourtheory.Thesearepoliciesthatareparticularlysensitivetosocialpolarization,whichweexpecttobeassociatedwith(andmeasuredinpartby)lowtrustandweakercivicnorms.Forex-ample,dif®cultyinimplementingstabilizationprogramshasbeenattributedtopolarizationassociatedwithinequalitiesinin-comeandwealth,andwithethnictensions[BergandSachs1988;KeeferandKnack1995].Inmorepolarizedsocieties,groupsaremorewillingtoimposecostsonsociety,forexample,byfailingtocompromiseonareformprograminatimelyway.Highin¯ationandgovernmentdebtcouldthereforebeaproductinpartoflowsocialcohesiveness.TableIIshowsthatcoef®cientsforTRUSTandCIVICfallÐalthoughusuallyremainingsigni®cantÐwhencurrencydepreciation,orinternationalbankers'evalua-tionsofthecreditworthinessofgovernments,areaddedtotheAmoredirectapproachtosocialpolarizationistoexamineinequality,asourceofpolarizationthathasbeenlinkedtounsta-QUARTERLYJOURNALOFECONOMICS16.Iftrustisreducedbyuntrustworthybehaviorbygovernmentof®cials,itsimpactoneconomicperformancecouldbeassociatedwithpolicycredibility.ThispossibilityisasecondexplanationofwhytheTRUSTcoef®cientmaydeclinewhencurrencydepreciationisincludedintheequation.17.Thesevariablesareaveragesoverthe1980±1990period.InstitutionalInvestorconductsthesurveyofinternationalbankers;thismeasureofcreditwor-thinessisalsousedinKeeferandKnack[1995]. blemacropolicyandtouncertainpropertyrights.Therearetheoreticalreasonstoexpectinequalitytolowertrustandweakencivicnorms.(ThisissueisaddressedinmoredetailinSectionVII.)TheGinicoef®cientforincomeinequalityisstronglycorrelatedwithTRUST(r.65)andwithCIVIC(.43).WhenthisGiniisaddedtoourregressions(seeTableII),theTRUSTcoef®cientdeclinesto.059(butremainssigni®cantatthe.05level),whiletheCIVICslopedropsto.099.Inequalityisnotsig-ni®cantininvestmentequations,however,whereithaslittleimpactonTRUSTandCIVIC.Whilethereappearstobeanim-portantrelationshipbetweeninequalityandsocialcapital,thesocialcapitalvariablesremainsigni®cantinthreeoutoffourcaseswithinequalityinthemodel,suggestingthatTRUSTandCIVICinTableIarenotsolelycapturingredistributionalorothereffectsofinequality.Iftrustandcivicnormsareviewedasªcultureºvariablesthatchangeonlyslowlyovertime,TRUSTandCIVICasmea-suredintheearly1980sshouldexplaincross-countryvariationinlong-termeconomicperformance,asmeasuredbyinvestmentorgrowthratesover20-or30-yearperiods,orevenbypercapitaincomelevels.Thisassumptionmaybereasonable,asTRUSTvaluesfor1980and1990arecorrelatedat.91.Moreover,changesinTRUSToverthedecadeareuncorrelatedwithgrowthrates.However,thereareboththeoreticalandempiricalreasonsforcautionregardingtheassumptionthattrustandthestrengthofcivicnormsarestableoverlongperiodsoftime.Cooperativeequi-libriacanunravelveryquickly,asYugoslaviademonstrates.TheUnitedStates,theonecountrywithalongtimeseriesonthetrustsurveymeasure,showsasteadydeclineintrustfrom55to60percentinsurveysfromthelate1950sandearly1960s,tothemid-andupper-30sinthe1990s.Withthesecaveatswereporttestsoflong-termperformancebelow,investigatinglonger-periodgrowthandinvestmentratesaswellaslevelsofoutputperworker.ThelasttworowsofTableIIshowtheassociationofTRUSTandCIVIConinvestmentlevelsandincomegrowthforthe1970±1992and1960±1992periods.Investmentresultsareverysimi-DOESSOCIALCAPITALHAVEAPAYOFF?18.SeeBergandSachs[1988]andKeeferandKnack[1995].19.Inequalityismeasuredasnearto1980asthedatapermit.Icelandismissingdataforinequality.SourcesaretheWorldDevelopmentReportandMila-novic[1994].20.Initialincomeandinvestmentgoodspricesaremeasuredin1970or1960asappropriate. lartothosefortheshorterperiod;theTRUSTcoef®cientrisessomewhatwhilethestandarderrorfalls.FigureIIIdepictsthesimplecorrelationofTRUSTandinvestment/GDPforthe1960±1992period.Coef®cientsingrowthregressionsfallbymorethanone-half,relativetotheshorterperiod.TheeffectofTRUST,butnotCIVIC,remainssigni®cantatthe.05levelforaone-tailedTheweakerrelationshipbetweensocialcapitalandgrowthinthelongerperiodsisdrivenmainlybythreelow-trustcoun-triesthatgrewslowly,ifatall,inthe1980±1992period,butrap-idlyinthe1960sand1970sÐinpartthroughdramaticincreasesinoilpricesinthecaseofNigeria,largein¯owsofdebtinthecaseofBrazil,andacombinationofthetwointhecaseofMexico.An(admittedlyexpostfacto)explanationoftheexperiencesofthesecountriesmaybeconsistentwithourtheory.Iftrustisweak,leadersaremorelikelytodirectrevenuewindfallstowardQUARTERLYJOURNALOFECONOMICS Investment/GDP(1960±1992)andTrust21.Fortheselongerperiods,TRUSTandCIVICremainsigni®cantlyrelatedtoinvestmentwhenadditionalregressorslistedinTableII(appropriatelymodi-®edforthelongerperiod)areadded.Forgrowth,TRUSTandCIVICcoef®cientsremainpositiveineverycasebutarerarelysigni®cantwhenotherregressorsareadded. consumptionthantowardproductiveinvestments.Becauseoflowtrust,leaderscannotcrediblypromisesupportersfuturebene®tsfromworthwhileinvestments.Tomaintainsupport,leadersmustdivertresourcestosupporterswho,againbecauseoflowtrust,donot®nditadvantageoustoinvesttheseresourcesinproductiveinvestmentsinsidethecountryÐmakingthecountrymorevul-nerabletocrisis.Moreover,oncecrisishits,thelackofsocialcohe-sivenessre¯ectedinlowvaluesofTRUSTandCIVICmakesitlesslikelythataconsensusonneededpolicyreformscanbebuilt.WealsoanalyzetherelationshipbetweenTRUSTandlevelsofoutputperworker,physicalandhumancapitalperworker,andtotalfactorproductivity(TFP),followingthelevelsaccountingapproachofHallandJones[1996]andusingdataprovidedbythem.TableIIIreportstherelationshipofTRUSTwith(1)thelogofoutputperworker,netofmineralproduction,for1988(fromSummersandHeston[1991]),(2)theHallandJonesestimateof(thelogof)capitalperworkerfor1988(usinginvestmentdatafromSummersandHeston),(3)humancapitalperworker(theBarro-Lee[1993]attainmentmeasurefor1985),and(thelogof)TFPfor1988,estimatedbyHallandJonesasaresidual.Column1ofTableIIIreportscoef®cientsandstandarderrorsforTRUSTfromfourbivariateregressions:TRUSTispositivelyandsigni®-DOESSOCIALCAPITALHAVEAPAYOFF?TABLEIIIRUSTANDEVELSOFUTPUT,CCUMULATION,AND Equation1234 NoICRG,FullHall-yearsopen,NoICRG,DependentvariableJonesmodelTrustonlyyearsopenorlatitude logoutput/worker.0258.0035.0197.0397(.0061)(.0081)(.0085)(.0177)logcapital/worker.0336.0015.0223.0454(.0062)(.0107)(.0102)(.0198)School/worker.1533.0977.1312.1667(.0156)(.0298)(.0268)(.0427)logTFP.0023.0056.0025.0127(.0038)(.0048)(.0046)(.0094) CellsoftablereportTRUSTcoef®cients;White-correctedstandarderrorsareinparentheses.FullHall-Jonesmodelincludeslatitude(distancefromtheequator),percentEnglish-speaking,percentspeakingan-otherªinternationallanguage,ºadummyforªcapitalist-statistºsystems(ascontrastedtoªcapitalistº),frac-tionofyears,1950±1994,withopeneconomy(fromSachsandWarner[1995]),andtheICRGpropertyrightsindex(fromKnackandKeefer[1995]).Samplesizeis29. cantlycorrelatedwithoutput,capital,andschooling,whilethecorrelationwithTFPispositivebutinsigni®cant.HallandJones[1996]reportregressionsofeachofthesefourvariablesonasetofvariables(listedinthenotetoTableIII)thattheycontendareexogenousandthatmeasureªbasiccharacteris-ticsºofaneconomy.WhenTRUSTisaddedtotheirfullmodel,itscoef®cientissigni®cantonlyforschooling(column2ofTableIII).Theoryandevidenceprovidedelsewhereinthispaperstronglysuggestthatpolicyisendogenoustotrust:accordingly,incolumn3ofTableIIIwedeletefromtheregressionsthetwopolicyvari-ablesusedbyHallandJones:theICRGpropertyrightsindexfromKnackandKeefer[1995],andtheopennessmeasurefromSachsandWarner[1995].Asinthesimplebivariateregressions,output,capital,andschoolingareallsigni®cantlyrelatedtoTRUSTwhenthesepolicyvariablesaredropped.Themostpow-erfulªdeterminantºofoutput,factoraccumulation,andTFPinHallandJones[1996]islatitude(distanceindegreesfromtheequator),avariablewithanambiguoustheoreticalinterpreta-tion,andwhichishighlycorrelatedwithTRUST.Whenlatitudeandthetwopolicyvariablesaredroppedfromthemodelsincol-umn4ofTableIII,coef®cientsforTRUSTarealllargerthaninthebivariateregressions,althoughstillnotsigni®cantinthecaseofTFP.Causalitycouldgoinbothdirectionsforincomeandtrust,andforeducationandtrust.Forexample,trustcouldbeaprod-uctofoptimism[Uslaner1995]generatedbyhighorgrowingin-comes.Butiftrustwerelargelyaproductratherthanacauseofhighincomes,itshouldfollowratherthanleadpercapitaincomelevels.Infact,TRUSTismorestronglycorrelatedwithpercapitaincomesinlateryears,suchas1990,thanwithincomeinearlieryears,suchas1970.Similarly,educationmaystrengthentrustandcivicnorms,forexample,ifignorancebreedsdistrust,oriflearningreducesuncertaintyaboutthebehaviorofothers,orifstudentsaretaughttobehavecooperatively[Mueller1989,pp.364±365).QUARTERLYJOURNALOFECONOMICS22.Measurementerrormaybepartlyresponsibleforthislowcorrelation,asimplausiblyhighTFPestimatesareconcentratedamonglow-trustcountriesinoursample.Forexample,TFPestimatesarefarhigherforMexico,Italy,andBrazilthanforNorway,Finland,Denmark,andSweden(andhigherthanforGer-many,Switzerland,andtheUnitedStates).23.TheriseofJapaninthelatenineteenthcenturycoincidedwithagovernment-initiatedshiftfromªlimited-groupºmoralitytoamoregeneralizedmorality,usingthesystemofuniversalcompulsoryeducationasanimportanttoolinthiscampaign[Platteau1994]. However,TRUSTismorehighlycorrelatedwithrecentthanwithpastmeasuresofattainmentorenrollment,suggestingthatcau-salitydoesnotrunsolelyfromeducationtosocialcapital.V.ASSOCIATIONALCTIVITYANDTheimportanceoftrustinexpandingeconomicactivityisperhapsnotsurprising,althoughnotpreviouslyempiricallysub-stantiatedatthecross-countrylevel.Theroleofassociationalac-tivityisasubjectofgreatercontention.Putnamattributestheeconomicsuccessandgovernmentalef®ciencyofnorthernItaly,relativetothesouth,inlargeparttoitsricherassociationallife,claimingthatassociationsªinstillintheirmembershabitsofco-operation,solidarity,andpublic-spiritednessº[Putnam1993,pp.89±90].Thiscooperationandsolidarityisinvokedmostcom-monlytoresolvecollectiveactionproblemsatthelevelofsmallergroups,however.Iftheeconomicgoalsofagroupcon¯ictwiththoseofothergroupsorofunorganizedinterests,theoverallef-fectofgroupmembershipsandactivitiesoneconomicperfor-mancecouldbenegative.AdamSmith(quotedinGranovetter[1985,p.484])notedthatwhenªpeopleofthesametradeºmeetªevenformerrimentanddiversionºtheresultisoftenªaconspir-acyagainstthepublicºorªsomecontrivancetoraiseprices.ºInamoreextremeexample,Marxblamedtheinabilityofthenine-teenthcenturyFrenchpeasantrytooverthrowcapitalismontheabsenceofdensenetworksofsocialinteraction:thepeasantsdidnotenterintoªmanifoldrelationswithoneanotherinHardin[1982,p.189]).Similarly,Olson[1982]observesthathorizontalassociationscanhurtgrowthbecausemanyofthemactasspecialinterestgroupslobbyingforpreferentialpoliciesthatimposedisproportionatecostsonsociety.KeeferandKnack[1993]providesomeevidenceforthecon-¯ictingin¯uencesofassociationalactivityongrowth,usingavariablefromBanksandTextor[1963]calledªinterestarticula-tionºwhichassesses(onasubjectivescalerangingfrom1to4)howeffectivelygroupsarticulatetheirpolicypreferencestogov-ernment.Althoughtheabilityofgroupstoarticulatetheirinter-estsislikelytobeanimportantrestraintongovernment,italsoprovidesgroupsawaytocaptureprivatebene®tsattheexpenseofsociety.Consistentwiththeviewthatthesetwoeffectstendtocounteracteachother,ªinterestarticulationºprovestobeaninsigni®cantpredictorofgrowthwhenintroducedintoBarro-typecross-countrytests.DOESSOCIALCAPITALHAVEAPAYOFF? Weobtainasimilarresulthereonthein¯uenceofhorizontalassociations.RespondentsintheWorldValuesSurveyswereaskedwhethertheybelongedtoanyofthefollowingtypesofa)socialwelfareservicesforelderly,handicapped,orde-privedpeople;b)religiousorchurchorganizations;c)education,arts,music,orculturalactivities;d)tradeunions;e)politicalpartiesorgroups;f)localcommunityactiononissueslikepoverty,employ-ment,housing,racialequality;g)thirdworlddevelopmentorhumanrights;h)conservation,theenvironment,ecology;i)professionalassociations;j)youthwork(e.g.,scouts,guides,youthclubs,etc.).Ourmeasureofthedensityofassociationalactivity(GROUPS)istheaveragenumberofgroupscitedperrespondentineachcoun-try(seeDataAppendixforvalues).Thisindicatorunfortunatelydoesnotmeasuretheintensityofparticipationingroups.Assum-ingthatgroupmembershipsarecorrelatedwithlevelsofactivity,GROUPSconstitutesareasonableapproximationofPut-nam'snotionofthedensityofhorizontalnetworksinasociety.IndependentdataonunionmembershipsasaproportionofthelaborforceareavailablefromWallerstein[1989],foreighteenofthecountriesinoursample(forthelate1970s),permittingacheckonthereliabilityofthesurveydata.Thecorrelationofthisvariablewithitem(d)ontradeunionmembershipsisareassuring.68.InTableIVwetesttherelationshipbetweenGROUPSandeconomicperformance,usingmodelssimilartothoseinTableI.Groupmembershipisnotsigni®cantineithergrowthorinvest-mentequations.AnobviouspossibleexplanationforthisresultQUARTERLYJOURNALOFECONOMICS24.ªSportsandrecreationclubsºÐofparticularinterestinPutnam'sworkÐandothergroupswerealsoincludedinthesurvey,butfortoofewcountriestobeanalyzed.Groupmembershipsformostcountriesarefromthe1981waveofsur-veys,andfromthe1990wavefortheothers,aswithTRUSTandCIVIC.25.Putnam[1995b]usessimilarmeasuresofassociationsfromtheWVSand,fortheUnitedStates,fromtheGeneralSocialSurvey.26.Thesamplesizedropsfrom29inTableIto26inTableIV,astherearenodataongroupmembershipsforIndia,Nigeria,orTurkey.Thegreatermea-suredimpactofTRUSTandCIVICrelativetoGROUPSisnotduetothechangingsample,however:coef®cientsforTRUSTandCIVICinTableIactuallyrisewhenthosethreecountriesaredropped. isthattheharmfuleffectsofgroupsasrent-seekingorganiza-tionstheorizedbyOlson[1982]areoffsettinganypositiveeffectspositedbyPutnam[1993].Weexploredthispossibilityfurtherbyattemptingtodiffer-entiateªOlsonianºfromªPutnam-esqueºgroups.Groupsb,c,andjfromtheabovelistwereidenti®edasthosegroupsleastlikelytoactasªdistributionalcoalitionsºbutwhichinvolvesocialinter-actionsthatcanbuildtrustandcooperativehabits.ThetotalmembershipsperrespondentinthesethreeªPutnamºcategories(P-GROUPS)rangefrom.83(fortheUnitedStates)to.06(Fin-land).Groupsd,e,andiweredeemedmostrepresentativeofgroupswithredistributivegoals;totalmembershipsintheseªOlsonºgroupsperrespondent(O-GROUPS)rangefrom.76(Ice-land)to.12(Korea).MembershipinªOlsonºgroupsisnotsigni®cantineitherthegrowthorinvestmentequations(equations2and4).Perversely,DOESSOCIALCAPITALHAVEAPAYOFF?TABLEIVEMBERSHIPSAND Equation1234 DependentInvestment/GDP,variableGrowth,1980±19921980±1992 Constant1.1561.55821.67622.698(3.323)(3.618)(8.210)(8.528)0.2740.6480.448(0.153)(0.164)(0.414)(0.455)PRIM604.4214.8001.9332.968(1.641)(1.482)(5.302)(5.383)SEC604.1964.0654.2375.098(1.995)(2.061)(7.076)(7.328)(1.657)(1.767)(3.118)(3.316)(0.872)(3.306)O-GROUPS2.186(1.551)(5.980)(1.412)(3.890) .19.18.19.161.711.724.504.59Mean,D.V.1.541.5423.223.2 White-correctedstandarderrorsareinparentheses.Samplesizeis26. ªPutnamºgroupsappeartoharminvestment(equation4).Theseresultsarelittlechangedbytinkeringwiththede®nitionsofPut-namandOlsongroups,forexample,byleavingoutreligiousandchurchorganizationsfromP-GROUPS,onthegroundsthathier-archicalreligionsweakentrust[Putnam1993],orbydeletingpo-liticalpartiesandgroupsfromO-GROUPS,onthegroundsthatpartiesmayberelativelyªencompassing,ºaggregatingtheprefer-encesofmanydifferentgroups[Olson1982].ThisattemptatdistinguishingtypesofgroupsthusprovidesnoempiricalsupportforourconjecturethatgroupshaveneutraleffectsonperformancebecausepositiveªPutnamºforcesarecounterbalancedbynegativeªOlsonºin¯uences.Thesetestscanberegardedasonlypreliminary,however:thecategoriesofgroupsareoverlybroad,itisnotclearwhatsomeofthesegroupsdo,andthedepthofinvolvementisnotmeasured.WhileTRUSTandCIVICappeartocapturetrustandnormsofcooperation,ourdataongroupsdonotpermitustoconvincinglydistinguishbetweensociallyef®cientandinef®cientmember-shipsandactivities.VI.AINKSFROMRUSTTOThissectionfurtherexploresthechannelsthroughwhichtrustmightaffecteconomicoutcomes.Someofthesechannelsmustremainlargelyunexploredduetodatalimitations.FollowingargumentsinSectionII,weconsidertwopossiblechannelshere:theimpactoftrustonthestrengthofpropertyandcontractualrights,andtheimpactoftrustontheperformanceofgovernment.Weconstructedanindexofperceivedgovernmentperfor-mancefromitemsintheWVS.Respondentswereaskedªhowmuchcon®denceºtheyhadinvariousgovernmentalandsocietalinstitutions,withresponsesrangingfromªagreatdeal,ºtoªquitealot,ºtoªnotverymuch,ºtoªnoneatall.ºTheindexwasbuiltfromresponsestoitemsconcerningtheeducationsystem(largelygovernment-runinmostcountries),thelegalsystem,thepolice,andthecivilservice.Foreachofthesewecalculatedthepercent-QUARTERLYJOURNALOFECONOMICS27.Membershipingroupsclassi®edhereasªPutnamºgroupscouldsimplybeproxyingstrongerpreferencesforleisure,whichmightharmmeasurableeco-nomicperformance.TradeassociationsЪOlsonºgroupsÐdomorethanlobbyforlegalbarrierstoentryandtaxbreaks.Theymayhavepositiveeffectsoneconomicperformancebyestablishingethicalcodesandstandards[Bergsten1985]orbyreducingtransactionscosts,e.g.,byspreadinginformationabouttheidentityofcheaters.SeeGreif[1996]andGranovetter[1985]. ageofrespondentsineachcountrywitheitherªagreatdealºorªquitealotºofcon®dence.Themeanofthefourpercentagesisusedasameasureofperceivedoverallgovernmentperformance(seeDataAppendixforvalues).Equation1ofTableVregressesthisindexonTRUST,con-trollingforpercapitaincomeandeducationenrollments.onlysigni®cantdeterminantofgovernmentperformanceinthisequationisTRUST:eachtwo-percentage-pointriseisassociatedwithariseincon®denceofaboutonepercentagepoint.DOESSOCIALCAPITALHAVEAPAYOFF?TABLEVINKSFROMRUSTTO Equation12 Con®denceContractinBureaucraticPropertyrightsenforceabilityDependentgovernmentef®ciency(ICRG)(BERI)variable(WVS)ICRGBERI Constant0.6262.6622.0654.8122.6471.3951.266(0.074)(1.244)(0.755)(1.102)(1.100)(0.281)(0.318)GDP80.00850.2350.1340.2930.1700.0650.057(.0091)(0.102)(0.056)(0.149)(0.104)(0.029)(0.031)0.1622.1950.7505.7656.4830.0540.101(0.093)(1.263)(0.578)(1.468)(0.954)(0.333)(0.360)0.2152.2200.6751.9381.5860.4850.465(0.145)(1.412)(0.964)(1.847)(1.507)(0.426)(0.421)TRUST.00450.0500.0180.0650.0370.0150.014(.0013)(0.025)(0.010)(0.020)(0.029)(0.005)(0.005)Executive0.6160.037constraint(0.260)(0.039) .20.73.69.74.82.73.73.0961.33.6451.811.51.324.324Mean,D.V..6059.664.9116.316.32.752.75N28292829292828 White-correctedstandarderrorsareshowninparentheses.28.Switzerlandistheonlyonefromour29-countrysampleinwhichthesequestionswerenotasked.Whilerankingssuggestacertainamountofªfacevalid-ityºforthisindex,thereareseveralanomalies.Japanscorespoorly(46percent)andNigeriascoresveryhighly(73.1percent).Itisuncleartowhatextenttheseªcon®denceºitemsre¯ecthowresponsivethegovernmentistothenarrowlyself-interesteddemandsofrespondents,orhoweffectivelyitcarriesoutitslegallymandatedresponsibilities,orsomethingelse.29.Whereincomesarehigher,governmentsmayhaveaccesstomoretaxrevenuetoprovidebetterservices,andbetter-educatedcitizensmayplacemoreeffectivedemandsongovernmenttoprovideservicesef®ciently. Foralternativesubjectivemeasuresofgovernmentalperfor-mance,weusedatafromtwo®rmsthatevaluateriskstoforeigninvestors.Foroneindexofbureaucraticef®ciency,wecreateanadditiveindexfromtwovariablesfromInternationalCountryRiskGuide(ICRG):ªcorruptioningovernmentºandªqualityofthebureaucracy.ºThemaximumpossiblevalueforthisindexistwelve;higherscoresindicatelesscorruptionandhigher-qualitybureaucracies.Forasecondindexofbureaucraticef®-ciency,weaddtwovariablesfromBusinessEnvironmentalRisk(BERI):ªbureaucraticdelaysºandªtransportationandcommunicationsquality.ºThemaximumpossiblescoreiseight,withhigherscoresindicatingshorterdelaysandbetterTheseICRGandBERIvariablesareshowninKnackandKeefer[1995]tobestrongpredictorsofinvestmentandgrowth.Equations2and3ofTableVshowthatthesebureaucraticef®-ciencyindexesarepositivelyandsigni®cantlyrelatedtoTRUST.Thisevidencesuggeststhattherelationshipbetweenªsocialcapi-talºandgovernmentalperformancePutnamdiscoveredforItalymaygeneralizeacrosscountries.TableValsodemonstratesstrongrelationshipsbetweenTRUSTandtwosubjectivemeasuresofpropertyrightssecurity.Inequations4and5thedependentvariableisanindexoftwoICRGvariables,ªriskofrepudiationofgovernmentcontractsºandªriskofexpropriationºofassets.Themaximumpossiblevalueistwenty,withhigherscoresindicatinglowerrisks.Inequations6and7,thedependentvariableisªcontractenforce-ability,ºaBERImeasurewithamaximumvalueoffour.Consistentwithourtestsofthedeterminantsofgovernmentper-formance,wecontrolforincomeandeducation.Inequations4and6TRUSTissigni®cantlyrelatedtotheICRGandBERIprop-ertyrightsindicators.Equations5and7controlforformalsourcesofsecurepropertyrights,withªexecutiveconstraints,ºasubjectivevariablecodedbyGurr[1990]inhisPolityIIdataset.Countriesarescoredona1to7scaleontheextenttowhichtheexecutiveofacountryisabletorulebydecree(codedas1),orQUARTERLYJOURNALOFECONOMICS30.Theseindexeswerecreatedusingaveragesoverthe1980±1990periodforBERIand1982±1990forICRG.AllcountriesinoursampleareincludedinICRGdata.IcelandismissinginBERI.31.Highincomesmayproxythenumberandsizeofbusinesstransactions,whichwillberelatedtotheªdemandºforenforceablecontracts,andmaygivethegovernmentaccesstomorerevenuetoestablishaneffectivejudicialsystem. mustgaintheconsentofothersbeforeacting(coded7).Theinde-pendenteffectofTRUSTontheICRGindexisreducedsomewhat(equation5),butTRUSTremainsasigni®cantdeterminantoftheBERImeasure.Thesepreliminaryexplorationsintothelinksbetweensocialcapitalandgrowthhaveprovidedsomesuggestiveevidencethattrustmayimprovegovernmentalef®ciencyandincreaseinvest-ors'con®dencethatcontractswillbeenforced.Theevidenceisnotentirelyunambiguous,however.Wehavenotattemptedtodevelopcompletemodelsofgovernmentalperformanceorofthesecurityofpropertyrights,orenforceabilityofcontracts.TheICRGandBERImeasuresthatarestronglylinkedtoeconomicperformanceinKnackandKeefer[1995]arenotsigni®cantpre-dictorsofinvestmentandgrowthinthesmallsampleusedhere.Theªcon®denceingovernmentºindexconstructedfromWVSitemsdoesnotpredictinvestmentrates,andisofonlyborderlinesigni®canceinagrowthequation(slope.019).Thus,wecannotdirectlyexplainthepathsbywhicheconomicperfor-manceisrelatedtoTRUSTandCIVICinTableI.Neitherhaveweaddressedempiricallythepossibilityofreversecausality,asthebehaviorofgovernmentscanin¯uencelevelsoftrustandciviccooperation[Gambetta1988,pp.158±163;Hardin1992].Forex-ample,wherethepoliceviolatetraf®clawsandnorms,coopera-tiveequilibriaamongdriverswilltendtounravel.Finally,TableVresultsaremuchweakerwhenCIVICissubstitutedforourprimaryindicator,TRUST.VII.DETERMINANTSOFRUSTANDOOPERATIONUsingthesurveyresultsandotherdatasources,weunder-takeapreliminaryinvestigationofthedeterminantsoftrustandciviccooperation,anissueofspecialinteresttopolicymakers.Wetesttheeffectsofgroupmemberships,incomeinequalityandethnicpolarization,formalinstitutionsforprotectingpropertyandcontractrights,percapitaincome,andeducationrates.InSectionVwearguethatassociationalactivityhasambigu-ouseffectsoneconomicperformance:itmayincreasetrust,butalsofacilitaterent-seeking.Here,wearguethateventherela-DOESSOCIALCAPITALHAVEAPAYOFF?32.WhenGROUPSissubstitutedforTRUSTinTableV,itissigni®cantlyandpositivelyrelatedtotwoofthefourdependentvariables:ªcon®denceingov-ernment,ºconsistentwithPutnam's[1993]results,andtheBERIcontracten-forceabilitymeasure. tionshipbetweenassociationsandtrustisdouble-edged.Involve-mentinformalorinformalgroupsandassociations(thedensityofnetworksofhorizontalassociation)maybuildtrustandcivic-mindedbehavior[Putnam1993].Theunderlyingideaisthatsuchrelationshipseitherbreakdowninformationasymmetriesorcreateapatternofrepeatedinteractionsthatallowself-enforcingagreementstobereached;peoplewhobelongtosuchnetworksªtrustºotherswhobelongtothem,andaremorelikelytoexhibitcivicbehavior.However,manysocietiesarepolarizedbyethnic,political,re-ligious,orincomedifferences.Associationsinsuchsocietieswilloftenformalong(e.g.)ethniclines.Theserelativelyhomogeneousassociationsinheterogeneoussocietiesmaystrengthentrustandcooperativenormswithinanethnicgroup,butweakentrustandcooperationbetweenthosegroups.Thiseffectcreatesthepoten-tialforanegativerelationshipbetweenhorizontalassociationsandtrustornormsofciviccooperationwhenmeasuredatthenationallevel.Additionally,participationinformalgroupsmayconstituteonlyasmallpercentageofthesocialinteractionsthatcanbuildtrustandcooperativenorms.Wethereforehypothesizethatatbesthorizontalassociationsareonlyweaklyrelatedtotrustornormsofciviccooperation.Evenifapositivecorrelationwerefound,itcouldbeattributabletoreversecausality:high-trustindividualsaremorelikelytojoinformalassociationsinwhichmanytransactionsatleastinitiallywillinvolveinteractingwithstrangers.Socialpolarizationisasecondpossibledeterminantoftrustandciviccooperation.Polarizationbyde®nitionimpliesgreaterdistancesbetweenpreferencesofindividualsinasociety.Individ-ualsandgroupsinsuchasocietyhaveagreaterincentivetore-negeonpolicyagreements[KeeferandKnack1995].Whenpolicycoalitionsareunstable,trustrelationsamongindividualsoftenbreakdown.Inpolarizedsocieties,individualsarelesslikelytosharecommonbackgroundsandmutualexpectationsaboutbe-havior,soitismoredif®culttomakeself-enforcingagreements.Platteau[1994]discussesexamplesoftheimportanceofreli-giousandlinguistichomogeneityinfacilitatingtradeinWestQUARTERLYJOURNALOFECONOMICS33.Weingast[1993]providesmultipleexamples,includingtheformerYugo-slavia,arguingthatinstitutionalguaranteesofpolicybargainsrelatedtointer-ethnicdistributionalissuesdisappearedwiththefallofcommunistregimes,leadinginshortordertothebreakdownoftrustbetweenindividualsfromdiffer-entethnicgroups. Africa.Finally,polarizationcanincreaserent-seekingactivi-tiesÐwhetherlegal,throughthepoliticalsystem,orillegal,throughtheftÐthatunderminetrust.Throughanyofthesevari-ouschannels,polarizationcanerodetrustandweakencoopera-tivenorms.Weproxyforpolarizationusingincomeinequalityandethnichomogeneity.Athirdpossibledeterminantoftrustandcivicnormsistheformalinstitutionalstructureofacountry.Wheretheseinstitu-tionsrestrainarbitrarybehaviorbygovernmentleaders,forex-ample,governmentpoliciesarelikelytobemorecredible(ortrustworthy).Ifformalinstitutionsenforceprivateagreementsandlawsmoreeffectively,trustandadherencetocivicnormsamongprivatecitizensmaybestrengthened:ªInaHobbesianview...trustisunderwrittenbyastronggovernmenttoenforcecontractsandtopunishtheft.Withoutsuchagovernment,coop-erationwouldbenearlyimpossibleandtrustwouldbeirrationalº[Hardin1992,p.161].Thereislittleempiricalortheoreticalresearchtoprovideguidanceastootherdeterminantsoftrustandthestrengthofcivicnorms.Wecontrolherefor1980percapitaincome,andschoolenrollmentin1960,recognizingthatcausalityislikelynotTablesVIandVIIreportresultsonthedeterminantsoftrustandciviccooperation.Theeffectofincomeisalwayspositive,asexpected,andissigni®cantinmostoftheTRUSTequations.Theeffectofsecondaryeducationisuniformlypositive,andoftensig-ni®cant.Coef®cientsforprimaryeducationarealwaysnegative,andsometimessigni®cant.DOESSOCIALCAPITALHAVEAPAYOFF?34.Therelationshipbetweentrustandciviccooperation,ontheonehand,andpropertyrightsandgovernmentperformanceontheother,islikelytobecom-plex,witheachin¯uencingtheother.FormalinstitutionscanbesubstitutesforÐaswellascausesofÐtrustandciviccooperation.Societieswithlowtrustrequiremorerobustformalinstitutionsiftheyaretoundertaketheexchangesthatarecrucialtogrowth.35.Usingindividual-leveldataongroupmembershipsfortheUnitedStates,Germany,andSwedenfromtheWVSandothersources,StolleandRochon[1996]®ndmembershipsincreasetrust,controllingforage,education,andincome.They®ndmixedevidenceonwhethergroupsthataremorediverse(intermsofage,socioeconomicstatus,ideology,andsex)producemoretrustamongmembers.Theyacknowledgethattheirresultscouldbein¯uencedbyself-selection.Helli-well[1996b],usingindividual-levelWVSdatafortheUnitedStatesandCanada,®ndsgroupmembersaremoretrusting,controllingforeducationandregion.He®ndsnodifferenceinthestrengthofthiseffectacrossdifferenttypesofgroups.Notethatsuchindividual-levelanalyses,unlikeouraggregate-levelanalysis,can-notcapturetheexternaleffectsofgroupmembershipsonthetrustandtrustwor-thinessofnonmembers. TABLEVIETERMINANTSOFRUSTANDOOPERATION:EMBERSHIPSANDOVERNMENTAL Equation DependentvariableTRUSTCIVICTRUSTCIVICTRUSTCIVICTRUST Constant34.39139.05334.74639.17223.42540.21725.921(16.275)(3.600)(14.415)(3.387)(7.397)(1.951)(3.497)GDP801.4030.2131.0470.1221.1180.0131.256(0.604)(0.137)(0.645)(0.131)(0.547)(0.163)(0.532)(13.107)(3.168)(11.780)(2.954)(6.082)(1.604)(8.227)SEC6022.8243.16818.9872.68525.0082.70016.540(11.682)(2.288)(10.923)(2.218)(10.484)(2.351)(10.263)GROUPS5.052(6.378)(0.903)3.119(8.568)(1.128)O-GROUPS27.0193.896(14.035)(1.345)1.5220.345(0.933)(0.203)Independenceof .46.30.51.42.51.24.609.941.619.461.469.771.759.05Mean,D.V.37.439.237.439.235.839.435.6N26262626292928 White-correctedstandarderrorsareshowninparentheses.Icelandismissingdataonindependenceofcourts.India,Nigeria,andTurkeyaremissingdataongroupmemberships. AsPutnam[1995b]hasnoted,thereisastrongbivariatere-lationshipbetweentrustandgroupmembershipsacrosscoun-However,we®ndinequations1and2ofTableVIthathorizontalassociationshavenosigni®canteffectonTRUSTorCIVICwhenincomeandeducationarecontrolledfor,consistentwiththeambiguitiesnotedaboveontheeffectsofgroupmember-ships.Apossibleexplanationforthisresultisthatgroupswithsocialgoalsarebetterthanthosewithpoliticalgoalsatbuildingtrustandcooperativehabits.Inequations3and4weinvestigatethispossibilityusingthedistinctionmadeinSectionVbetweenPutnamgroupsandOlsongroups.Membershipsinthemorepo-liticallyorientedªOlsonºgroupsareassociated,surprisingly,withstrongertrustandcooperativeattitudes,whiletheªPutnamºDOESSOCIALCAPITALHAVEAPAYOFF?TABLEVIIETERMINANTSOFRUSTANDOOPERATION:OLARIZATION Equation1234 DependentvariableTRUSTCIVICTRUSTCIVIC Constant57.93847.70425.71739.883(12.108)(2.112)(4.645)(1.710)GDP801.3410.0541.7760.224(0.516)(0.143)(0.473)(0.150)(7.137)(1.511)(8.219)(1.766)SEC6017.4250.9065.9681.313(9.566)(2.667)(9.350)(2.413)Gini(income)(0.173)(0.027)Ethnic0.3490.064homogeneity(0.107)(0.023)Lawyers(1963)1.2540.012(0.194)(0.068) .55.31.73.309.531.687.321.68Mean,D.V.35.639.435.839.4N28282929 White-correctedstandarderrorsareshowninparentheses.36.However,themarkeddeclineintrustintheUnitedStatessincethe1960shasnotbeenaccompaniedbyadeclineingroupmemberships.Evenifthetwovariableswerecorrelated,causalityarguablyrunsprimarilyfromtrusttogroupsratherthantheotherwayaround,asmoretrustingindividualsmayhaveagreaterpropensitytoaf®liatewithstrangersingroups. groupshavenoeffectontrust(equation3)andappeartoreduceciviccooperation(equation4).ªExecutiveconstraintsºispositivelyand(marginally)sig-ni®cantlyrelatedtobothTRUST(equation5)andCIVIC(equa-tion6).Eachone-pointriseintheseven-pointExecutiveConstraintsscaleisassociatedwithariseinTRUSTof1.5per-centagepoints.Asasecondproxyforformalinstitutions,weuseavariablecalledªIndependenceofthecourtsºfromHumana[1987].Thisvariableisafour-pointscale,withhighernumbersindicatinggreaterindependenceofthejudiciaryfromthechiefexecutive.ItisstronglyrelatedtoTRUST(equation7),witheachone-pointriseassociatedwithaneight-percentage-pointincreaseinTRUST.ThereisnorelationshipbetweenindependenceofthejudiciaryandCIVIC,however(resultsnotshown).Theseresultsonformalinstitutions,althoughperhapsfraughtwithmultipledirectionsofcausation,constituteimportantevidenceforthein-teractionofformalandinformalinstitutions.TableVIIanalyzestherelationshipbetweenpolarizationandourmeasuresofsocialcapital.TheGinicoef®cientforincomein-equalityisstronglyassociatedwithlowertrustandciviccoopera-tion(equations1and2).BothTRUSTandCIVICincreasesig-ni®cantlywithethnichomogeneity(equations3and4).Witheachthree-pointincreaseinthepercentagebelongingtothelargestethnicgroup,TRUSTrisesbyonepoint.Each®fteen-pointin-creaseinhomogeneityincreasesCIVICbyonepoint.arealsoassociatedwithlowtrust,whetherasasignalofso-cialpolarizationorthroughsomeotherchannel.Eachone-percentage-pointincreaseinlawstudents(asapercentageofallpostsecondarystudentsin1963)isassociatedwithadeclineinTRUSTofmorethanonepercentagepoint(equation3).LawyersareunrelatedtoCIVIC,however(equation4).Thissectionoffersevidenceforthefollowingpropositions.In-comeequalityandchecksonexecutivepowerareassociatedwithhigherlevelsoftrustandstrongernormsofciviccooperation.EthnicandlinguisticdivisionscoincidewithweakenedtrustandQUARTERLYJOURNALOFECONOMICS37.TheseresultsÐaswiththoseinSectionVongroupmembershipsandeconomicperformanceÐarenotsensitivetodeletingreligiousorganizationsfrom38.Usingindividual-levelWVSdata,Helliwell[1996b]foundthatU.S.andCanadianrespondentsclassifyingthemselves®rstintermsofethnicity(e.g.,Asian-AmericanorFrench-Canadian)andonlysecondarilyintermsofnationality(e.g.,AmericanorCanadian)werelesstrusting,controllingforeducationand civicnorms.Groupmemberships,incontrast,areunrelatedtotrustandnorms.Wealsoexploredtheeffectsofseveralotherpossibledetermi-nantsoftrustandciviccooperationwhicharelesswell-developedconceptuallyhereforreasonsofspace.Urbanization,population,populationdensity,andgovernmentsizeallprovedinsigni®-Putnam[1993]viewsªhierarchicalºreligionssuchasCa-tholicismasbeinginimicaltohorizontaltiesandtrust.Inglehart[1990]andFukuyama[1995],citingWeber,linkProtestantismtohighertrust.LaPortaetal.[1997]reportthattrustislowerincountrieswithªdominanthierarchicalreligionsºasmeasuredbyofthepopulationthatisCatholic,OrthodoxChristian,orMos-lem.We®ndthatProtestantismisassociatedwithsigni®cantlygreatertrust:each®ve-percentage-pointriseinthenumberofProtestants(fromTaylorandJodice[1983])isassociatedwithaone-pointriseinTRUST.WhenpercentProtestant,percentCath-olic,andpercentMoslemareallincludedinaTRUSTregression,coef®cientsforthelattertworegressorsaresigni®cantlynega-tive,whilepercentProtestantispositivebutnotsigni®cant.VIII.CONCLUSIONANDMPLICATIONSThenotionofsocialcapitalisemergingwithgreaterfre-quencyindiscussionsofdevelopment,whetherofpoorcountriesorofchronicallypoorareasofindustrializedcountries.Theelas-ticityofthetermsocialcapitalhasmadeprogressinthesediscus-sionsdif®cult.Nevertheless,allconceptshaveincommontheideathattrustandnormsofciviccooperationareessentialtowell-functioningsocieties,andtotheeconomicprogressofthosesocieties.Thispapermakesthreecontributionstothesediscus-sions.First,itprovidessomeideaoftheimportanceofthesedi-mensionsofsocialcapital,supplyingthestrongestevidencetodatethattrustandciviccooperationhavesigni®cantimpactsonaggregateeconomicactivity.Oursecondcontributionisanegativeone.DisputesaboutDOESSOCIALCAPITALHAVEAPAYOFF?39.Asdiscussedabove,theWVSindicatorsappeartomeasuregeneralizedtrustandcivicnorms.Inequalityandethnicheterogeneitymaywellstrengthencooperationandtrustwithincertainlimitedgroups.40.Taylor[1976]claimsthatlargegovernmentsªcrowdoutºnorms.Bu-chanan[1965]amongothersarguesthattheforceofethicalrulesweakenswithlargernumbers.41.Resultsareavailableonrequest. theroleofsocialcapitalarisewhenoneattemptstode®nehowtrustandnormsofciviccooperationinasocietycanbestrength-ened.Putnam[1993]hassuggestedthatdensehorizontalnet-worksreinforcetrustandcivicnorms.However,we®ndthathorizontalnetworksÐasmeasuredbymembershipingroupsÐareunrelatedtotrustandcivicnorms(controllingforeducationandincome)andtoeconomicperformance.ThestrongerimpactoneconomicperformanceoftrustandcivicnormssuggeststhatifdecliningsocialcapitalintheUnitedStateshasadverseimpli-cationsforgrowth,itistheerosionoftrustandciviccooperationasdocumentedbyKnack[1992]thatareofgreaterconcernthanthedeclineinassociationallifeemphasizedbyPutnam[1995a,Ourthirdcontributionistodemonstratetheimportanceoftwosourcesoftrustandcivicnorms.Lowsocialpolarization,andformalinstitutionalrulesthatconstrainthegovernmentfromactingarbitrarily,areassociatedwiththedevelopmentofcooper-ativenormsandtrust.Theseresultscarryseveralimplicationsforpolicy.Trust'sre-lationshiptogrowthinourstudyisespeciallylargeinpoorercountries,whichmaybeattributabletotheirlesswell-developed®nancialsectors,insecurepropertyrights,andunreliableen-forceabilityofcontracts.Interpersonaltrustseemstobemoreim-portantinfacilitatingeconomicactivitywhereformalsubstitutesareunavailable.This®ndingsuggeststhatwhereinterpersonaltrustislowandunlikelytoimproverapidly,institutionalreformsprovidingbetterformalmechanismsforthereliableenforcementofcontractsandaccesstocreditareevenmoreimportantthanwheretrustishigher.Promotinghorizontalassociationsthroughencouragingtheformationofandparticipationingroupsmaybecounterproduc-tive,accordingtoour®ndings.Groupmembershipsarenotdi-rectlyrelatedtoeconomicperformance,andareunrelatedtotrust.Onthepositiveside,secondaryeducationisshownheretobeassociatedwithtrust,althoughdeterminingcausalityisprob-lematic.Reformsinthisarea,alreadystronglysupportedbyin-ternationalagenciesanddonorsforotherreasons,mayimproveQUARTERLYJOURNALOFECONOMICS42.Moreover,surveyevidenceindicatesthatmembershipingroupshasbeenfairlystableinrecentyearsintheUnitedStates.Putnam'sªdeclineºresultsfromadjustingfortheriseineducationlevelsovertime,whichhejusti®esonthebasisofapositivecross-sectionalrelationshipbetweenyearsofeducationandgroup A Con®denceEthnicCountryTrustCivicGroupsO-groupsP-groupsingovernmenthomogeneity Norway61.240.751.090.240.630.7298Finland57.240.640.400.060.290.6690Sweden57.141.571.080.270.640.6588Denmark56.040.340.970.240.610.7695Canada49.639.741.030.520.290.7070Australia47.838.271.010.450.350.6498Netherlands46.238.361.110.530.250.6399U.S.45.440.551.500.830.420.6181U.K.44.440.070.920.380.360.5482Switzerland43.240.890.730.220.29.72Iceland41.641.071.700.630.760.73100Japan40.841.790.380.140.210.4699Ireland40.237.510.850.480.240.7394Korea,South38.039.640.470.310.120.61100Spain34.538.750.450.230.140.5575India34.342.65...0.6772Austria31.841.450.760.260.370.6099SouthAfrica30.536.990.840.520.160.7073Belgium30.238.080.560.260.200.6057Germany29.839.830.740.220.350.5499Argentina27.039.500.470.190.210.2891Italy26.341.230.380.120.200.4499France24.836.260.420.160.180.6294Nigeria22.939.19...0.7332Chile22.736.800.590.330.140.6478Portugal21.436.890.430.210.140.4599Mexico17.734.550.570.280.140.5358Turkey10.042.43...0.6182Brazil6.737.580.680.310.160.5588 economicperformancethroughincreasinginterpersonaltrust.Our®ndingsalsoreinforcethecaseforreducingincomedispari-tiesindevelopingcountries.ThesepoliciesÐoftenadvocatedonothergroundsÐhavenotprovedeasytoimplement,however.Thebuildingofsocialcapitalhasbeenbroachedasasigni®cantnewroadtodevelopment.Ourresultssuggestthatthisroadisnolessdif®cultthanthemoreheavilytraveledones.NIVERSITYANDIRIS,UNIVERSITYOFARYLANDArrow,Kenneth,ªGiftsandExchanges,ºPhilosophyandPublicAffairs,I(1972),Banks,ArthurS.,andRobertS.Textor,ACross-PolitySurvey(Cambridge,MA:MITPress,1963).Barro,Robert,ªEconomicGrowthinaCross-SectionofCountries,ºJournalofEconomics,CVI(1991),407±444.Barro,Robert,andJong-WhaLee,ªInternationalComparisonsofEducationalAt-JournalofMonetaryEconomics,XXXII(1993),363±394.Berg,Andrew,andJeffreySachs,ªTheDebtCrisis:StructuralExplanationsofCountryPerformance,ºJournalofDevelopmentEconomics,XXIX(1988),Bergsten,GordonS.,ªOntheRoleofSocialNormsinaMarketEconomy,º,XLV(1985),113±137.Besley,Timothy,ªNonmarketInstitutionsforCreditandRiskSharinginLow-IncomeCountries,ºJournalofEconomicPerspectives,IX(1995),115±127.Buchanan,James,ªEthicalRules,ExpectedValues,andLargeNumbers,ºLXXVI(1965),1±13.Clague,Christopher,ªRuleObedience,OrganizationalLoyalty,andEconomicDe-JournalofInstitutionalandTheoreticalEconomics,CXLIX(1993),393±414.Coleman,JamesS.,ªSocialCapitalintheCreationofHumanCapital,ºJournalofSociology,XCIV(1988),S95-S120.ÐÐ,FoundationsofSocialTheory(Cambridge,MA:HarvardUniversityPress,DeLong,J.Bradford,andLawrenceH.Summers,ªEquipmentInvestmentandEconomicGrowth,ºQuarterlyJournalofEconomics,CVI(1991),445±502.Fukuyama,Francis,Trust:TheSocialVirtuesandtheCreationofProsperityYork:TheFreePress,1995).Galor,Oded,andJ.Zeira,ªIncomeDistributionandMacroeconomics,ºReviewofEconomicStudies,LX(1993),35±52.Gambetta,Diego,Trust:MakingandBreakingCooperativeRelationsBlackwell,1988).Granovetter,Mark,ªTheStrengthofWeakTies,ºAmericanJournalofSociologyLXXVIII(1973),1360±1380.ÐÐ,ªEconomicActionandSocialStructure:TheProblemofEmbeddedness,ºAmericanJournalofSociology,XCI(1985),481±510.Greif,Avner,ªReputationandCoalitionsinMedievalTrade:MaghribiTraders,ºJournalofEconomicHistory,LIX(1989),857±882.ÐÐ,ªContracting,Enforcement,andEf®ciency:EconomicsBeyondtheLaw,ºPresentedattheWorldBankConferenceonDevelopmentEconomics,Wash-ington,DC,April1996.Gurr,TedRobert,PolityII:PoliticalStructuresandRegimeChange,1800±1986(AnnArbor,MI:Inter-UniversityConsortiumforPoliticalandSocialRe-search,1990).QUARTERLYJOURNALOFECONOMICS Hall,RobertE.,andCharlesI.Jones,ªTheProductivityofNations,ºmimeo,Stan-fordUniversity,1996.Hardin,Russell,CollectiveAction(Baltimore,MD:ResourcesfortheFuture,ÐÐ,ªTheStreet-LevelEpistemologyofTrust,ºAnalyse&Kritik,XIV(1992),Helliwell,John,ªEconomicGrowthandSocialCapitalinAsia,ºNBERWorkingPaperNo.5470,1996a.ÐÐ,ªDoBordersMatterforSocialCapital?EconomicGrowthandCivicCultureinU.S.StatesandCanadianProvinces,ºNBERWorkingPaperNo.5863,Helliwell,John,andRobertPutnam,ªEconomicGrowthandSocialCapitalinItaly,ºEasternEconomicJournal,XXI(1995),295±307.Humana,Charles,WorldHumanRightsGuide(London,UK:HodderandStough-ton,1987).Inglehart,Ronald,CultureShiftinAdvancedIndustrialSociety(Princeton,NJ:PrincetonUniversityPress,1990).ÐÐ,CodebookforWorldValuesSurveys(AnnArbor,MI:InstituteforSocialRe-search,1994).Keefer,Philip,andStephenKnack,ªWhyDon'tPoorCountriesCatchUp?ACross-NationalTestofanInstitutionalExplanation.ºIRISCenterWorkingPaperNo.60,UniversityofMaryland,CollegePark,1993.Keefer,Philip,andStephenKnack,ªPolarization,PropertyRightsandtheLinksbetweenInequalityandGrowth,ºIRISCenterWorkingPaperNo.153,Uni-versityofMaryland,CollegePark,1995.King,RobertG.,andRossLevine,ªFinanceandGrowth:SchumpeterMightBeQuarterlyJournalofEconomics,CVIII(1993),717±737.Knack,Stephen,ªCivicNorms,SocialSanctions,andVoterTurnout,ºandSocietyIV(1992),133±156.Knack,Stephen,andPhilipKeefer,ªInstitutionsandEconomicPerformance:Cross-CountryTestsUsingAlternativeInstitutionalMeasures,ºandPolitics,VII(1995),207±227.Kyriacou,GeorgeA.,ªLevelandGrowthEffectsofHumanCapital,ºC.V.StarrWorkingPaperNo.91-26,1991.LaPorta,Rafael,FlorencioLopez-de-Silanes,AndreiShleifer,RobertW.Vishny,ªTrustinLargeOrganizations,ºAmericanEconomicReview,(1997),333±38.Loury,Glenn,ªADynamicTheoryofRacialIncomeDifferences,ºinP.A.Wallace,andA.LeMund,eds.,Women,MinoritiesandEmploymentDiscrimination(Lexington,MA.:LexingtonBooks,1977).Milanovic,Branko,ªDeterminantsofCross-CountryIncomeInequality:AnAug-mentedKuznets'Hypothesis,ºWorldBankPolicyResearchWorkingPaperNo.1246,1994.Mueller,Dennis,PublicChoiceII(Cambridge,UK:CambridgeUniversityPress,Murphy,Kevin,AndreiShleifer,andRobertW.Vishny,ªTheAllocationofTalent:ImplicationsforGrowth,ºQuarterlyJournalofEconomics,CVI(1991),Narayan,Deepa,andLantPritchett,ªCentsandSociability:HouseholdIncomeandSocialCapitalinRuralTanzania,ºWorldBankPolicyResearchWorkingPaperNo.1796,1997.Olson,Mancur,TheRiseandDeclineofNations(NewHaven,CT:YaleUniversityPress,1982).Platteau,Jean-Philippe,ªBehindtheMarketStageWhereRealSocietiesExist,ºJournalofDevelopmentStudies,XXX(1994),533±77,753±817.Putnam,Robert(withRobertLeonardiandRaffaellaY.Nanetti),MakingDemoc-racyWork(Princeton,NJ:PrincetonUniversityPress,1993).ÐÐ,ªBowlingAlone:America'sDecliningSocialCapital,ºJournalofDemocracyVI(1995a),65±78.ÐÐ,ªTuningIn,TuningOut:TheStrangeDisappearanceofSocialCapitalinPS:PoliticalScienceandPolitics,XXVIII(1995b),664±683.Sachs,JeffreyandAndrewWarner,ªEconomicReformandtheProcessofGlobalBrookingsPapersonEconomicActivity(1995:1),1±95.DOESSOCIALCAPITALHAVEAPAYOFF? Solow,Robert,ªButVerify,ºTheNewRepublic(September111995),p.36.Stolle,Dietlind,andThomasR.Rochon,ªSocialCapital,ButHow?AssociationsandtheCreationofSocialCapital,ºPresentedattheConferenceforEuro-peanistsinChicago,March,1996.Sullivan,MichaelJ.,MeasuringGlobalValues(NewYork:Greenwood,1991).Summers,Robert,andAlanHeston,ªThePennWorldTable:AnExpandedSetofInternationalComparisons,1950±1988,ºQuarterlyJournalofEconomicsCVI(1991),327±368.Taylor,Michael,AnarchyandCooperation(NewYork:Wiley,1976).Taylor,Charles,andDavidJodice,WorldHandbookofPoliticalandSocialIndica-,3rdedition(NewHaven,CT:YaleUniversityPress,1983).Uslaner,Eric,ªFaith,Hope,andCharity:SocialCapital,Trust,andCollectiveAction,ºmimeo,UniversityofMaryland,CollegePark,1995.Wallerstein,Michael,ªUnionOrganizationinAdvancedIndustrialDemocracies,ºAmericanPoliticalScienceReview,LXXXIII(1989),481±501.Weingast,Barry,ªThePoliticalFoundationsofDemocracyandtheRuleofLaw,ºIRISCenterReprintNo.56,UniversityofMaryland,CollegePark,1993.White,HalbertL.,ªAHeteroskedasticity-ConsistentCovarianceMatrixEstima-torandaDirectTestforHeteroskedasticity,º,XLVIII(1980),QUARTERLYJOURNALOFECONOMICS