ENOUG OR DON AT ALL RI NEEZY AND LDO USTICHINI Economist usuall assum tha moneta - PDF document

PAYENOUGHORDON’TPAYATALL*URIGNEEZYANDALDORUSTICHINIEconomistsusuallyassumethatmonetaryincentivesimproveperformance,andpsychologistsclaimthattheoppositemayhappen.Wepresentanddiscussasetofexperimentsdesignedtotestthesecontrastingclaims.Wefoundthattheeffectofmonetarycompensationonperformancewasnotmonotonic.Inthetreatmentsinwhichmoneywasoffered,alargeramountyieldedahigherperformance.However,offeringmoneydidnotalwaysproduceanimprovement:subjectswhowereofferedmonetaryincentivesperformedmorepoorlythanthosewhowereofferednocompensation.Severalpossibleinterpreta-tionsoftheresultsarediscussed.I.IAccordingtostandardeconomicreasoning,anincreaseintheŽnancialincentivesprovidedforanactivitywillimproveperfor-mance.Thispredictionisaconclusionofverybasicassumptionsineconomictheory:performanceispositivelyrelatedtoeffort;effortisunpleasant,andmoneyisgood.Weshouldthereforeobserveamonotonicandincreasingrelationshipbetweenmone-tarycompensationforanactivityandtheperformancelevelofthatactivity.1Themainaimofthispaperistoprovideatestofthisprediction,inacontrolledlaboratoryenvironment,whichpaysparticularattentiontothecomparisonbetweenthetotalabsenceandthepresenceofmonetaryrewards.Ourmainresultisthatperformancevariesinanonmonotonicwaywithincentives.Themonotonicrelationshippredictedbythetheorymayfailinconcretesituations,eitherinreallifeorinexperiments,becausefactorsdifferentfrommoneyandeffortmayenterintothedecisionoftheagent.Forinstance,apersonmaybereluctanttoworkforverysmallcompensationbecausethisfactmightsignalhisgeneralwillingnesstoacceptasmallwage,andthusweakenhisfuturebargainingposition.Adifferentreason,whichismorecommonlysuggestedintheeconomicsliterature,isthatpeoplefollowsocialnormsthatprescribeabehaviorindependentlyofany*WethankColinCamerer,GaryCharness,DouglasDejong,MartinDufwen-berg,ErnstFehr,theeditorEdwardGlaeser,GeorgKirchsteiger,MatthewRabin,RobertRosenthal,ArielRubinstein,twoanonymousreferees,andananonymouseditorforhelpfulcommentsandsuggestions.SpecialthanksgotoEricvanDammeforhiscommentsaswellasforhishelpinraisingthemoneyneededforthe1.Thisargumentrequiresthatchangesincompensationaresmallenoughsothattheincomeeffectsarenegligibleascomparedwiththesubstitutioneffect.Inourexperimentsandintheliteraturewediscuss,thisisalwaysthecase.r2000bythePresidentandFellowsofHarvardCollegeandtheMassachusettsInstituteofTechnology.TheQuarterlyJournalofEconomics,August2000791 monetarycompensation.Donatingbloodmaybeconsideredadutytothecommunitythatoneshouldperformwhenpossible.Amonetarycompensationmaydestroythissenseofdutyandproduceanetdecreaseintheaction.2Adifferentsocialnormthatmaybeunderminedbymonetarycompensationisreciprocity.SupposethatanactionisoriginallyperformedinreturnforapreviousbeneŽt,butthatmoneyispaidforit.Thenthecompensa-tionratherthanthereciprocitywillprobablybetakenasamotivationforthataction.Theincentiveforreciprocityisde-stroyed,andtheactionbecomeslessappealingonitsownmerits.3Aswewillsee,noneoftheseexplanationsseemsadequateforourresults.Theissueoftheeffectofrewardsonbehaviorhasbeendebatedinpsychologythroughoutthepastfourdecades.Behavior-isttheoryhadthesameopinionasstandardeconomics,althoughforcompletelydifferentreasons.Accordingtoinstrumentalcondi-tioning,rewardofferedforanactivitywhichisinitselfneutralorevenmildlyunpleasant,willeventuallyassociateapositivevalencetothatactivity.Sointhelongrunapastrewardhasapositiveeffectontheperformanceofthatactivity.4Thisconclusionofbehavioristpsychologywaschallengedatthebeginningoftheseventiesbythecognitivepsychologyschool.Theyofferedanalternativeview:anactivityhasamotivationofitsown,independentofanyreward,calledintrinsicmotivation.Areward,differentfromthisintrinsicmotivation(inparticular,butnotonly,amonetaryreward)mayreplacetheintrinsicmotiva-tion.5Theneteffectmaybeareductionoftheoverallmotivation,2.Titmuss{1970}claimedthatmonetarycompensationmightunderminethesenseofcivicduty.HeconsidersthespeciŽcexampleofblooddonationinTitmuss{1971},wherehearguesthattheintroductionofmonetarycompensationwillmakethequalityofblooddonatedworse.Arrow{1972}discusseshisthesis:hepredictsthatanincreaseinpricewilleventuallyproduceanincreaseinsupply.Morerecently,theworkofFreyandseveralcoauthors{Frey1994;Frey,Oberholzer-Gee,andEichenberger1996;FreyandOberholzer-Gee1997}haspresentedanddefendedtheideathatpriceincentivesmaycrowdoutmotivation.Kohn{1993a,1993b}hascriticizedincentiveplansbecausetheymakepeoplelessenthusiasticabouttheirwork.3.IntheexperimentsofFehr,Gachter,andKirchsteiger{1996}theintroduc-tionofexplicitincentivesreducedtheperformanceofworkersinaŽrm-workerrelationshipbecausethenormofreciprocitywascompromised.ThispointisdiscussedinFehrandGachter{1998},andmoreextensively,withadditionalevidence,inFehrandRockenbach{2000}.InhisŽeldstudyonmanagementbehavior,Bewley{1995,1997}notesthatreal-lifemanagersknowwellthatitisnotwisetodependonŽnancialincentivesaloneasmotivators.4.AclearexpositionofthispointofviewisinSkinner{1953}.5.DeŽnitionsandmeasurementofintrinsicmotivationarestillcontroversial:butabasicconditionfortheexistence(andempiricalevidence)ofintrinsicmotivationisthattheactivityshouldbeexercisedevenwhenrewardisabsent.QUARTERLYJOURNALOFECONOMICS792 andhenceareductionoftheactivityitself.Wecanformulatethesameideainthelanguage,morefamiliartoeconomists,ofpreferences:iftherewarddirectlyaffectstheutilityofanindi-vidualinanegativeway(becauseitreducestheintrinsicmotiva-tion),thenperformancemaydeclinewiththeincreaseinmone-taryincentive.6Themainconclusionsofthesestudieswerethatpositiverewards,inparticularmonetaryrewards,haveanegativeeffectonintrinsicmotivation.Ifapersonisrewardedforperform-inganinterestingactivity,hisintrinsicmotivationdecreases.ThenegativeeffectissigniŽcantonlyiftherewardiscontingentontheperformance;subjectswhoarepaidaŽxedpositiveamount,independentoftheirperformance,donotdisplayreductioninintrinsicmotivation.71.TwoExperimentalTestsInthispaperwetestexperimentallytheeffectsofmonetaryincentivesonperformance.IntheŽrstexperimentagroupof160studentsattheUniversityofHaifawereaskedtoanswerasetof50questionstakenfromanIQtest.ThestudentswerepaidaŽxedamountof60NIS(NewIsraeliShekel8)forparticipatingintheexperiment.Theyweredividedintofourdifferentgroups,corre-spondingtofourdifferenttreatments.TheŽrstgroupwassimplyaskedtoanswerasmanyquestionsastheycould.Tosubjectsinthesecondgroupwepromisedanadditionalpaymentof10centsofaNISpereachquestionthattheyansweredcorrectly.Tosubjectsinthethirdgroupwepromisedanamountof1NIS,andtothoseinthefourthgroupanamountof3NISperquestion.Weobservedthattheaveragenumberofquestionsansweredcor-rectlydeclinedfromslightlymorethan28intheŽrstgroupto23questionsinthesecond.Thenumberincreasedtomorethan34inthethirdgroup,andwasstableat34inthefourthgroup.Aswe ThethesiswassuggestedinDeci{1971},andfurtherdiscussedamongothersinDeci{1975},Deci,Cascio,andKrusell{1973},andKruglansky,Alon,andLewis{1972}.5Aratherlargesetofexperimentsshowedthataloweringindeedoccurred:anearlyoverviewofthisliteratureanditsexperimentalevidenceisinLepperandGreene{1978}.6.Thisisthemodelofmotivationcrowding-out,presentedinFreyandOberholzer-Gee{1997},forinstance.TheirmodelisdiscussedinSectionIVbelow.7.Aconsensusopinioninexperimentalpsychologyisfarfrombeingreached.CameronandPierce{1994}andEisenbergerandCameron{1996}haveprovidedmeta-studiesonthetopicoftheeffectofrewardsonmotivation,evaluatingmorethantwodecadesofstudiesontheissue.Theyalsoprovideausefulreviewoftheliteraturewehavediscussed.TheŽnalconclusionisstillunclear:forinstance,theyŽndanegativeeffectoftangiblerewards,andapositiveeffectofverbalrewards.8.Atthetimeoftheexperiment,NIS3.55$1.PAYENOUGHORDON’TPAYATALL793 argueinthenextsection,wherethedetailsoftheresultsarepresented,theeffectwasstatisticallysigniŽcant.Oursecondexperimentinvolvedhighschoolchildren,againinIsrael,whoweredoingvolunteerwork.Everyyear,onapredeterminedday,studentsgofromhousetohousecollectingmonetarydonationsthathouseholdsmaketosocietiesforcancerresearch,assistancetodisabledchildren,andsoon.Inourexperiment,wedivided180ofthesestudentsintothreegroups.TheŽrstservedasacontrolgroup:wesimplygaveasmallspeechtothesubjectsrecallingtheimportanceoftheactivitytheyweregoingtoperform.Tothesecondgroup,inadditiontothespeech,wemadeapromisetopay1percentofthetotalamountcollected.Tothethirdwepromisedtopay10percentoftheamountcollected.InbothcasesitwasmadeclearthatthepaymentwasŽnancedbyus,andnotbythesocieties.InthissecondtestweobservedthattheamountcollectedwassmallerinthesecondthanintheŽrstgroup.Theaverageamountforthethirdgroupwashigherthaninthesecondgroup,butstilllowerthanintheŽrst.Again,theresultswerestatisticallysigniŽcant.Sincemonetaryincentives,atlowvalues,wouldappeartohaveadetrimentaleffectonperformance,weinvestigatedwhetherthesubjectswereawareofthis.InaŽnalsetofexperimentsweaskedsubjectstodecidewhatincentivetheywouldprovidetoothersubjectsworkingontheirbehalf.Inthisexperimentsub-jectsintheŽrstgroupwerepaidaccordingtotheperformanceofthesubjectstowhomtheyweregivingtheincentive.Theycouldchoosebetweenano-rewardandalow-reward.Theincentivetheydecidedtopaywassubtractedfromtheirpayoff.Themajoritychosethelowincentive.Thisincentivewasmorecostly,andaswehaveseenwasinducingaworseperformance,soitisthewrongcontractintheprincipal-agentrelationship.Ourexperimentsfocusonperformance,amatterofcentralinterestforeconomics.Thisvariablealsoprovidesanobjective,quantitativemeasureoftheeffect.Inparticular,westudythedifferentialeffectofsmallandlargerewards.Incontrast,theexperimentalpsychologyliterature,motivatedbythedistinctionbetweenintrinsicandextrinsicmotivation,hasdifferentmea-suresoftheintrinsicmotivationtobethedependentvariable.Themeasurescommonlyusedaretheamountoftimefreelyspentontheactivityandthereportbythesubjectsonthemotivation.9Thereisaseconddifferencebetweenourstudyandthe9.ThesearethetwovariablesthatCameronandPierce{1994}andEisen-bergerandCameron{1996}considerintheirmeta-studies.QUARTERLYJOURNALOFECONOMICS794 experimentsperformedinthepsychologyliterature.Psycholo-gistsstudybehaviormodiŽcationthroughconditioning(inthecaseofthebehavioristschool)orlearning(forthecognitiveschool).Wedonot.Toillustratethedifference,wemayconsidertheclassicexperimentreportedinDeci{1971}.Hehadcollegestudentsplaywithapuzzleinthreesuccessivesessions.IntheŽrstsessionparticipantswerelefttoplayfreely.Inthesecondsessionsubjectsinonegroupreceivedpaymentiftheysolvedthepuzzle,whilethecontrolgroupdidnot.Inathirdsessionthesubjectswereagainlefttoplayfreely.TheamountoftimespentonfreeactivityintheŽrstandthirdsessionwastakenasameasureofintrinsicmotivation.Decifoundthatinthethirdsessiontheexperimentalgroupspentlesstimethanthecontrolgroupplayingwiththepuzzle,andheconcludedthattherewardofferedhaddecreasedtheintrinsicmotivationofsubjectsintheŽrstgroupoverthethreesessions.Westudythebehavioralresponsetodifferentrewardsinasingle-stagesetup.Thecompari-sonisacrossindividuals,notacrosssuccessiveperiodsforthesameindividualfollowingthereward.Finally,wetestboththeeffectsoftheintroductionofreward,andtheeffectsofanincreaseinthereward.Inexperimentaleconomicsthequestionis‘‘howmuchclosertothepredictionofeconomicsandgametheorydoesanincreaseinmonetaryrewardsbringthebehaviorofsubjects?’’Thefocusisonchangesinrewards,alwayskeepingthesepositive.10Webroadenedthistoallowforastart-upeffectandshowedthattheoverallpatterncanbenonmonotonic,thusaccountingforexperimentalevidenceinpsychologythatseemedinconsistentwiththeŽndingsoftheexperimentaleconomicsliterature.II.THEEXPERIMENTS1.TheIQExperiment:DesignTheexperimentwasconductedattheUniversityofHaifa.Thesubjectswere160maleandfemaleundergraduatestudents10.FreyandcoauthorshaveconductedŽeldstudiesoftheeffectonintrinsicmotivationfromthepointofviewofeconomists.Inparticular,FreyandOberholzer-Gee{1997}studyanswerstoaquestionnaireaskingpeopleabouttheirattitudetowardanuclearwasterepositoryintheirregion.Theirstudyisbasedonstatedbehaviorinhypotheticalchoicesituations,inaninterviewfacetoface.Sincenobodylikestobeconsideredgreedy,theinconsequentialstatementsmightbebiasedbythedesiretoshowoff.Thequestionswereaskedsequentiallyofthesamesubjects.Thedependentvariablewasnotperformance,butwillingnesstoex-changetheinconveniencesofnuclearwasteagainstmoney.Intheirreview,SmithandWalker{1993}Žndthatincreasingrewardsbringsthebehaviorclosertothepredictionsofeconomictheory,andreducesthevariancearoundthemean.Inthemorerecentreview,CamererandHogarth{1999}concludethattheevidenceismorecontroversial.PAYENOUGHORDON’TPAYATALL795 fromallŽeldsofstudy,withanaverageageof23.Thesubjectstookpartintheexperimentdividedintofourdifferentgroupsof40studentseach,correspondingtofourdifferenttreatmentsthatwedescribebelow.Atthebeginningoftheexperiment,eachstudentwasprom-isedaŽxedpaymentofNIS60forparticipation.Theywerethentoldthattheexperimentwouldtake45minutes,andtheywouldbeaskedtoansweraquizconsistingof50problemstakenoutofapsychometrictestusedtoscanapplicantstotheuniversity.ThistestissimilartotheGMATexam:theparticipantsweretoldthatthiswasasortofIQtest.Theproblemsinthequizwerechosentomaketheprobabilityofacorrectanswerdependmostlyoneffort.Inparticular,emphasiswasplacedonquestionsinvolvingreason-ingandcomputationratherthangeneralknowledge.Inthefourdifferenttreatmentssubjectswerepromiseddifferentadditionalpaymentsforeachcorrectanswer.IntheŽrstgroupnomentionwasmadeofanyadditionalpayment.Inthesecondgroupsubjectswerepromisedanadditionalpaymentof10centsofaNISperquestionansweredcorrectly.Theamountpromisedwasofstep1NISand3NIS,respectively,forthethirdandfourthgroup.Aftertheintroduction,thequizwasdistributed.Participantswerenotallowedanymaterialontheirtablesexceptthequizitself,andweretoldthatonlythosewhostayeduntiltheendoftheexperimentwouldbepaid.Noclarifyingquestionsbystudentswereallowedduringthistime.Attheendoftheexperimentparticipantsweretoldwhereandwhentogotocollecttheirearnings.TheinstructionsarepresentedinAppendix1.2.TheIQExperiment:ResultsAppendix2reportsthenumberofcorrectanswersforeachsubject.SummarystatisticsarepresentedinTableI.Theaveragenumberofcorrectanswerswas28.4outof50questionsintheŽrstgroup.Theaveragewas23.07inthesecondgroup,wheresubjectsweregettinganadditional10centspercorrectanswer.Theaveragewasthenhigher,evencomparedwiththeŽrstgroup,andequalto34.7inthethirdgroup(oneNIS),and34.1inthefourthgroup(threeNIS).AnonparametricMann-WhitneyUtestbasedonrankscanbeusedtoinvestigatewhetherthesampleofcorrectanswerscamefrompopulationswiththesamedistribution.InTableIIwereporttheresultsofapairwisecomparisonofthedifferenttreatments.AQUARTERLYJOURNALOFECONOMICS796 numberintheintersectionofrowandacolumnindicates,forthecorrespondingpairoftreatments,theprobabilityofgettingatleastasextremeabsolutevaluesoftheteststatisticasweobserve,giventhatthetwosamplescomefromthesamedistribution.ThedifferencebetweenthedistributionsinthezeropaymentandinthelowpaymentissigniŽcant,ata.9levelofsigniŽcance.Thedifferencebetweenthedistributionsinthehighpayofftreatments(1NISand3NIS)isnotsigniŽcant.Finally,thedistributionsintheselattertreatmentsaresigniŽcantlyhigherthanthedistributionsinthecaseofthezeroand10centsmarginalpayoffs.Forinstance,thep-valuesforthecomparisonbetweenthe1NIStreatmentandthezeroand10centspaymentare.0687and.0004,respectively.Thedifferenceamongtreatmentspersistsifwecomparesubgroupswithsimilarperformance.Forinstance,werankedtheTABLEISUMMARYSTATISTICSFORTHEIQEXPERIMENT,FORTHEDIFFERENTTREATMENTSTheLowerFractionistheFractionofSubjectsWhoGaveaNumberofCorrectAnswersLessthan16 Nopayment10centsNIS1NIS3Average28.423.0734.734.1Standarddeviation13.9214.728.889.42Median31263737Averagetop203934.942.3541.6Standarddev.top205.256.793.634.18Averagebottom2017.811.2527.0526.6Standarddev.top2011.5610.225.076.8220thquantile4035444380thquantile2002625Lowerfraction15%27.5%0%0% TABLEIIMANN-WHITNEYUTESTSBASEDONRANKSWITHPAIRWISECOMPARISONSOFMEDIANSOFCORRECTANSWERSBYTREATMENT Nopayment10centsNIS110cents.0875——NIS1.0687.0004—NIS3.0708.0006.6964* (Prob.. z ,wherezistheteststatistic).Anasteriskindicatesthatforthatcomparisonwecannotreject(ata.9levelofsigniŽcance)thehypothesisthatthetwosamplescomefromthesamedistribution.PAYENOUGHORDON’TPAYATALL797 tophalfandthebottomhalfoftheparticipantsaccordingtoperformance,andwethentestedthesigniŽcanceofthedifferencebetweendistributions.TheresultsarepresentedinTablesIIIaandIIIb.Asimilarcomparisoncanbemadebetweenthedistributionofthetoptenparticipants.Thedifferencebetweenthe1NISand3NISisclearlynotsigniŽcant.Thedifferencebetweenthe10centsand1NISaswellas3NISissigniŽcant.Thediversityinindividualperformancemaybeduetodifferencesinseveraldifferentfactors,suchasskill,generalknowledge,andpreferencesformoneyandeffort.Ourresultsseemtoindicatethattheeffectoftheintroductionofmonetaryincentivesandtheirchangeaffectinthesamewayindividualswithdifferentcharacteristics,ashighertalent,orhigherwilling-nesstoputouteffort.4.TheDonationExperiment:DesignToillustratethenextstudyapremiseisnecessary.InIsraelafew‘‘donationdays’’takeplaceeveryyear.Eachofthesedaysisdevotedtoasocietythatcollectsdonationsfromthepublicforsomepurpose,suchascancerresearch,disabledchildren,etc.High-schoolstudentsgofromdoortodoortocollectthedonations.TABLEIIIaMANN-WHITNEYUTESTSBASEDONRANKSWITHPAIRWISECOMPARISONSOFMEDIANSOFCORRECTANSWERSBYTREATMENT,USINGTHEBEST20OBSERVATIONSOFEACHTREATMENT Nopayment10centsNIS110cents.0381——NIS1.0146.0003—NIS3.0339.0008.4222* TABLEIIIbMANN-WHITNEYUTESTSBASEDONRANKSWITHPAIRWISECOMPARISONSOFMEDIANSOFCORRECTANSWERSBYTREATMENT,USINGTHE20WORSTOBSERVATIONSOFEACHTREATMENT Nopayment10centsNIS110cents.0299——NIS1.0120.0000—NIS3.0270.0001.7552* QUARTERLYJOURNALOFECONOMICS798 Normally,thestudentsareorganizedintogroupsaccordingtotheclassinwhichtheystudy,andeachgroupisthendividedintopairsofstudentswhoworktogetherasateam.Eachpairreceivesacertainnumberofcoupons,whichserveasreceiptsforthedonors.Theamountcollectedbyeachpaironthedonationdaydependsmostlyontheeffortinvested:themorehousestheyvisit,themoremoneytheycollect.Thisisespeciallytruebecausethestudentsdonothaveto‘‘sell’’thedonation,sincemostpeoplearealreadyfamiliarwithitfromtelevisionannouncementsandWehad180high-schoolstudentsaroundtheageof16participatingwiththreetreatmentlevels.Thereweretwogroupsofparticipants,eachwithŽfteencouples,foreachtreatment.EachpairreceivedcouponsamountingtoNIS500altogether.Inthediscussionthatfollowswereportjointlytheresultsforthetwogroupsateachtreatmentlevel.IntheŽrsttreatmentanexperimenterappearedbeforeeachofthegroupsandtoldthemabouttheimportanceofthedonationtheyweretocollect,andthatthesocietywishedtomotivatethemtocollectasmuchmoneyaspossible.Theyweretoldthattheresultsofthecollectionwouldbepublished,sothattheamountcollectedbyeachpairwouldbecomepublicknowledge.Thesecondtreatmentwasconductedsimilarly:butafterthesamespeech,eachpairwaspromised1percentoftheamountthatthetwoofthemcollected.Finally,inthethirdtreatmenteachpairwaspromised10percentoftheamounttheycollected.Inthesecondandthirdtreatmentsitwasmadeclearthatthepaymentwasmadefromfundsadditionaltothedonation,pro-videdbyus,andthatthesocietieswouldreceivethetotalamountofthedonationasusual.Theactivityofcollectingdonationsthenwentonasusual,accordingtotheproceduredescribedabove.5.TheDonationExperiment:ResultsThepreciseamountcollectedbyeachgroupforthethreedifferenttreatmentsisreportedinAppendix3.WereportthemostimportantsummarystatisticsinTableIV.Theaverageamountcollectedwas238.67over500forgroupsintheŽrsttreatment(withnopayment).Theaveragefellto153.67inthesecondgroup.Itwas219.33,higherthaninthesecondtreatment(butlowerthantheŽrst)inthethirdtreatment.TotestthesigniŽcanceoftheseresults,weusethenonpara-metricMann-WhitneyUtestbasedonrankstoinvestigatePAYENOUGHORDON’TPAYATALL799 whethertheamountsofmoneydonatedcamefromthesamedistribution.TheresultsofthetestarereportedinTableV.ThedifferencebetweentheaveragecollectionintheŽrstandinthesecondgroupissigniŽcant,ata.9levelofsigniŽcance.Whenthepayoffincreasedto10percentoftheamountcollected,theaveragecollectionwas219.33.TheamountscollectedinthistreatmentweresigniŽcantlyhigherthantheamountscollectedinthe1percenttreatment,butnotsigniŽcantlyhigherthantheamountscollectedwhennopayoffwasgiven.Wecomparedthetoptencollectionsineachtreatment.Thedifferencebetweentheamountscollectedinthe1percenttreat-mentandtheamountscollectedintheothertwotreatmentswassigniŽcant.AsintheIQexperiment,thisresultindicatesthatthedifferencebetweentreatmentsisuniformamongsubjectswithhighandthosewithlowperformance.III.HOWSUBJECTSPERCEIVETHEEFFECTSOFMONETARYINCENTIVESTheevidencewehavepresentedseemstoindicatethattheeffectofmonetaryincentivescanbe,forsmallamounts,detrimen-TABLEIVSUMMARYSTATISTICSFORTHEDONATIONEXPERIMENT,FORTHEDIFFERENTTREATMENTS Nopayment1percent10percentAverage238.6153.6219.3Standarddeviation165.77143.15158.09Median200150180Averagetop20375.33272348Standarddeviationtop20111.9298.64110.46Averagebottom2010235.3390.66Standarddeviationbottom2066.1352.0863.9720thquantile10005080thquantile450250400 TABLEVMANN-WHITNEYUTESTSBASEDONRANKSWITHPAIRWISECOMPARISONSOFMEDIANSOFAMOUNTSOFMONEYCOLLECTEDBYTREATMENT Nopayment1percent1percent.0977—10percent.7054*.0515 QUARTERLYJOURNALOFECONOMICS800 taltoperformance.Sinceincentivesareusuallydesignedtoaffectperformanceinanoptimalway,itisinterestingtoinvestigatewhetherthiseffectisanticipated.Toaddressthisquestion,weconductedatestbasedonboththeIQanddonationexperiments.Inbothcasessubjectswerepromisedapaymentproportionaltotheperformanceofadifferentperson(their‘‘agent’’),andtheyhadtochoosetheincentiveschemefortheagent.1.TheIQExperimentwithPrincipalandAgentsThesubjectsintheexperimentwere53studentsintheroleofprincipals.Theyweretoldthattheywouldbematchedwithanotherplayer.Theyweregivenashortintroductioninwhichtheyreceivedanexplanationofthetaskthattheir‘‘agents’’wouldperform,namely,answeringquestionsfromtheadmissiontestthatthesubjectstookintheIQexperiment.Theywerethentoldthateachoneofthemwouldbepaid1NISforeverycorrectanswergivenbyhisagent.Theprincipalshadtochoosewhetherapaymentof10centsofNISorazeropaymentwastobemadetotheiragentforeverycorrectanswer.Thispaymentwouldbepaidoutoftheamountof1NISpaidtotheprincipal.Theprincipalsweretoldthattheagentwouldknowinadvancehowmuchhewasgoingtobepaidforeverycorrectanswer,butthathewouldnotknowthattheprincipalhadtodecideŽrstwhethertopayhimnothingor10cents.Hewouldnotevenknowthataprincipalexisted.Thiswastheonlydecisiontheprincipalshadtomake.Attheendoftheexperimentparticipantsweretoldwhereandwhentogotocollecttheirearnings.Outofthe53subjects46subjects,aproportionof87percent,chosetopay10centsforeverycorrectansweroftheagent.Withthischoicetheyreducedtheirincomeintwoways:byprovidingapaymenttotheagent,andbyreducingtheperformanceoftheagentbecauseofthenegativeeffectoflowrewards.2.TheDonationExperimentwithPrincipalandAgentsInthedonationexperimentwealsohadagroupofstudentswhoplayedtheroleof‘‘principals.’’Theyweretoldthattheywouldbematchedrandomlywithonepairwhohadalreadycollectedmoney,andwouldbepaid5percentofwhatthispairhadcollected.TheprincipalshadtodecidewhethertheywantedustochoosethepairfromthegroupthatdidnotreceiveanypayofforPAYENOUGHORDON’TPAYATALL801 fromthegroupthatreceived1percentofwhattheyhavecollected.Thepaymenttotheagentwasmadeoutofthe5percent.TheresultsconŽrmedwhatweobservedintheprevioustest.Outofthe25participants,19(thatis,aproportionof76percent)preferredtobematchedwithanagentwhowaspaid1percentoftheamounthecollected.IV.INTERPRETATIONOFTHERESULTS1.SummaryoftheResultsThefollowingfactsseemtobeclearfromtheanalysisoftheresults.Ifwecomparethetreatmentinwhichmonetarycompen-sationwasnotevenmentionedwiththeoneinwhichitwas,thenwemayconcludethatthemonetarycompensationproducesareductionintheperformance.Butinthesetoftreatmentsinwhichamonetarycompensationisoffered,ahighermonetaryincentiveproducedahigherperformance.Thisresultindicatesadiscontinuityatthezeropaymentoftheeffectofmonetaryincentive:forallpositivebutsmallenoughcompensations,thereisareductioninperformanceascomparedwiththezerocompen-sation,or,better,withthelackofanymentionofcompensation.Also,subjectsinthesamepopulationasthosewhoexhibitthisbehaviordonothaveaclearorstrongperceptionofthisdisconti-nuity.Theyseemtoindiscriminatelyapplytherule,whichisvalidintheregionofpositivecompensations,thathigherpaymentprovideshigherperformance.2.IntrinsicandExtrinsicMotivationAŽrstinterpretationoftheseresultsispossiblealongthelinesprovidedbycognitivepsychologists:theactivityhasanintrinsicmotivation,andtheintroductionofamonetaryreward,whichisanextrinsicmotivation,displacestheŽrst,andtheneteffectmaybeareductionintheactivity.WehavetoseewhetherthisexplanationisjustiŽedinthecaseofourexperiments.AŽrstversionoftheinterpretationisthatthesubjectisweightingintrinsicandextrinsicmotivation,andtheintrinsicmotivationisreduceddirectlybythemonetarycompensation.AmodelpresentingthisversionformallyisprovidedinFreyandOberholzer-Gee{1997}.Inthismodeltheagenthasautilityfunctionoftheactivityaandthemonetaryrewardr.Thefunctionhasoneterm,u(a,r),withthestandardfeatures,addedtoasecondQUARTERLYJOURNALOFECONOMICS802 termdescribingtheintrinsicmotivation,m(a,r).Thislattertermcapturesthenegativeeffectofmonetaryrewardonintrinsicmotivation,becausethederivativeofmwithrespecttorisnegative.Theoptimallevelofactivityasafunctionoftherewardristhencomputedintheusualway.Comparativestaticanalysisshowsthattheoptimalactivitymaybedecreasinginr.Thismodelisinsightful,andhasthevirtueofreducingthediscussiontoaspecialcaseofstandardeconomicanalysis.How-ever,itseemsunabletoexplainthediscontinuity:asmallchangefromzerotoaverylowpaymentshouldreduce(iftheutilityfunctionhascurvature)asmallchangeintheactivity.Asecondversionisbasedonacognitiveexplanationofthedisplacement,basedonattributiontheoryalongthelinesŽrstsuggestedinFestinger{1957}andHeider{1958}.ThisexplanationofthedisplacementispresentedinBem{1965,1967}andKelley{1967,1971}.Forinstance,Bemsuggeststhatpeopleinterprettheiractionsasanyoutsiderdoes,bylookingatthereasonsandmotivesforwhattheydo.Ifasubjectobserveshimselfdoinganactionwhennoexogenouscompensationisprovided,thenheinterpretshismotiveasanintrinsicmotivation.Ifamonetarycompensationisprovided,thenthesamesubjectwillinterprethismotiveasbeingthemonetaryreward.Inthisversionoftheinterpretationthediscontinuityisexplained:theintroductionofsmallcompensationhastheeffectofchangingtheperception,andthischangeislarge,independentoftheamount.Butthisexplanationisonlyappropriateinsequentialexperiments,whereasubjecthastheopportunityofobservinghimselfactingforsomeincentive,modifyhisself-perception,andthenchangehisbehavior.Itislessappropriateinoursingle-stagesetup.Thenextinterpretationseemstousthemostconvincing.3.IncompleteContractsThecontractdescribingtheexperimentisanincompletecontract.Allsuccessiveinstructionsgivenbytheexperimenterprovidepartialcompletionofitsterms.Whennomonetarycompensationforthecorrectanswersisprovided,thecontractisprobablyinterpretedas‘‘Apaymentof60NISisprovidedforparticipationintheexperiment,andInowknowthatthisparticipationconsistsofansweringthequestions.’’Some(butnotall)ofthesubjectsfeltthatitwastheirsideofthebargaintoanswerthosequestions.PAYENOUGHORDON’TPAYATALL803 Theintroductionofadditionalcompensationforcorrectan-swerschangestheperceptionofthecontract:‘‘SixtyNISwerepaidforshowingup.Theactivityofansweringthequestionsisnowpaidbyrate.’’Thisrateisnowthereferencepointtodecidetheappropriateintensityoftheactivity.Whentherateislow(10cents),theactivityislow:thereisnonecessaryconnectionwiththelevelofactivityprovidedinthatdifferentsituation,whereansweringquestionsis‘‘due.’’Animportantpredictiongivenbyourexplanation(butalsobythesecondversionofthepreviousinterpretation)isthatthechangeinperception,oncerealized,ishardtoreverse.ThisisindeedthecaseinarelatedŽeldstudy{GneezyandRustichini2000}thatweconductedontheeffectofpenalties.Thestudywasconductedinagroupofday-carecenters,whereparentswerecominglaterthantheduetimetocollecttheirchildren.InthetestgroupweintroducedatthefourthweekofthestudyaŽneforlatearrival.TheŽnewas10NISforadelayoftenminutesormore.TheŽnewascancelledattheseventeenthweek.TheeffectwasanincreaseinthenumberoflatearrivalsaftertheintroductionoftheŽne.InFigureIwereporttheaveragedelaysforthecontrolandthetestgroup,inthetwenty-weekperiod.Thisresultisconsistentwithashiftinperceptionofanincompletecontract,asintheinterpretationwehavejustpro-vided.Furthersupportforthisexplanationcomesfromthebehaviorofparentsaftertheseventeenthweek:thenumberofFIGUREIAverageNumberofLate-ArrivingParentsEachWeek,byGroupTypeQUARTERLYJOURNALOFECONOMICS804 latearrivalsstayedconstant,asonewouldexpectaftertheperceptionofthecontracthaschanged.AninterestingextensionoftheIQexperimentdesignedtotestthistheorywouldbetotaketheparticipantsoftheIQquizexperiment,where10centshasbeenofferedpereachquestionansweredcorrectly,andinasecondexperimentaskthemtoperformthesametask,fornocompensation.Weconjecturethattheperformancewillbeworsethantheoneofsubjectswhohavebeenofferedzerocompensationfromthestart,andofcourseworsethantheperformanceofthesameparticipantsintheŽrst4.HowSmallIsaSmallAmount?Inourexperimentsthesubjectswhowerepaid10centsfor1NISforeachrightanswergaveaworseperformancethandidthosewhowerenotgivenanypayment.Tencentsmaysoundlikeverysmallcompensation,almostinsultingandthereforethepracticalimplicationsofourŽndingsminimal;buttwoqualiŽca-tionsarenecessary.First,notallsmallcompensationsmaybeconsideredinsulting.Forexample,considerthepracticeofpayingbackasmallamountforrecyclingasoftdrinkbottle,whichiscommoninmanyEuropeancountries.Anecdotalevidencesug-geststhatpeoplearelesswillingtorecyclewhenthissmallcompensationisoffered,thantheyareinplaceswherenomoneyisoffered,andfailuretorecycleissimplyconsideredbadbehavior.Itisunlikelythattheamountofferedisconsideredinsulting.Adifferentexplanationinthiscasemightbethatpeopleareafraidoflooking‘‘cheap’’formakingtheeffortofrecyclingtocollectthesmallamount.Second,insultingcompensationsarenotnecessarilysmallcompensations.Thereasonisthattheamountofmoneyofferedchangestheperceptionthatpeoplehaveof‘‘whatthecontractisabout.’’Inparticular,itmaynotbesafetoassumethataddinganincentiveleavestheutilityoftheotherincentivesunchanged.Acertainamountofmonetarycompensationmaybeperceivedastoosmallwhencomparedwiththeotherrelevantfactors,evenifitisnottoosmallinitself.Wemaythinkofreal-lifesituationswhereanontrivialamountofmoneymaysounddisproportionatelysmallcomparedwithotherfactors.Forinstance,anincreaseinsalaryby$200permonthtoaprofessor,ascompensationforasmallerPAYENOUGHORDON’TPAYATALL805 office,maybeworsethannocompensation.Similarly,smallhonorariaforseminarspeakersmaybecounterproductive.11Thisfactorislikelytobemoreimportantwhenfactorslikehealthorreputationareatstake.Sowhileinourexperimentsitisclearthat‘‘toosmall’’issomewherebetween10centsand1NIS,theexactdeterminationofthisquantityinexperimentalorreal-lifesituationsislikelytobedifficultandsubtle.Ourtheoryhasaninterestingimplicationforexperimentaleconomics.Itseemswidelyacceptednowthatasufficientlyhighrewardissufficienttoreducethevarianceoftheobservedbehavioraroundthemeanvalueofthebehavioraspredictedbythetheory.Theseeminglynaturalimplicationthatasmallrewardsimplyproduceshighervariancemightbefalse:atthelowendofthescaleofrewards,theremightbeparadoxicalbehavior,ofthetypeobservedinourexperiments.Infact,thebehaviorwithsmallpaymentmaybe,asitisinourexperiments,moredistantfromthepredictionthanthebehaviorwithzeroproportionalpayment.Thismaybeimportantincaseswherelargepaymentsareimpossible,forexample,becauseofethicalorlegalreasons.Therulethat‘‘asmallpaymentisbetterthannothing’’mightbeabadrule.125.TheTwoExperimentsThereisofcourseanimportantdifferencebetweenthetwoexperiments.Inthedonationstudythereisanintrinsicmotiva-tionthatisclearlyidentiŽable:thealtruisticreasonthatis,afterall,themotivationforthestudentsbeforetheybecamesubjectsinourexperiment.Thisdifference,forinstance,mayexplainasigniŽcantdifferenceinbehaviorbetweenthetwostudies.IntheIQtesttheperformancewithasubstantialpayment(1NIS)reachesalevelwellabovetheoneinthetreatmentwithzeropayment(34.7correctanswersagainst28.4).Inthedonationstudy,evenapaymentof25NISperpersondoesnotbringtheperformancebacktothelevelachievedinthezeropaymentgroup(219.3NISagainst238.67).Ofcourseanevenlargerpaymentmightbeenough.Sotheissueofasystematicdifferenceinbehaviorinthetwoenvironmentsseemsinteresting,andthepointisworthfurtherstudy.11.Wethankananonymouseditorforsuggestingthispoint.12.AsimilarpointisarguedinthereviewbyCamererandHogarth{1999}.QUARTERLYJOURNALOFECONOMICS806 V.CONCLUSIONSInthispaperwehaveprovidedquantitativelypreciseevi-dence,inacontrolledenvironment,oftheeffectoftheintroductionofmonetarycompensationonperformance,whichincludesaprecisecomparisonofthecasesinwhichtherewardwasgivenindifferentquantitiesornotgivenatall.Theresulthasbeenthattheusualpredictionofhigherperformancewithhighercompensa-tion,whenoneisoffered,hasbeenconŽrmed:buttheperformancemaybelowerbecauseoftheintroductionofthecompensation.Onthebasisofthispreciseevidencewemaybeginthesearchforasatisfactoryexplanation.Furtherresearch,intheoryandexperiments,isnecessary,andwehaveindicatedsomeofthepromisingdirections.Inthemeantime,themostconvincingexplanationseemstoustobebasedoncognitivearguments:contracts,socialorprivate,areusuallyincomplete,andregulateaninteractioninasituationofincompleteinformation.TheintroductionofarewardmodiŽessomeofthetermsofthecontract,butalsoprovidesinformation.Thenewbehaviorpro-ducedbythecontractisaresponsetothecombinationofanewpayoffstructureandthenewinformation.ThedifficultyisthatthestandardBayesianupdatingofinformationseemsunsuitedforthissituation.APPENDIX1:INSTRUCTIONSFORTHEIQEXPERIMENTITheinstructionsaresimple,andifyoufollowthemcarefullyyoumayearnaconsiderableamountofmoney.Theexperimentwilltakeabout45minutes.Intheexperimentyouareaskedtoansweraquizof50problemstakenfromapsychometrictestusedtoscanapplicantstotheuniversity.ItisasortofIQtest.YouwillbepaidNIS60forshowinguptotheexperiment. Thefollowingsentencewasnotincludedintreatment1:‘‘Inaddition,youwillbepaidNIS0.1’’(intreatment2,NIS1intreatment3,NIS3intreatment4)foreverycorrectansweryougive. Themoneywillbepaidtoyou,privatelyandincash,attheendoftheexperiment.Doyouhaveanyquestions?PAYENOUGHORDON’TPAYATALL807 APPENDIX2:THENUMBEROFCORRECTANSWERSGIVENINTHEIQEXPERIMENTBYPARTICIPANTSACCORDINGTOTREATMENTS Obs.#NopaymentObs.#10centsObs#NIS1Obs.#NIS314941508149121502484244824712250348434483471234744544438446124455424540854612544642463986451264474247368744127448404835884412843937493589441294210375035904313041113751349141131411237523492411323913365332934113339143654329440134391536553195401353816355630963813638173457269738137371834582698381383719345926993813937203160261003714037213161241013414137223162231023314236233163231033314336242964221043314434252965211053114533262466211063114631272367211073014731282368191082914831292369191092914928302270131102915027312271111112815126322072811228152253320730113261532534187401142315421357750115221552036376011622156203707701172215719380780118211581939079011920159174008001201716016 QUARTERLYJOURNALOFECONOMICS808 TECHNION,HAIFA,ISRAELBOSTONUNIVERSITYREFERENCESArrow,K.J.,‘‘GiftsandExchanges,’’PhilosophyandPublicAffairs,I(1972),343–362.Bem,D.J.,‘‘AnExperimentalAnalysisofSelf-Persuasion,’’JournalofExperimen-talandSocialPsychology,I(1965),199–218. ,‘‘Self-Perception:AnAlternativeInterpretationofCognitiveDissonancePhenomena,’’ PsychologicalReview, LXXIV(1967),183–200. Bewley,T.,‘‘ADepressedLaborMarketasExplainedbyParticipants,’’AmericanEconomicReview,PapersandProceedings,LXXXV(1995),250–254. ,‘‘ADepressedLaborMarketasExplainedbyParticipants,’’manuscript,DepartmentofEconomics,YaleUniversity,1997.APPENDIX3:THEAMOUNTOFMONEYCOLLECTEDBYSTUDENTSINTHEDONATIONEXPERIMENTACCORDINGTOTREATMENTS Obs.#NopaymentObs.#1%Obs.#10%10310610203206203033063204403406420580350653061003606650710037067100812038068100912039069120101304030701401115041507115012150428072150131504310073150141904412074150152004515075180162004615076200172404715077200182504818078240192504921079250203005023080250213305124081290223405224082290233505325083350244205425084380254505525085400265005630086410275005733087460285005840088500295005940089500305006050090500 PAYENOUGHORDON’TPAYATALL809 Camerer,C.F.,andR.M.Hogarth,‘‘TheEffectsofFinancialIncentivesinExperiments:AReviewandCapital-Labor-ProductionFramework,’’1999,CaliforniaInstituteofTechnologyDiscussionPaperNo.1059,in Journalof RiskandUncertainty, XIX(1999),7–42. Cameron,J.,andW.D.Pierce,‘‘Reinforcement,Reward,andIntrinsicMotivation:AMeta-Analysis,’’ ReviewofEducationalResearch, LXIV(1994),363–423. Deci,E.,‘‘EffectsofExternallyMediatedRewardsonIntrinsicMotivation,’’JournalofPersonalityandSocialPsychology,XVIII(1971),105–115. ,IntrinsicMotivation(NewYorkandLondon:PlenumPress,1975).Deci,E.,E.L.Cascio,andJ.Krusell,‘‘SexDifferences,VerbalReinforcementandIntrinsicMotivation,’’PaperpresentedattheMeetingoftheEasternPsycho-logicalAssociation,Washington,May1973.Eisenberger,R.,andJ.Cameron,‘‘DetrimentalEffectsofReward.RealityorMyth?’’ AmericanPsychologist, LI(1996),1153–1166. Fehr,E.,S.Gachter,andG.Kirchsteiger,‘‘ReciprocityasaContractEnforcementDevice,’’LXV(1996),833–860.Fehr,E.,andS.Gachter,‘‘ReciprocityandEconomics:TheEconomicImplicationsofHomoReciprocans,’’ EuropeanEconomicReview, XLII(1998),845–859. Fehr,E.,andB.Rockenbach,‘‘TheHiddenCostsofEconomicIncentives.Contrac-tualContingenciesCrowdOutVoluntaryCooperationandReciprocity,’’Univer-sityofZurichDiscussionPaper,2000.Festinger,L.,ATheoryofCognitiveDissonance(Evanston,IL:Petersen,1957).Frey,B.S.,‘‘HowIntrinsicMotivationIsCrowdedinandout,’’ Rationalityand Society, VI(1994),334–352. Frey,B.S.,F.Oberholzer-Gee,andR.Eichenberger,‘‘TheOldLadyVisitsyourBackyard:ATaleofMoralsandMarkets,’’ JournalofPoliticalEconomy, CIV (1996),1297–1313. Frey,B.S.,andF.Oberholzer-Gee,‘‘TheCostofPriceIncentives:AnEmpiricalAnalysisofMotivationCrowding-out,’’ AmericanEconomicReview, LXXXVII (1997),746–755. Gneezy,U.,andA.Rustichini,‘‘AFineIsaPrice,’’JournalofLegalStudies,XXIX(2000),1–18.Heider,F.,ThePsychologyofInterpersonalRelations(NewYorkCity,NY:Wiley,1958).Kelley,H.H.,‘‘AttributionTheoryinSocialPsychology,’’NebraskaSymposiumonXV(1967),192–238. ,AttributioninSocialInteraction(NewYorkCity,NY:GeneralLearningPress,ModuleSeries,1971).Kohn,A.,PunishedbyRewards(Boston,MA:HoughtonMifflin,1993a). ,WhyIncentivePlansCannotWork, HarvardBusinessReview, LXXI(1993b), 54–63. Kruglansky,A.W.,S.Alon,andT.Lewis,‘‘RetrospectiveMisattributionandTaskEnjoyment,’’JournalofExperimentalSocialPsychology,VIII(1972),493–501.Lepper,M.R.,andD.Greene,TheHiddenCostsofReward:NewPerspectivesinthePsychologyofHumanMotivation(Hillsdale,NJ:LawrenceErlbaumAssoci-ates;NewYork:1978).Skinner,B.F.,ScienceandHumanBehavior(NewYork,NY:Macmillan,1953).Smith,V.L.,andJ.M.Walker,‘‘MoneyRewardsandDecisionCostinExperimen-talEconomics,’’EconomicInquiry,XV(1993),245–261.Titmuss,R.M.,TheGiftRelationship(London,UK:AllenandUnwin,1970). ,‘‘TheGiftofBlood,’’Trans-action,VIII(1971);nowinThePhilosophyofWelfare,SelectedWritingsbyR.M.Titmuss,B.Abel-SmithandK.Titmusseds.(London,UK:AllenandUnwin,1987).QUARTERLYJOURNALOFECONOMICS810