WewouldliketothankDanArielyNittaiBergmanBrunoBiaisArthurCampbellErnstFehrBobGibbonsLorenzGotteBengtHolmstromJonathanLevinStevenLippmanRobertMarquezMurielNiederle 1IntroductionPreviousres ID: 254242
Download Pdf The PPT/PDF document "IsPay-for-PerformanceDetrimentaltoInnova..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
IsPay-for-PerformanceDetrimentaltoInnovation?FlorianEdereryandGustavoMansozJuly14,2012AbstractPreviousresearchineconomicsshowsthatcompensationbasedonthepay-for-performanceprincipleiseectiveininducinghigherlevels WewouldliketothankDanAriely,NittaiBergman,BrunoBiais,ArthurCampbell,ErnstFehr,BobGib-bons,LorenzGotte,BengtHolmstrom,JonathanLevin,StevenLippman,RobertMarquez,MurielNiederle, 1IntroductionPreviousresearchontheeectivenessofincentivepayarguesthatpayinganagentbasedonhisperformanceinducestheagenttoexertmoreeorttherebyimprovingproductivity.Thereisampleevidencesupportingthisthesisindierenttypesofstudies,rangingfromquasi-experimentsinempiricaldatatoeconomiceldexperimentsandlaboratorystudies.Forexample,Lazear(2000)showsthattheproductivityofwindshieldinstallersinSafeliteGlassCorporationincreasedwhenmanagementchangedtheircompensationfromxedwagestopiece-ratepay.Shearer(2004)ndssimilarevidenceinarandomizedeldexperimentwithCanadiantreeplanters.Dickinson(1999)showsthatsubjectsinalaboratoryexperimenttypemoreletterswhentheircompensationismoresensitivetoperformance.Asintheaboveexam-ples,mostoftheexistingevidenceoftheeectofnancialincentivesonperformancecomesfromstudyingsimpleroutinetasks,inwhicheortisthemaindeterminantofproductivity.1Incontrast,asubstantialbodyofexperimentalandeldresearchinpsychologyprovidesevidencethat,intasksthatrequireexplorationandcreativity,pay-for-performancecanactu-allyundermineperformance.McGraw(1978),McCullers(1978),Kohn(1993),andAmabile(1996)summarizethendingsofthislineofresearchbystatingthatpay-for-performanceencouragestherepetitionofwhathasworkedinthepast,butnottheexplorationofnewuntestedapproaches.Thesestudiesconcludethatintasksthatinvolvecreativityandinno-vation,monetaryincentivesshouldnotbeusedatalltomotivateagents.Takentogethertheavailableevidenceseemstosuggestthatstandardperformancepayworkswellforcertaintypesoftasks(physicaleort)butnotforothers(creativity).Inthispaperwepresentevi-dencethatisfullyinlinewiththisview,but,moreimportantly,wealsoshowthatcorrectlystructuredperformance-basednancialincentivescanmotivateinnovationincreativetasksandcanleadtosuperiorresultsthanthoseachievedintheabsenceofnancialincentives.Ourndingsprovidesupportfortheideathatforanytask\theeectsofincentivesdependonhowtheyaredesigned"(p.206)asGneezy,Meier,andRey-Biel(2011)suggestintheconclusionsoftheircomprehensivereviewarticleonincentiveeects.Inacontrolledexperimentalsettinginwhichsubjectsperformataskthatinvolvesatrade-obetweenexplorationandexploitation,weprovideevidencethatincentiveplansthattolerateearlyfailureandrewardlong-termsuccessleadtomoreinnovationandbetterper-formancethanxedwagesorstandardpay-for-performanceincentiveschemes.Wealsondevidencethatthethreatofterminationcanundermineincentivesforinnovationwhilegoldenparachutescanalleviatetheseinnovation-reducingeects.Finally,weshowthatrisk-aversion 1Detrimentalincentiveeectsoneorthavealsobeenstudiedineconomicswithmorenuancedconclusionsastowhenmonetaryrewardsleadtobetterperformance(GneezyandRustichini2000a,GneezyandRustichini2000b,FreyandJegen2001,Ariely,Gneezy,Loewenstein,andMazar2009).1 isanimportantfactorinexplainingtheseresults.Ourresultshaveimplicationsforvarioussituationsinwhichenablingentrepreneurshipisanimportantconcern.Infact,\stimulatinginnovation,creativityandenablingentrepreneurship"isatoppriorityformanagementandwidelyregardedasthe\greatesthumanresourcechallenge"facingorganizationsaccordingtoCEOsurveys.2Thesesituationsrangefromthedesignofcompensationpackagesoftopexecu-tivesandmiddlemanagersinlargecorporationstostructuringcompensationforentrepreneursinstart-upcompanies.Forexample,someofthemostinnovation-andcreativity-drivenrmsincludingpharmaceuticalandadvertisingcompanieshaveadoptedrewardsystemsthattol-erateorevenrewardthefailureofemployees.3Inthecontextofexecutivecompensation,toleranceforearlyfailurecanbeachievedthroughcommonlyusedpracticessuchasmanage-rialentrenchment,goldenparachutes,oroptionrepricing.Thesepracticesareoftencriticizedbecausetheyprotectorevenrewardthemanagerafterpoorperformanceand,therefore,caninducethemanagertoshirkordivertfundsfromthecorporation.Therehavebeensev-eralproposalstorestricttheuseofsomeofthesepractices.4Ourresultssuggestthatwhencombinedwithappropriatelong-termincentives,thesepracticescanbeeectiveinmotivatinginnovation.Regulationsthatrestricttheirusecouldthushaveanadverseeectoninnovation.Innovationisoftendenedastheproductionofknowledgethroughexperimentation(Arrow1969,Weitzman1979).AspointedoutbyMarch(1991),thecentralconcernthatariseswhenlearningthroughexperimentationisthetensionbetweentheexplorationofnewuntestedapproachesandtheexploitationofwell-knownapproaches.Asubstantialmanagementlit-eraturefollowingMarch(1991)hasinvestigatedtheorganizationalfactorsthatfavoreitherofthetwoformsofinnovationandhowtooptimallybalancethetensionbetweentheminambidextrousorganizations.5Thisliteraturehasabstractedawayfromhowdierentincen-tivesystemsmightaectthetrade-o,anissuewhichwespecicallyaddressinthepresentpaper.Weconductaseriesoflarge-scaleeconomiclaboratoryexperimentswith379subjectsinwhichsubjectsfacethetensionbetweenexplorationandexploitation.Subjectscontroltheoperationsofacomputerizedlemonadestandandmustchoosebetweenne-tuningtheprod-uctchoicedecisionsgiventothembythepreviousmanagerorchoosingadierentlocationandradicallyalteringtheproductmixtodiscoverabetterstrategyTostudytheimpactofdierentincentiveschemesonproductivityandinnovationinthe 2In\CEOChallenge2004:PerspectivesandAnalysis,"TheConferenceBoard,Report1353,theauthorsreportthatthehighestresponse(31%)amongsurveyedCEOsisthatstimulatinginnovationisofthegreatestconcerntotheircompany.3Foradiscussionofthisissueinthepopularpresssee\BetterIdeasThroughFailure:CompaniesRewardEmployeeMistakesToSpurInnovation,GetBackTheirEdge",WallStreetJournal,September27,2011.4See,forexample,BebchukandFried(2004)and\RewardsforFailure,"BritishDTIconsultation,June2003.5AcomprehensiveoverviewofthisliteraturecanbefoundinthespecialissueonorganizationallearninginManagementScience(Argote,McEvily,andReagans2003).2 abovetasks,werstconsiderthreedierentbaselinetreatmentgroups.Theonlydierencebetweenthesetreatmentgroupsisthecompensationoeredtosubjects.Subjectsintherstgroup(xedwage)receiveaxedwageineachperiodoftheexperiment.Subjectsinthesecondtreatmentgroup(pay-for-performance)aregivenastandardpay-for-performance(orprotsharing)contract,receivingaxedpercentageoftheprotsproducedduringtheexperiment.Subjectsinthethirdtreatmentgroup(exploration)areallocatedacontractthatistailoredtomotivateexploration.Theircompensationisaxedpercentageoftheprotsproducedduringthesecondhalfofeachofthetwoexperiments.6Ourmainhypothesisisthatsubjectsundertheexplorationcontractexploremoreandaremorelikelytondasuperiorstrategythansubjectsunderthexed-wageorstandardpay-for-performancecontracts.Twofeaturesoftheexplorationcontractencouragesubjectstoexplore.First,toleranceforearlyfailurepermitssubjectstofailatnocostinthersthalfoftheexperimentwhiletheyexplorenewstrategies.Second,theprospectofpayforperformancelateronencouragessubjectstolearnbetterwaysofperformingthetask.There-fore,eventhoughtheexplorationcontractcombineselementsofthexed-wageandpay-for-performancecontracts,itisplausiblethatperformanceundertheexplorationcontractissuperiortoperformanceunderboththexed-wageandthepay-for-performancecontracts.Ourresultsstronglysupportthemainhypothesisstatedabove.Wendthatsubjectsgiventheexplorationcontractendtheexperimentinthebestlocation80%ofthetime,whilesubjectsgiventhexed-wageandthepay-for-performancecontractsendtheexperimentinthebestlocationonly60%and40%ofthetime,respectively.Toexplainthesedierencesweexaminethereasonsbehindtherelativelypoorperformanceofsubjectsunderthexedwageandpay-for-performancecontracts.Althoughsubjectsgiventhexed-wagecontractexplorealot,theyarenotassystematicintheirexplorationassubjectswhoaregivenanexplorationcontract.Forexample,whenweanalyzethenotessubjectstakeinatableweprovidetothematthebeginningoftheexperiment,wendthatonly55%ofthesubjectsunderthexed-wagecontractcarefullykeeptrackoftheirchoicesandprots;whenfacingtheexplorationcontract,82%ofthesubjectskeeptrackoftheirchoicesandprotsusingthetable.Subjectsunderthepay-for-performancecontract,ontheotherhand,tendtodirecttheireorttowardsne-tuningthepreviousmanager'sproductmixinsteadofsearchingforbetterlocations.Duringtherst10periodsoftheexperiment,subjectsundertheexplorationcontractchoosealocationotherthantheinitialdefaultlocation80%ofthetime,whilesubjectsunderthepay-for-performancecontractdosoonly50%ofthetime.Dierentattitudestowardsrisk 6ThisexplorationcontractisbasedonManso(2011)whoshowsthatthecombinationoftoleranceforearlyfailureandrewardforlong-termsuccessisoptimaltomotivateinnovation.Theexplorationcontractcanbeimplementedinpracticethroughstockgrantsthatincreasewiththetenureofthemanagerinthermindependentlyofthemanager'sperformance.Alternatively,itcanbeimplementedviastockoptiongrantswithlongvestingperiods,andoptionrepricingincasethemanagerperformspoorlyearlyon.3 canalsoaecttheoutcomeunderthedierentcontracts.Wendthatriskaversionplaysanimportantroleinexplainingdierencesintheexplorationbehaviorandperformanceofthesubjectsunderthepay-for-performancecontract.Underthepay-for-performancecontract,morerisk-aversesubjectsarelesslikelytondtheoptimalstrategyandtheyobtainloweraverageprotsthanlessrisk-aversesubjects.Tostudytheeectsofterminationoninnovationandperformance,weintroducetwonewtreatmentgroupsinthelemonadestandexperiment:aterminationtreatmentgroupandaterminationwithgoldenparachutetreatmentgroup.Subjectsinbothgroupsreceivetheexplorationcontractandarealsotoldthattheexperimentwillendearlyiftheirprotsintherst10periodsarelowerthanacertainthreshold.Subjectsintheterminationwithgoldenparachutetreatmentgrouparetoldthattheywillreceiveareparationpaymentiftheexperimentendsafter10periods.Ourhypothesisisthatsubjectsintheterminationtreatmentarelesslikelytondtheoptimallocationthansubjectsintheexplorationtreatment.Wefurtherhypothesizethatsubjectsintheterminationwithgoldenparachutetreatmentgrouparemorelikelytondtheoptimallocationthansubjectsinthepureterminationtreatmentgroup.Thishypothesisissupportedbythedata:only45%ofthesubjectsintheterminationtreatmentgroupndtheoptimallocationwhereas65%ofthesubjectsintheterminationwithgoldenparachutetreatmentgroupndtheoptimallocation.Finally,wedemonstratethatourresultsarerobusttomodicationsintheexperimentaldesign.Werstaddresspotentialsignalingeectsofincentivecontractsinthelemonadestandexperiment.Inaddition,weshowthatinasecondexperiment,thegoldprospectingexperiment,whichusesthechoseneortapproachandawithin-subjectsdesignandallowsustofullycontrolthebeliefsofsubjectsinaparametrizedsetting,incentivecontractsthattoler-ateearlyfailureencourageexplorationandaresuperiortobothstandardpay-for-performanceandxedwages.Severalrecentpapersinvestigatetheeectsofinstitutionalfeaturesoninnovationsuchasdebtor-friendlybankruptcylaws(AcharyaandSubramanian2009),stringentlaborlaws(Acharya,Baghai-Wadji,andSubramanian2009),takeoverpressure(Sapra,Subramanian,andSubramanian2008,Atanassov2008,ChemmanurandTian2011),leverage(LiuandWong2011)aswellasthefailure-tolerantattitudeofinstitutionalinvestors(Aghion,Reenen,andZingales2008)andofventurecapitalists(TianandWang2010,Chemmanur,Tian,andLoutskina2011).Anotherburgeoningliteratureinvestigatestheinterplaybetweennon-competeagreementsandinnovation,inparticularinrelationtoCaliforniaandSiliconValley(Gilson1999,AlmeidaandKogut1999,SorensonandStuart2003),butcausalevidenceremainsthin(Fallick,Fleischman,andRebitzer2006,Marx,Strumsky,andFleming2009).Ourworkismostcloselyrelatedtocontributionsthatinvestigatetherelationbetween4 explicitincentivestructuresandinnovation.7Long-termincentivesfortheheadsofresearchanddevelopmentdepartments(LernerandWulf2007),goldenparachutesforCEOs(Francis,Hasan,andSharma2010),andlongerstockoptionvestingperiodlengths(YanadoriandMarler2006)areassociatedwithmore(heavilycited)patents.Finally,Azoulay,Zivin,andManso(2011)addresswhetherfundingpolicieswithtoleranceforearlyfailureandlonghori-zonstoevaluateresultsmotivatecreativityinscienticresearch.Allthesepapersprovidesomesupportforthethesisthattoleranceforearlyfailurecoupledwithrewardforlong-termsuccessmotivatesinnovation.Becausetheyusenaturallyoccurringdata,however,theyeitherrestrictthemselvestomerelyshowingcorrelationpatterns,ortheyaresubjecttothecriticismthatthevariationintheincentiveschemesmaynotbecompletelyexogenous.Inourpaper,weareabletostudytheeectsofincentivesoninnovationbyexogenouslyvaryingcompensationschemesinacontrolledlaboratoryenvironment.Acommonapproachtothestudyofincentivesusinglaboratoryexperimentsistogivesubjectsacostfunctionandrequirethemtochooseaneortlevel(Bull,Schotter,andWeigelt1987,Fehr,Gachter,andKirchsteiger1997,NalbantianandSchotter1997).MeyerandShi(1995)andBanks,Olson,andPorter(1997)investigatethetensionbetweenexploita-tionandexplorationinanexperimentalsettingusingmonetaryeortsandreportresultsinbroadaccordancewiththetheoreticalbanditmodel.Neitherofthesepapersconsiderstheeectofdierentincentiveschemesonsubjectbehavior.8Experimentalresearchershavealsoconductedincentivestudiesinboththelabandtheeldinwhichsubjectshavetoexertrealeorttocompleteroutinetaskssuchastypingletters(Dickinson1999),decodinganumberfromagridofletters(Sillamaa1999),crackingwalnuts(FahrandIrlenbusch2000),stu-inglettersintoenvelopes(FalkandIchino2006),orpickingfruit(Bandiera,Barankay,andRasul2005).Thesetasks,however,areinadequatetostudyincentivesforinnovation.Inthispaper,wethereforeintroduceanewtasktoinvestigatethecausalimpactofincentivesonexploration.2ExperimentalDesignWeestablishanenvironmentinwhichwecanmeasuretheeectsofdierentincentiveschemesoninnovationandperformance.Forthispurposeweconductexperimentsinwhichpartici-pantshavetosolvearealeorttaskinwhichthetrade-obetweenexplorationandexploita- 7RelatedtheoreticalcontributionsthatalsofocusontheinterplayofincentivesandinnovationincludeAghionandTirole(1994),Hellmann(2007)andHellmannandThiele(2011).8OtherrelatedexperimentalpapersusingsearchproblemsakintoourlemonadestandexperimentanddealingwithinnovationmodeledasacomputationallycomplexproblemincludeGabaix,Laibson,Moloche,andWeinberg(2006)andMeloso,Copic,andBossaerts(2009).5 tiontakescenterstage.2.1ProceduresandSubjectPoolTheexperimentswereprogrammedandconductedwiththesoftwarez-Tree(Fischbacher2007)attheHarvardBusinessSchoolComputerLaboratoryforEconomicResearch(HBSCLER).ParticipantswererecruitedfromtheHBSCLERsubjectpoolusinganonlinere-cruitmentsystem.Atotalof379subjectsparticipatedinourexperiments.Aftersubjectscompletedtheexperimentweelicitedtheirdegreeofriskaversionandam-biguityaversion.Wedescribetheexactproceduresforriskandambiguityaversionelicitationintheappendix.9Subjectswerethenprivatelypaid.Asessionlasted,onaverage,60minutes.Experimentalcurrencyunitscalledfrancswereused.Theexchangeratewassetat100francs=$1andtheshow-upfeewas$10.Subjectsonaverageearned$24.2.2TaskDescriptionSubjectstaketheroleofanindividualoperatingalemonadestand.Theexperimentlasts20periods.Ineachperiod,subjectsmakedecisionsonhowtorunthelemonadestand.Thesedecisionsinvolvethelocationofthestand,thesugarandthelemoncontent,thelemonadecolorandtheprice.Asisnaturalforthistypeoftask,someofthevariablesarediscretechoices(location,color)whileothersaremorecontinuous(sugar,lemoncontent,price)thusyielding6,181,806possiblecombinations.Theexactparametersofthegameareprovidedintheappendix.Attheendofeachperiod,subjectslearntheprotstheyobtainedduringthatperiod.Theyalsolearncustomerreactionsthatcontaininformationabouttheirchoices.Customerfeedbackisimplementedbyhavingthecomputerrandomlyselectoneofthecontinuouschoicevariablestoprovideabinaryfeedbacktothesubject.Thisfeedbackisonlyinformativeforthelocationinwhichthesubjectchosetosellinthecurrentperiod.10Subjectsdonotknowtheprotsassociatedwitheachoftheavailablechoices.Attachedtotheinstructionsisaletterfromthepreviousmanagerwhichisreproducedintheappendix.Thelettergiveshintstothesubjectsaboutastrategythathasworkedwellforthismanagerandoersanaccuratedescriptionofagoodbusinessstrategyforoneparticularlemonade 9Forourstatisticalanalysiswesplitsubjectsintoalessandamorerisk/ambiguity-aversegroupbasedonthemedianobservationforeachmeasureinthesample.10Forexample,ifthesubjectchoseasugarcontentthatisabovetheoptimallevelfortheparticularsaleslocation,thefeedbacktakestheform:\Manyofyourcustomersatthislocationtoldyouthatthelemonadeistoosweet."Wechosetogiveonlylimitedfeedbackbecause,inpractice,itisexpensiveforarmtocollectinformationfromcustomersandfeedbackisonlylikelytobeforthcomingforcombinationsthathavebeentriedbytherm.6 standlocation.Thestrategysuggestedbythepreviousmanagerinvolvessettingthestandinthebusinessdistrict,choosingahighlemoncontent,alowsugarcontent,ahighpriceandgreenlemonade.Themanager'sletteralsostatesthatthemanagerhastriedseveralcombinationsofvariablesinthebusinessdistrictlocation,butthathehasneverexperimentedwithsettingupthestandinadierentlocation.Itfurthersuggeststhatdierentlocationsmayrequireaverydierentstrategy.Theparticipantsintheexperimentthusfacethechoicebetweenne-tuningtheproductchoicedecisionsgiventothembythepreviousmanager(exploitation)orchoosingadierentlocationandradicallyalteringtheproductmixtodiscoveramoreprotablestrategy(ex-ploration).Thestrategyofthepreviousmanagerisnotthemostprotablestrategy.Themostprotablestrategyistosetthelemonadestandintheschooldistrict,andtochoosealowlemoncontent,ahighsugarcontent,alowpriceandpinklemonade.Thepayosinthegamewerechoseninsuchawaythatwithoutchangingthedefaultlocationtheadditionalprotsearnedfromimprovingthestrategyinthebusinessdistrictarerelativelysmall.Ontheotherhand,changingthelocationtotheschoolrequiredlargechangesinatleasttwoothervariablestoattainanequallyhighprotassuggestedbythedefaultstrategy.Inadditiontothepreviousmanager'sletter,theinstructionscontainatableinwhichsubjectscaninputtheirchoices,prots,andfeedbackineachperiod.Subjectsaretoldthattheycanusethistabletokeeptrackoftheirchoicesandoutcomes.Weusetheinformationsubjectsrecordinthistableasonemeasureoftheireortduringtheexperiment.2.3TreatmentGroupsandTestableHypothesesWeinitiallyimplementthreetreatmentconditionsinordertoexaminehowdierentincentiveschemesaectinnovationsuccess,explorationbehavior,timeallocationandeortchoices.Theonlydierencebetweenthegroupsisthewaysubjectsarecompensated.Thethreeincentiveschemesareasfollows:IncentiveScheme1(FixedWage):\Youwillbepaidaxedwageof50francsperperiod."IncentiveScheme2(Pay-for-Performance):\Youwillbepaid50%oftheprotsyoumakeduringthe20periodsoftheexperiment."IncentiveScheme3(Exploration):\Youwillbepaid50%oftheprotsyoumakeduringthelast10periodsoftheexperiment."Therewere51,46and47subjectsineachofthesethreetreatments.Ourmainhypothesisconcernstheextenttowhichthedierentpaymentschemesconsideredinourtreatmentgroupsaecttheexplorationactivityofsubjects.Inparticular,buildingonthetheoreticalpredictionsofManso(2011)wehypothesizethatsubjectsundertheexplorationcontract7 conditionshouldndtheoptimalbusinessstrategymoreoftenthansubjectsintheothertreatments.MainHypothesisSubjectsundertheexplorationcontractchooseabusinessstrategythatisclosertotheoptimalbusinessstrategythansubjectsunderthexed-wageandpay-for-performancecontracts.Theabovehypothesisconstitutesthecentralpartofthepresentpaperandmayappearsurprisingatrstglance,sincetheexplorationtreatmentisahybridofxedwageandpay-for-performancetreatments.Thus,aplausiblenullhypothesiswouldbethatthechoicesandperformanceofsubjectsintheexplorationtreatmentlieinbetweenthoseofthexedwageandthepay-for-performancetreatments.Similarly,anullhypothesisbasedonmodelsofrepeatedeort(Rogerson1985,HolmstromandMilgrom1987,Sannikov2008)wouldpredictthattheexplorationcontractinducesshirkingduringthersttenperiodsandhigheortandgoodperformanceinthelattertenperiodsoftheexperiment.Thus,incentiveschemesthattolerateearlyfailureareinadequatebecausetheyleadtolowereortandproductivitythanstandardpay-for-performanceincentiveschemes.However,asManso(2011)formallyshowsinatheoreticalmodelthatfeaturesthesametrade-obetweenexploitationandexplorationinadditiontoacostlyeortchoiceasinourexperiment,contractssuchastheexplorationcontractusedhereareeectiveinmotivatingexploration(andconsequentlyinnovation).11Therearetwoimportantreasonsforthedierenceinperformancebetweensubjectsintheexplorationtreatmentandsubjectsintheothertwotreatments.First,sincethecompensationofsubjectsunderthepay-for-performancecontractdependsontheirperformancefromtheveryrstperiod,theyarelesswillingtoexplorethansubjectsundertheexplorationcontract.Thisisbecauseasubjectwhoisgiventhepay-for-performancecontractanduseshisrstfewperiodstoexploredierentstrategies,islikelytoobtainlowerprotsandwouldonlylowerhiscompensationduringthoseperiodsascomparedwiththecompensationhewouldreceivefromthedefaultstrategy.Inanenvironmentwhereexplorationisdesirable,pay-for-performancecompensationthereforeleadstoinsucientexploration.Incontrast,theexplorationtreatmentwithitstoleranceforearlyfailureencouragessubjectstoexploreearlyon.Thus,relativetothepay-for-performancecontract,theexplorationcontracttiltsthescalestowardsexplorationforthesubjects'exploitation-explorationtrade-o.Second,subjectswhoaregivenaxedwagecontractshouldalsobewillingtoexplorebecausetheydonotfaceanycostsfromfailingwhiletheyexploredierentstrategies.Thus,onemightreasonablyexpectthatexplorationactivityisequallyhighamongsubjectsinthe 11ForafulldiscussionoftheparticularmodelandformalresultswerefertheinterestedreadertoManso(2011),SectionIV.B.Here,wefocusontheintuitionofthemainndingsthatarerelevantforthepresentpaper.8 xedwagetreatment.However,underaxed-wagecontractsubjectsdonothaveanyexplicitincentivesforgoodperformanceandtheyshouldminimizethecostlycontemplationeortnecessarytondthebestbusinessstrategy.Thus,thefailureofthexedwagecontractisduetothelackofincentivestoexertcostlyeortthatitprovidestothesubjects.Inthetwofollowingsub-hypothesesweinvestigateindetailtheproblemsofthexedwageandpay-for-performancecontractsinexplorationsettings.ExplorationSub-HypothesisSubjectsundertheexplorationcontractaremorelikelytoexplorethansubjectsunderthepay-for-performancecontract,whoaremorelikelytofocusonexploitationactivities.Sincethepay-for-performancecontractfailstoencourageexploration,wehypothesizethatsubjectswillexplorelessthansubjectsundertheexplorationcontract.Whiletheaboveexplorationhypothesisexplainsthedierentialeectsofexplorationandpay-for-performancecontracts,itdoesnotpredicthowsubjectsunderthexed-wagecontractwillbehave.Subjectsunderthexedwagecontractdonothaveexplicitincentivesforgoodperformance.Wethereforehypothesizethattheywillminimizetheireort.ShirkingSub-HypothesisSubjectsunderthexed-wagecontractexertlesseortthansubjectsundertheexplorationcontract.Whilewepredictthatsubjectsundertheexplorationcontractaremorelikelytoexplorethansubjectsunderthepay-for-performancecontractandlesslikelytoshirkthansubjectsinthexed-wagecontract,itneednotbethecasethattheyalsoproducebetteraverageperformancethansubjectsunderthesetwoothercontracts.InManso(2011),agentsareassumedtoberisk-neutral.Incontrast,inourexperimentsomeofthesubjectsarerisk-averseandthereisconsiderableheterogeneityinattitudesto-wardsrisk.Whilerisk(orloss)aversiondoesnotqualitativelychange(butactuallystrength-ens)themainresultsregardingtoleranceofearlyfailure,itintroducesinterestingheteroge-neoustreatmenteectsbetweenmoreandlessrisk-aversesubjects.Compensationispoten-tiallyquitevariableinthepay-for-performancetreatment,inparticularwhensubjectschoosetoexplorebydivergingfromthedefaultstrategy.Thus,inthepay-for-performancetreatmentweexpectmorerisk-aversesubjectstoexplorelessthanlessrisk-aversesubjects.Ontheotherhand,wepredictthatthereisnosignicantdierenceinexplorationbehaviorbetweenmoreandlessrisk-aversesubjectsinthexedwagetreatmentwherecompensationisconstantbydesign.Wethenturntoinvestigatinghowthethreatofearlyterminationin\ruencesexplorationbehaviorandperformance.12Earlyterminationcanunderminetheexplorationbehaviorin- 12SeealsoManso(2011),SectionVI.9 ducedbytheexplorationcontractbyeliminatingthetoleranceforearlyfailure.Wefurthershowthatthiseectcanbemitigatedbytheuseof\goldenparachutes"orreparationpay-mentswhichsubjectsreceiveincaseofearlyterminationsincethesepaymentsreintroducethetoleranceforearlyfailurethatisrequiredtoencourageexploration.Weintroducetwoadditionaltreatmentgroupsthatenableustoinvestigatetheeectsofterminationandgoldenparachutes.IncentiveScheme4(Termination):\Youwillbepaid50%oftheprotsyoumakeduringthelast10periodsoftheexperiment.However,iftheprotsyoumakeduringtherst10periodsoftheexperimentarebelow800francs,theexperimentwillendafter10periods."IncentiveScheme5(TerminationwithGoldenParachute):\Youwillbepaid50%oftheprotsyoumakeduringthelast10periodsoftheexperiment.Iftheprotsyoumakeduringtherst10periodsoftheexperimentarebelow800francs,theexperimentwillendearlyandyouwillreceiveapaymentof250francs."Therewereatotalof71and78subjectswhoparticipatedintheterminationandthegoldenparachutetreatments.Pureterminationinhibitsexplorationactivitiesbecauseitdoesnotoersucienttoleranceforearlyfailure.13Whilethethreatofterminationproducesstrongincentivesforgoodperformance,italsoforcesindividualstofocusonproducinggoodperformancefromtheverybeginningandthusreducestheincentivesforexploration.Incon-trast,inthegoldenparachutetreatmentweexpectsubjectstoexplorealittlemoreintensivelythanintheterminationtreatmentatthebeginningoftheexperimentdespitethependingthreatofterminationsincethegoldenparachutepaymentprovidesthemwithsomeinsuranceincaseoffailure.TerminationHypothesisSubjectsundertheterminationcontractarelesslikelytondtheoptimalbusinessstrategythansubjectsundertheexplorationtreatment.Furthermore,subjectsunderthegoldenparachutetreatmentaremorelikelytondtheoptimalbusinessstrategythansubjectsintheterminationtreatment.3ResultsWepresenttheresultsobtainedinourexperimentscomparingtheoutcomeacrossthevedierenttreatments(xed-wagecontract,pay-for-performancecontractandexplorationcon-tract,termination,terminationwithgoldenparachute).Subjectswererandomlyassignedto 13Thepredictionthatterminationhasanadverseeectonexplorationdependscruciallyonourchoiceoftheterminationthresholdwhichischosensuchthatitcanbeachievedwithoutexploring.10 thedierenttreatmentsandtherearenosignicantex-postdierencesinage,gender,riskaversionandself-reportedincome.3.1Innovation,ExplorationBehaviorandEortChoiceWerstfocusontheexplorationbehaviorofsubjectsacrosstherstthreedierenttreat-ments.Subjectsdonotmakeanydecisionsthatin\ruenceanyothersubjectsandwehaveatotalof20perioddecisionsforeachofthe379individualsubjects.However,sincedecisionsarecorrelatedovertimeforeachsubject,eachindividualsubjectparticipatingintheexper-imentconstitutesonlyoneindependentobservation.Unlessotherwisenoted,nonparametrictestsarethereforebasedonaveragesoftherelevantvariablesoftheindividualsubjectandthroughoutouranalysisweclusterstandarderrorsattheleveloftheindividualsubject.Ourrstresultshowsthatthepredictionthattheexplorationcontractleadstomoreinnovationthantheothertwocontractsisconrmedbythedata.Result1(innovation):Subjectsunderthexed-wageandpay-for-performancecontractsaresignicantlylesslikelytochoosetosellattheschool(highestprotlocation)inthenalperiodoftheexperimentthansubjectsundertheexplorationcontract.Subjectsundertheexplorationcontractcomeclosest(intermsofmaximumandlast-periodprot)tondingtheoptimalbusinessstrategy.InitialsupportingevidenceforResult1comesfromFigure1whichshowstheproportionofsubjectsunderthexed-wage,pay-for-performance,andexplorationcontractconditionschoosingtoselllemonadeinaparticularlocationinthenalperiod.Consistentwithourexplorationhypothesis,subjectsundertheexplorationcontractsettingaremorelikelytosellattheschoolwhichisthelocationwiththehighestprotsinthenalperiodoftheexperimentthansubjectsunderthexed-wageandpay-for-performanceconditions.Whereasintheexplorationcontractconditionmorethan80%ofsubjectschoosetoselllemonadeattheschool,only40%ofsubjectschoosetodosointhepay-for-performanceconditionand60%choosetodosounderthexed-wagecontract.UsingMann-Whitney-Wilcoxonranksumtestsoftheindividualsubjectaverageswecanshowthatthesedierencesarehighlysignicantbetweentheexplorationcontractandthexed-wagecontract(p-value0.0042)andtheexplorationandthepay-for-performancecontract(p-value0.0001).Thedierenceislessmarkedbetweenthexed-wageandthepay-for-performancecontract(p-value0.0865).14 14Inaddition,weestimatedasubjectxed-eectslogitmodelwherethedependentbinaryvariabletakesthevalue1ifthenallocationchoiceistheschoolwhichistheoptimallocationchoiceintheexperiment,and0otherwise.Theindependentvariablesarebinaryvariablesforthethreedierentcontracts.Asbefore,thecoecientestimatesshowthatsubjectsunderthepay-for-performance(p-value0.0001)andxed-wagecontract(p-value0.0054)aresignicantlylesslikelytochoosetosellintheschoolinthenalperiodofthe11 0 .2 .4 .6 .8Proportion of subjects by location in final period Fixed wageP-f-P contractExploration contract Business districtSchool Figure1:Proportionofsubjectsbylocationinthenalperiodoftheexperimentforthexed-wage,pay-for-performanceandexplorationcontracts.Finally,thedierenceinperformancebetweenthexedwageandthepay-for-performancetreatmentisinlinewiththenegativeeectsofperformancepayfoundinthepsychologyliterature.Wealsoexaminehowclosesubjectscometondingtheoptimalstrategyoverthecourseoftheexperimentintermsoftheprotstheyachieve.Thiscaneasilybemeasuredbyexaminingthemaximumperperiodprotachievedbysubjectsthroughoutthecourseoftheexperiment.Perperiodprotisamorecomprehensivemeasurethanlocationchoice.Itcapturesthemulti-dimensionalaspectofthetaskwhichinvolvesthechoiceofseveralvariables.Onaverage,subjectsundertheexplorationcontractachievethehighestmaximumperperiodprots(145francs)whilesubjectsunderthexed-wage(128francs)andthepay-for-performance(117francs)contractsperformworseonthisdimension.Thesamepatternholdsfornalperiodprotwheretherespectivevaluesare140(exploration),120(xedwage)and111francs(pay-for-performance).Asbeforethedierencesinmaximumperperiodprotaswellasnalperiodprotbetweentheexplorationcontractandtheothertwocontractsarehighlysignicant(Mann-Whitney-Wilcoxontest:p-valuesof0.013and0.0001formaximumprot,p-valuesof0.009and0.0001fornalperiodprot)whilethedierencebetweenthexed-wageandthepay-for-performancecontractisnotstatisticallysignicant(p-value0.1144formaximumprot,p-value0.28fornalperiodprot).Toexplainwhysubjectsundertheexplorationcontractaremorelikelytondtheopti-mallocationandbusinessstrategythansubjectsundertheothertwocontracts,weanalyze experimentthansubjectsintheexplorationcontract.Thenegativeeectonndingtheoptimallocationinwhichtosellisparticularlypronouncedforthepay-for-performancecontractwhilethedierencebetweenxed-wageandpay-for-performancecontractsisnotassignicant(p-value0.0865).12 dierentmeasuresofexplorationandeort.Thenextresultshowsthatsubjectsundertheexplorationcontractexploremorethansubjectsunderthexed-wagecontractwhilesubjectsunderthepay-for-performancecontractexploretheleast.Result2(explorationbehavior):Subjectsunderthepay-for-performancecontractexplorelessthansubjectsunderthexed-wagecontractandtheexplorationcontractwiththelatterexploringthemost.Usingthedierentchoicevariablesavailabletotheagentswecanconstructseveralmea-suresofexplorationactivity.Subjectsinthepay-for-performanceconditionexplorelocationsotherthanthedefaultlocation(businessdistrict)lessoftenthansubjectsundertheothertwocontractswithsubjectsundertheexplorationcontractchoosingtoexplorethemostof-ten.Whilesubjectsundertheexplorationcontractchoosealocationotherthanthedefaultlocationin82%and85%ofcasesintherstandthelast10periods,subjectsunderthexed-wagecontractchoosetodosoonlyin60%and63%ofcasesandtheproportionsareaslowas51%and48%forsubjectsinthepay-for-performancecontract.Thetoleranceforearlyfailureoftheexplorationcontractrelativetothexed-wageandpay-for-performancecontractsencouragedindividualstoattemptnewuntriedapproachesintherst10periods.UsingMann-Whitney-Wilcoxonranksumtestsoftheindividualsubjectaveragesrevealsthatthisdierenceinlocationchoicebehaviorbetweenthedierentcontractsisstatisticallysig-nicant.Intherst10periodssubjectsundertheexplorationcontractchoosetoexploreadierentlocationmoreoftenthansubjectsunderthexed-wagecontract(p-value0.0053)andthepay-for-performancecontract(p-value0.0001).Thedierenceinexplorationbehaviorasmeasuredbylocationchoiceintherst10periodsisnotstatisticallysignicantbetweensubjectsunderthexed-wageandthepay-for-performancecontracts(p-value0.1482),butsubjectsunderthexed-wagecontractchoosetoexploresignicantlymoreoftenthansub-jectsunderthepay-for-performancecontractinthelast10periodsoftheexperiment(p-value0.0985).Thisparticularformofexplorationactivityisalsore\rectedinFigure2whichshowstheaveragesubject-specicstandarddeviationinstrategychoicesforthethreecontinuouschoicevariables(sugarcontent,lemoncontentandprice)duringtherstandlast10periodsoftheexperiment.Thisstandarddeviationmeasurecapturesvariationinallthevariablesofthismulti-dimensionalchoiceproblem.First,thevariabilityofactionchoicessignicantlydeclinesoverthecourseoftheex-perimentinthepay-for-performance(p-value0.0005)andtheexplorationcontracts(p-value0.0001).Thisoccursbecauseinperiods11to20thebeneciallearningeectsofexplorationrelativetoexploitationarenolongeraslargeasatthebeginningoftheexperimentsincethetimehorizonisshorter.Incontrast,thevariabilityofactionchoicesonlydecreasesslightlyin13 .4 .6 .8 1 1.2Standard deviation of strategy choices Fixed wageP-f-P contractExploration contract Periods 1-10Periods 11-20 Figure2:Averagesubject-specicstandarddeviationofstrategychoicesforthethreecontinuousvariables(sugarcontent,lemoncontent,price)inperiods1-10and11-20oftheexperimentforthexed-wage,pay-for-performanceandexplorationcontracts.thexed-wagecontractandthisdeclineisnotstatisticallysignicant(p-value0.2194).Sinceagentsarenotpenalizedforlowprots,explorationbehaviorinthexed-wagecontractisexclusivelydrivenbyintrinsicmotivesandsubjectsmaythereforecontinuetoexploreeventhoughtheadditionalbenetsofexplorationaresmall.Second,thevariabilityofactionchoicesintherst10periodsissignicantlyhigherintheexplorationcontractthaninthepay-for-performance(p-value0.0012)andthexed-wagecontracts(p-value0.0027).Thisshowsthatsubjectsundertheexplorationcontractexperimentandconsciouslymakeverydierentactionchoicesinadirectedattempttondmorepromisingstrategies.Incontrast,inthepay-for-performancecontractthestandarddeviationofactionchoicesismuchlowerassubjectsopttone-tunethedefaultvalues.Thisisalsotrueforsubjectsunderthexed-wagecontractwhoexplorelessthansubjectsundertheexplorationcontractduringtherst10periods.However,becausesubjectsintheothertwotreatmentsexploremuchlessinthelaterperiodsoftheexperimentwhentheircompensationisdirectlylinkedtotheirperformance,thevariabilityofactionchoicesofsubjectsunderthexed-wagecontractishigher(thoughnotalwayssignicantlyso)thaninthepay-for-performance(p-value0.0246)andtheexplorationcontracts(p-value0.6567).Therelativelyhighexplorationbehaviorofsubjectsunderthexed-wagecontractinthelast10periodsoftheexperimentalsoexplainswhytheyaremorelikelytondthehighest-protlocationthansubjectsunderthepay-for-performancecontractwhoexploretheleastovertheentirecourseoftheexperimentamongthethreecontracttreatmentgroups.Wealsoexpectthevariabilityofprotstomirrorthevariabilityofactionchoices.Thisisindeedthecase.First,thevariabilityofprotssignicantlydeclinesovertimewiththe14 declineinvariabilitybeingparticularlymarkedfortheexplorationcontractandthepay-for-performancecontracts.Second,thevariabilityofprotsintherst10periodsissignicantlyhigherforsubjectsundertheexplorationcontractthansubjectsundertheothertwocontracts,whilethereisnosignicantdierenceinprotvariabilityacrosssubjectsunderthethreecontractsinthelast10periods.WeuseCoxhazardratemodelstoanalyzethedynamicsthatgovernthestrategychoicesofindividualsintheexperiment.Thisallowsustotestwhetherthedierenttreatmentcondi-tionsalsoin\ruencewhether,oncetheyhavedecidedtoexplore,subjectscontinuetoexploreandwhatotherfactorscontributetomakingthempersistintheirexplorationactivities.Weclassifysubjectsashavingenteredanexplorativephaseassoonastheychoosealocationotherthanthedefaultlocation(businessdistrict)suggestedbythepreviousmanager.Anexplorativephaseendswhensubjectsmakeonlysmallchangestostrategychoicesrelativetothepreviousperiodorswitchbacktothedefaultlocation.15Ascanbeseenfromcolumn1ofTable1,thehazardrateofendinganexplorativephaseissignicantlyhigherunderthepay-for-performancecontractthanundertheexplorationcontract.Thehazardrateisalsohigherinthexed-wagecontractalthoughthiseectisnotstatisticallysignicant.Moreover,higherprotssignicantlydecreasethehazardrateassubjectsareencouragedtopersistintheirexplorationeort.Column2ofTable1showsthattheestimatesfortherst10periodsarequalitativelysimilar.Finally,answersintheopen-endedpost-experimentalquestionnaireinwhichallsubjectswereaskedtodescribetheirstrategiesandtheeectthecompensationschemehadontheirchoicesalsore\rectedthedescribedexplorationpattern.Subjectsundertheexplorationcon-tractspontaneouslyarguedthatthetoleranceforearlyfailureofthecompensationschemeaswellasthestrongrewardsforsuccessinlaterperiodsin\ruencedtheirstrategicchoices,causingthemtoexperimentwithuntestedlocationsandactionchoicesearlyonandthentochooseandne-tunethebestavailablestrategybeginninginperiod11.Sofar,ourresultshavelargelyfocusedonexplorationbehavior.However,wealsopredictedthatsubjectsunderthexed-wagecontractshouldexertlesseortthansubjectsundertheothertwocontractssincetheircompensationdoesnotdependontheirperformanceintheexperiment.Result3(timeallocationandeortchoice):Subjectsunderthexed-wagecontractspendlesstimemakingandevaluatingdecisionsandexertlesseortrecordingtheirpreviouschoicesandoutcomesintheexperimentthansubjectsunderthepay-for-performanceandexploration 15Inparticular,anexplorativephaseisdenedasendingwhenasubjectswitchesbacktothedefaultlocationorwhenasubjectdoesnotchangelocationandlemonadecolorandalsodoesnotchangelemoncontent,sugarcontentandpricebymorethan0.25units.Asarobustnesscheckwealsousedotherdenitionsthresholdsfortheendofanexplorationphase.Theresultingmagnitudesandsignicancelevelsareverysimilar.15 CoxHazardRateModelsPeriod1-20Period1-10Period1-20Period1-10b/seb/seb/seb/se rhFixedWage0.2170.564***(0.136)(0.204)Pay-for-Performance0.334***0.632***(0.126)(0.201)Termination0.487***0.861***(0.110)(0.185)Parachute0.297**0.580***(0.123)(0.199)tProts-0.001***-0.001***-0.001***-0.001***(0.000)(0.000)(0.000)(0.000) Pseudo-R20.0110.0130.0140.015N2418106829951458 Table1:EstimatesfromaCoxhazardratemodelreportingthehazardratesforexitinganexplorationphasewiththeexplorationcontractasthebaseline.Separateestimationsareshownfortheentire20periodsoftheexperimentandtherst10periods.Robuststandarderrorsarereportedinbrackets.Statisticalsignicanceusingclusteringattheindividualsubjectlevelattheten,veandonepercentlevelisindicatedby*,**and***.contracts.Aprincipaldecidingwhethertopayagentsaxedwagemightworrythatabsentanyintrinsicmotivationandimplicitincentivestheagentwillchoosetominimizecostlyeort.Similarly,inourexperiment|wheresubjectshavetomentallyfocusandrecordpastchoicestotrytomaximizetheirperformance|subjectswhosecompensationdoesnotdependontheirperformancemaychoosetominimizecostlyandtime-consumingcontemplationanddeliberationeort.Indeed,manysubjectsunderthexed-wagecontractclaimedinthepost-experimentalquestionnairethattheyattemptedtominimizethetimeandeortnecessarytocompletetheexperimentsincetheirperformancedidnotaecttheircompensation.Thispatternisalsoborneoutinourdata.Whilesubjectsunderthexed-wagecontractspendonlyanaverageof24secondsonthedecisionscreen(wheresubjectsentertheirstrategychoices),subjectsundertheexplorationandthepay-for-performancecontractsspend31and30secondsrespectively.Thatis,overtheentiredurationoftheexperiment,subjectsundertheexplorationandthepay-for-performancecontractconditionspendalmost30%moretimeonthedecisionscreenthansubjectsunderthexed-wagecondition.Thesedierencesarestatisticallysignicant(p-valuesof0.0014and0.0175)overthecourseoftheentireexperimentaswellasinsubperiods.Moreover,subjectsintheexplorationcontracttreatmentspendsignicantlymoretimeonthedecisionscreenthansubjectsinthexed-wagetreatment(p-value0.022)evenduringtherst10periodsof16 theexperimentwhentheyreceivenocompensation.Thisdierenceintimespentbetweentheexplorationandpay-for-performancecontractsisnotsignicant(p-value0.8477).Thisevidencestandsincontrasttodynamicprincipal-agentmodelsofrepeatedeort(Rogerson1985,HolmstromandMilgrom1987,Sannikov2008),whichpredictthattheex-plorationcontractshouldinducemoreshirkingduringthersttenperiodsoftheexperimentthanthepay-for-performancecontract.Thesemodelsfailtoincorporatethelearningpro-ducedbytheexplorationofnewstrategies,whichpotentiallyenhancesperformanceinlaterperiods,andmaythusprovideincentivesfortheagenttoexerteortinearlyperiods,evenwhenhiscompensationdoesnotdependonproductivityinthoseearlyperiods.Theresultsabovesuggestthatexperimentationandlearningcanindeedbeimportantcomponentsinincentiveproblems,andshouldbetakenintoaccountwhendesigningcompensationschemesforinnovativetasks.Furthermore,inadditiontospendinglesstimemakingdecisions,subjectsunderthexed-wagecontractalsoexertlesseortbyenteringlessinformationintothesheetgiventothemthansubjectsunderthepay-for-performanceandexplorationcontracts.Figure3showsthatacrossthethreecontractsthereisaconsiderablysmallerproportionofsubjectsunderthexed-wagecontractwhollouthalformoreoftheeldsinthedecisiontablethanintheothertwocontracttreatments.Thisdierenceineortchoiceisstatisticallysignicantbetweentheexplorationcontractandthexed-wagecontract(p-value0.0053)aswellasbetweenthepay-for-performanceandthexed-wagecontract(p-value0.0804).Intherst10periodsoftheexperimentsubjectsundertheexplorationcontractaresignicantlymorelikelytorecordinformationthansubjectsinthexed-wagecontract(p-value0.0111)therebyrefutingoncemoretheshirkingpredictionofthestandardrepeatedmoral-hazardmodel.Thedierenceineortexertedduringtherst10periodsbetweensubjectsundertheexplorationcontractandthepay-for-performancecontractisnotsignicant(p-value0.5782).Thedierenceineortchoicebetweentheexplorationcontractandthepay-for-performancecontractispositivebutnotstatisticallysignicant(p-value0.29).Ontheonehand,subjectsunderthepay-for-performancecontractaregivenmorepowerfulincentivesoverallsincetheircompensationdependsonperformancebothintherstandthelast10periodsoftheex-periment.Ontheotherhand,sincesubjectsundertheexplorationcontractchoosetoex-perimentwithverydierentstrategiesintherst10periodsasweshowedinResult3,theyneedtoexertmoreeortwhenevaluatingtheirdecisionsthansubjectsunderthepay-for-performancecontract.ThisisalsovisibleinFigure3whichshowsthateortdeclinesinthepay-for-performancecontract.Thisoccurssincesubjectsinthepay-for-performancecontractessentiallystopexploringandexperimentingwithdierentchoicesveryearlyintheexperi-mentandthereforetheybarelychangetheirchoicesinthelast10periods.Sincethereislittle17 .5 .6 .7 .8Proportion of subjects exerting high effort Fixed wageP-f-P contractExploration contract Periods 1-10Periods 11-20 Figure3:Proportionofsubjectswhocompletemorethanhalfoftheeldsinthedecisionrecordtableforthexed-wage,pay-for-performanceandexplorationcontracts.change,theydonothavetorecordtheirchoicesascarefullyassubjectsintheexplorationcontracttreatment.Wealsonotethattimeallocationandeortchoiceinthexed-wageisstrictlygreaterthanzerosincesomeofthesubjectsaresucientlymotivatedbyintrinsicrewardstoexerteort.Aninspectionofeortchoicesbysubjectsinthexed-wagetreatmentrevealsabimodaldistribution.Subjectseitherfullyrecordordonotrecordanyoftheirpastchoices.Moreover,subjectsinthexed-wagetreatmentwhoexertmoreeortaremorelikelytosuccessfullyinnovate:65%ofthemendupsellingattheschoolinthenalperiodcomparedto47%ofthesubjectswhoexertlesseort,butthisdierenceisnotstatisticallysignicant(p-value0.2047).However,maximumprotsaresignicantlyhigherforsubjectsinthexed-wagetreatmentwhoexertmoreeort(p-value0.0298).163.1.1AveragePerformanceHavingconrmedthattheinnovationsuccess,explorationbehavior,timeallocationandeortchoiceacrossthedierentcontractsareinaccordancewithourtheoreticalpredictions,wenowshowthatsubjects'overallperformanceintheexperimentasmeasuredbyaverageprotishighestintheexplorationcontract.Result4(performance):Subjectsundertheexplorationcontractproducehigheraverageprof-itsthansubjectsunderthepay-for-performanceandxed-wagecontracts. 16Forastudyoftheeectofintrinsicmotivationoninnovationproductivity,seeSauermannandCohen(2010).18 PreliminaryevidenceforResult4comesfrominspectingtheaverageprotforthethreecontracts.Thisperformancemeasureishighestintheexplorationcontract(111francs)andthedierenceinperformancebetweentheexplorationcontractandthepay-for-performance(96francs)andthexed-wagecontract(102francs)isstatisticallysignicant(Mann-Whitney-Wilcoxontest:p-valuesof0.0009and0.0253).Thisdierenceinperformanceexistsdespitethefactthattheaveragewagereceivedbysubjectsundertheexplorationcontractislowerthanintheothertwocontracts.17Wenowinvestigatewhetherattitudestowardriskcanexplainthedierencesinperfor-mancedocumentedinResult4.Result5(riskaversion):Underthepay-for-performancecontractmorerisk-aversesubjectsaresignicantlylesslikelytoexploreandtochoosetosellintheoptimallocationinthenalperiodoftheexperiment.Theyalsoproducesignicantlylowerprots.Attitudestoriskhaveasimilar(thoughstatisticallyinsignicant)eectintheexplorationcontract,whilenosystematiceectsofriskarefoundforthexed-wagecontract.Usingthedatafromtheseparateriskaversionexperimentweclassifysubjectsintomoreandlessrisk-aversegroups.TheleftpanelofFigure4providesarstindicationforthesignandmagnitudeoftheeectofriskaversiononthelikelihoodofndingthebeststrategy.InthisgureweuseourriskaversionmeasurestofurtheranalyzethenalperiodlocationchoiceaswedidinFigure1.Weseparatelypresentnallocationchoicesformoreandlessrisk-aversesubjectsforeachofthethreecontracts.Inthepay-for-performancecontract,morerisk-aversesubjectsarelesslikelytondtheoptimallocationastheyarelesslikelytoexplorethanthelessrisk-aversesubjects.Thisinnovation-reducingeectofriskisstatisticallysignicantinthepay-for-performancecontracttreatment(p-value0.0170),butitisnotstatisticallysignicantintheothertwotreatments.18Thislowerrateofinnovationsuccesscausedbyriskaversionisdrivenbythelowerlevelsofexplorationunderthepay-for-performancecontractbecauseinthistreatmenttheproportionoflocationchoicesotherthanthedefaultlocation(p-value0.0075)aswellasthevariabilityofactionchoices(p-value0.0181)aresignicantlylowerforsubjectswithhigherriskaversion.However,intheexplorationcontractwheresubjects'failureistoleratedinearlyperiodsoftheexperimentandcompensationhasamuchsmallerriskycomponenttheeectissmallerinmagnitudeandnotstatisticallysignicant.Thesame 17Duringtherst10periodsoftheexperimentthethreecontractsarequitesimilarintermsofaverageprots.Afterperiod10,theaverageprotsunderthedierentcontractsbegintodivergeassubjectsundertheexplorationcontractreverttoandsubsequentlyne-tunethebeststrategytheyfoundduringtherst10periodsoftheexperiment.18However,evenamonglessrisk-aversesubjectsthereexistsastatisticallysignicantinnovation-reducingeectofstandardpay-for-performancerelativetotheexplorationcontract(p-value0.0367).19 0 .2 .4 .6 Fixed wageP-f-P contractExploration contract less R-Amore R-Aless R-Amore R-Aless R-Amore R-A Business districtSchool 90 100 110 120 130 140 Fixed wageP-f-P contractExploration contract less R-Amore R-Aless R-Amore R-Aless R-Amore R-A Maximum profitLast period profit Figure4:Proportionofsubjectsbylocationinthenalperiodoftheexperiment(left)andmaximumprot,lastperiodprotandaverageperperiodprotofsubjects(right)forthexed-wage,pay-for-performanceandexplorationcontractsadjustingfordierencesinriskaversion.istrueinthexed-wagecontractwherecompensationentailsnorisk.19Itisimportanttonoteatthispointthat,aswithallsubgroupanalysis,theheterogeneoustreatmenteectswithrespecttoriskaversionneedtobetreatedwithcaution.Itispossiblethatthesemaymerelybedrivenbycorrelatedomittedvariables(e.g.,cognitivere\rectionability(Frederick2005))aswecanonlyexogenouslyvarytheincentiveschemesadministeredtotheexperimentalsubjects.Sincemorerisk-aversesubjectsunderthepay-for-performancecontractarelesslikelytoexploreandthereforelesslikelytoselllemonadeintheoptimallocationinthenalperiod,theyalsoproducelowerprotsascanbeseenintherightpanelofFigure4.Thisprot-reducingeectofriskaversioninthepay-for-performancecontractislargeinmagnitudeandstatisticallysignicantformaximumprot(p-value0.0563)andnalperiodprot(p-value0.0382),butitisnotstatisticallysignicantforaverageprot(p-value0.1846).Furthermore,asinthecaseofthenalperiodlocationchoice,riskaversionalsohasasmallnegativebutstatisticallyinsignicanteectonprotmeasuresintheexplorationandthexed-wagecontracttreatment.20Therecouldbereasonsinadditiontoriskaversionforthedierenceinaverageprotsacrossthethreetreatmentgroups.Forexample,inourexperimentsubjectsarenotgiven 19Qualitativelysimilarresultsholdfortheambiguityaversionmeasure.Theeectsareofthesamesignastheeectsofriskaversion,buttheyaregenerallysmallerinmagnitudeandinsomecasesnotstaticallysignicant.20Wealsospeculatethatlossaversion,similartoriskaversion,mayfurtherexacerbatetheexploration-reducingeectsofpay-for-performancecompensation.Underthepay-for-performancecontractanyloss-averseagentmaysetthepayoofthedefaultstrategyasareferencepointandwillnotdeviatefromthisstrategytoavoidanylossesrelativetothereferencepoint.20 preciseinformationabouttheprotsassociatedwitheachoftheavailablechoices.21Thedierencesinaverageprotsacrossthethreetreatmentgroupscouldthusbeduetosubjectsbeingpessimisticaboutthereturnstoexploration.Theexplanationmirrorsthatgivenintheabovetwoparagraphswithpessimisminplaceofriskaversion.3.2TerminationWenowshowthatthethreatofterminationreducestheprobabilitythatsubjectssuccessfullyinnovatebecausethethreatofearlyterminationreducesexplorationactivities.Furthermore,theadverseeectsofterminationarelesspronouncedinthegoldenparachutetreatment.Result6(termination):Thethreatofterminationhasadverseeectsoninnovationsuccessandexplorationactivities,butgoldenparachutesalleviatethesenegativeeects.Riskaversionfurtherreducesinnovationsuccess,explorationactivitiesandperformanceintheterminationtreatment.Figure5showsnalperiodlocationchoicesintheexplorationcontract,terminationandgoldenparachutetreatmentswhereinthecaseofthelattertwotreatmentsweeliminatedsubjectsthatareterminatedaftertherst10periods.Thethreatofterminationinthepureterminationandgoldenparachutetreatmentsignicantlyreducestheprobabilitythatsubjectsendupchoosingtosellatthebestlocationinthenalperiodoftheexperimentrelativetotheexplorationcontracttreatment(p-values0.0001and0.0200)whiletheuseofgoldenparachutesraisestheinnovationsuccessprobability(p-value0.0485)relativetotheterminationtreatment.22Focusingonsubjectsthatarenotsubjecttoterminationweagainndthatterminationhasaninnovation-reducingeectsinceaveragemaximumprotintheexplorationcontracttreatment(145francs)issignicantlyhigherthaninthetermination(126francs)andthegoldenparachutetreatments(134francs).Therespectivep-valuesforthedierencesare0.0037and0.0772.Comparingthemaximumprotsfortheterminationandgoldenparachutetreatmentsshowsthattheuseofgoldenparachutesslightlymitigatestheseadverseeects,thoughtheeectisnotsignicant(p-value0.1784).Theadverseeectofterminationismorepronouncedifweconsiderthefullsampleofsub-jectsandonlyfocusontherst10periods.Theaveragemaximumprotinthetermination 21Thisisalsothecaseinsomeoftheaforementionedpsychologyexperimentswhichndthatsubjectsunderaxed-wagecontractperformbetterthansubjectsunderapay-for-performancecontract.22Thesameresultsholdwhenfocusingexclusivelyonthenallocationchoiceaftertherst10periodsusingallthesubjectsintheterminationandgoldenparachutetreatments.Thethreatofterminationreducestheprobabilityofndingthebestlocationrelativetotheexplorationtreatment(p-values0.0063and0.0562)andtheuseofreparationpaymentsincreases(thoughnotstatisticallysignicantly)theinnovationsuccessprobabilityinthegoldenparachutetreatmentrelativetotheterminationtreatment(p-value0.3176).21 0 .2 .4 .6 .8Proportion of subjects by location in final period Exploration contractTerminationGolden parachute Business districtSchool Figure5:Proportionofsubjectsbylocationinthenalperiodoftheexperimentfortheexplorationcontract,terminationandgoldenparachutetreatments.andthegoldenparachutetreatmentsisagainsignicantlylowerthanintheexplorationcon-tracttreatment(p-values0.0032and0.0037).However,thedierencebetweentheterminationandthegoldenparachutetreatmentsisnotstatisticallysignicant(p-value0.7989).Asinouranalysisofthethreebaselinetreatments,wecantracethedierencesinin-novationsuccessbacktodierencesinexplorationbehavior.Tothisendweagaincomparethenumberoftimessubjectschoosetodeviatefromtheproposedstrategyandtoexplorealocationotherthanthebusinessdistrict.Toguardagainstpotentialselectioneectsarisingfromattritionwefocusexclusivelyonchoicesintherst10periods.Asexpected,explorationislowerintheterminationtreatmentwheresubjectsshyawayfromexploringotherlocationsintherst10periods.Whiletheaverageproportionoflocationchoicesotherthanthedefaultlocationis82%intheexplorationcontractitisonly47%intheterminationtreatmentand59%inthegoldenparachutetreatment.Thisexploration-reducingeectofthethreatofter-minationisstatisticallysignicant(p-values0.0001and0.0009).Moreover,goldenparachutesincreaseexplorationactivitiesrelativetothepureterminationtreatment(p-value0.0495).Inthepost-experimentquestionnairesubjectsarguedthatthethreatofterminationforcedthemtoconcentrateonsellinginthebusinessdistrictandleftnoleewayforexploration.Furtherevidencefortheexploration-reducingeectofthethreatofterminationandtheexploration-increasingeectofreparationpaymentscomesfromcomparingthevariabilityofactionchoicesintherst10periodsforthefullsampleofsubjects.Thesubject-specicstandarddeviationofactionchoicesintherst10periodsishighestintheexplorationcontract(standarddeviation1.09).Thismeasureissignicantlylowerintheterminationtreatment(standarddeviation0.74,p-value0.0014)andinthegoldenparachutetreatment(standarddeviation0.79,p-value0.0071).Theuseofgoldenparachutesslightlyincreasesexploration22 0 .2 .4 .6 Exploration contractTerminationGolden parachute less R-Amore R-Aless R-Amore R-Aless R-Amore R-A Business districtSchool 90 100 110 120 130 140 Exploration contractTerminationGolden parachute less R-Amore R-Aless R-Amore R-Aless R-Amore R-A Maximum profitLast period profit Figure6:Proportionofsubjectsbylocationinthenalperiodoftheexperiment(left)andmaximumprot,lastperiodprotandaverageperperiodprotofsubjects(right)fortheexplorationcontract,terminationandgoldenparachutetreatmentadjustingfordierencesinriskaversion.activityrelativetotheterminationtreatment,butthiseectisnotstatisticallysignicant(p-value0.2821).23Usingthesamehazardratemodelasinouranalysisofthebaselinetreatmentsthoughconcentratingexclusivelyontherst10periodswecaninvestigatehowlikelysubjectsaretopersistintheirexplorationactivitiesinthedierenttreatments.Column3ofTable1showsthatbothintheterminationtreatmentandinthegoldenparachutetreatmentsubjectsaresignicantlymorelikelytostopexploringthanintheexplorationcontract.Moreover,subjectsintheterminationtreatmentarealsosignicantlymorelikelytostopexploringthansubjectsunderthegoldenparachutetreatment(p-value0.0663).Column4ofTable1reportsestimatesfortherst10periodsshowingstatisticallysignicantdierencesinthehazardratebetweentheexplorationcontractandtheterminationaswellasthegoldenparachutetreatment.Thedierencebetweenterminationandgoldenparachuteisalsostatisticallysignicant(p-value0.0604).Riskaversionplaysanimportantroleintheterminationtreatmentascanbeseenintheleftpanel,whichshowsnalperiodlocationchoice,andintherightpanelofFigure6,whichpresentsthedierentprotmeasures.Morerisk-aversesubjectsintheterminationtreatmentarelesslikelytosellintheschoolinthenalperiodoftheexperimentandtheyachievelowermaximum,nalperiodandaverageprots.Throughout,thereisastatisticallysignicantnegativeeectofriskaversionintheterminationtreatmentonthecorrectnalperiodlocation 23Thedierentproportionsofsubjectswhoareterminatedintheterminationandthegoldenparachutetreatmentsarealsoinlinewithsubjectsexploringmoreinthelattercase.Whileintheterminationtreatment13outof71subjects(18%)donotmeetorexceedtheterminationthreshold,21outof78subjects(27%)areterminatedinthegoldenparachutetreatment,butthedierenceisnotstatisticallysignicant(p-value0.2124).23 choice(Mann-Whitney-Wilcoxontest:p-value0.0041)aswellasmaximumprots(p-value0.0023),nalperiodprots(p-value0.0041)andaverageprots(p-value0.0037).Thisndingisinlinewithourpreviousanalysiswherewefoundsimilarlystrongeectsofriskaversionforthepay-for-performancecontractwhichalsoinducesindividualstoachieveprotsfromtheverybeginningoftheexperimentinsteadoflearningthroughexploration.Incontrast,likeforourndingfortheexplorationcontracttreatment,thereisnostatisticallysignicanteectofriskaversioninthegoldenparachutetreatment.Finally,intheterminationtreatmentahighdegreeofriskaversionsignicantlydecreasessubjects'propensitytoexplore.Intheterminationtreatmentthenumberoftimessubjectschoosetodeviatefromtheproposedstrategyandtoexplorealocationotherthanthebusinessdistrictintherst10periodsissignicantlylowerforsubjectswhoaremorerisk-averse(p-value0.0114).Similarly,intheterminationtreatmentthevariabilityofactionchoicesintherst10periodsisalsosignicantlylowerformorerisk-aversesubjects(p-value0.0040).Therearealsosmallnegativeeectsofriskaversiononexplorationactivityinthegoldenparachutetreatment,buttheseeectsareneverstatisticallysignicant.4RobustnessInthissubsection,weshowthatourresultsarerobusttomodicationsintheexperimentaldesign.4.1SignalingEectsofIncentiveContractsIntheanalysiswepreviouslyconducted,eachsubjectonlyeversawoneparticularincentivecontract.Thesubjectswerenotmadeawarethatavarietyofdierentincentiveschemeswereadministeredtodierentsubjects.Thismeansthatsubjectsmightmakedierentinferencesfromthedierentcontractstheyaregivenaboutwhatthebeststrategytoplayis.Forexample,whilesubjectsunderthepay-for-performancecontractmightinferthatthebeststrategyisnottoexplore,subjectsundertheexplorationcontractmightinferthatthebeststrategyistoexplore.Toaccountforthesepotentialsignalingeectsweadministeredanothertreatmentinwhichsubjectswereabletoseethatbothpay-for-performanceandexplorationcontractswereavailable.Inthistreatment,afterhavingobservedthesetofpossiblecontracts(pay-for-performanceorexploration),theincentiveschemethatwouldbeappliedtoeachsubjectwasdeterminedbyhavingthesubjectrolladice.Afterhavingobservedtheoutcomeofthedicerolltheexperimentercircledtherelevantcompensationschemeandcrossedouttheirrelevantcompensationscheme.Atotalof70subjectsparticipatedinthistreatmentofwhich24 0 .2 .4 .6 .8Proportion of subjects by location in final period P-f-P contract (dice)Exploration contract (dice) Business districtSchool Figure7:Proportionofsubjectsbylocationinthenalperiodoftheexperimentforthepay-for-performance(diceroll)andexploration(diceroll)contracts.32subjectsrolledthediceandreceivedapay-for-performancecontractand38subjectsrolledandreceivedanexplorationcontract.Figure7conrmsourresultsabouttheimportanceofcorrectlystructuredincentivesformotivatinginnovation.Asbefore,subjectswhoaregivenanexplorationcontractaresigni-cantlymorelikely(Mann-Whitney-Wilcoxontest:p-value0.0152)tochoosethebestlocationinthenalperiodoftheexperimentthansubjectswhoreceiveapay-for-performancecon-tract.Subjectswithanexplorationcontractalsoagainachievesignicantlyhighermaximumprots(138francs)andhighernalperiodprots(134francs)thansubjectsunderapay-for-performancecontract(120francs,118francs).Therespectivep-valuesforthecomparisonsare0.0372and0.0654.Asbeforethisdierenceininnovationsuccessisdrivenbythedierencesinexplorationbehaviorthatincentiveschemesinduce.Theproportionoflocationchoicesotherthanthede-faultlocationissignicantlyhigherforsubjectswhoobtainanexplorationcontractfollowingtheirdiceroll(p-value0.0045)andthevariabilityofstrategychoicesisalsohigher,althoughthisdierenceisnotsignicant(p-value0.1343)duetothesmallersamplesize.Mirroringourpreviousresults,subjectsunderthepay-for-performancecontractalsohavelowaverageprotsalthoughthiseectisnotstatisticallysignicant(p-value0.1591).Fur-thermore,riskaversionagainhasaninnovation-andprot-reducingeectinthepay-for-performancetreatment.Inthepay-for-performancetreatmentthereisastatisticallysigni-cantnegativeeectofriskaversiononthecorrectnalperiodlocationchoice(p-value0.0583),butthereisnosignicanteectintheexplorationcontracttreatment.Thenegativeeectofriskaversionwhensubjectsobtainapay-for-performancecontractisalsoapparentinthelowerprotsformorerisk-aversesubjects,butthiseectisnotstatisticallysignicantdueto25 thesmallsamplesize.4.2GoldProspectingExperimentInthelemonadestandexperiment,subjectsexertedrealeortandhadtoformbeliefsabouttheoutcomesassociatedwitheachoftheavailableactions.Asanotherrobustnesscheckweconductedasecondexperimentusingthechoseneortapproachandawithin-subjectsdesign.Thisallowsustofullycontrolthebeliefsofsubjectsinaparametrizedsetting.4.2.1ProceduresandSubjectPoolTheexperimentwasprogrammedandconductedwiththesoftwarez-Tree(Fischbacher2007)attheCaliforniaSocialScienceExperimentalLaboratory(CASSEL)atUCLA.ParticipantswererecruitedfromtheCASSELsubjectpoolusinganonlinerecruitmentsystem.Atotalof30subjectsparticipatedintheseexperiments.Weemployedawithin-subjectsdesignsothatallsubjectsexperiencedallthreetreatmentconditions.Theexperimentalsessionlasted50minutes.Duringtheexperiment,experimentalcurrencyunitscalled\goldnuggets"wereusedtokeeptrackofmonetaryearnings.Theexchangeratewassetat3goldnuggets(gn)=$1,andtheshow-upfeewas$10.Onaverage,subjectsearned$33.4.2.2TheTaskSubjectsplayseveralseparateinstancesofagoldprospectinggamethatexactlymirrorsthetheoreticalmodelinManso(2011).Eachgoldprospectinggameconsistsof2periods.Ineachperiod,eachsubjectchoosestoprospectgoldinoneoftwomountainsortostayathome.Eachoftheseoptionshasanassociatedrevenueandcost.Theoption\Stayathome"alwaysyields0gnanditcoststhesubject0gntochoosethisoption.Incontrast,theoption\Mountain1"yields100gnwithaprobabilityof50%and0gnwithaprobabilityof50%anditcosts0.25gntochoosethisoption.Eachpayorealizationof\Mountain1"isanindependentdraw.Finally,theoption\Mountain2"allowsforlearning.Withprobabilityp,\Mountain2"isagold-richmountain,inwhichcaseityields100gninbothperiod1andperiod2wheneverthisoptionischosen.Withaprobabilityof1 p,\Mountain2"isagold-poormountain,inwhichcaseitalwaysyields0gnwheneverthisoptionischosen.Thecostofchoosing\Mountain2"is0.75gn.Eachinstanceofthegoldprospectinggameisanindependentdraw.Thus,eachsubjectcanonlylearnaboutthequalityof\Mountain2"forthatparticularinstanceofthegoldprospectinggame.26 Theparticipantsintheexperimentthusfacethechoicebetweenanoutsideoptionof0gn(\Stayathome"),anoptionwithaknownandconstantpayodistribution(\Mountain1")andanoptionwithanunknownpayodistribution(\Mountain2")thatallowsforlearning.Weelicitsubjectchoicesusingthestrategymethod:subjectsareaskedtoprovideafullcontingentplanofactionforallpossibleoutcomes.Thus,ontheinputscreenforeachgoldprospectinggamesubjectsenter(a)theirchoiceforperiod1,(b)theirchoiceforperiod2conditionalonapayoof0gninperiod1,and(c)theirchoiceforperiod2conditionalonapayoof100gninperiod1.Attheendofeachgoldprospectinggame,subjectslearnhowmuchrevenuetheyobtainedandareremindedofthecosttheyincurred.Inaddition,attheendofeachgoldprospectinggamesubjectsareinformedabouttheircumulativepayobalance.4.2.3TreatmentGroupsandPredictionsWeconductfoursetsoftreatmentswhichdierintermsoftheprobabilitypthatMountain2isagold-richmountain.Theprobabilityptakesthevalues5%,35%,45%oritisunknowntothesubjects.Whenpisunknownwechooseitsuchthatpisequalto50%.Ineachofthefourtreatmentsets,eachsubjectplaysthegoldprospectinggameunderthreedierentcompensationschemes.24Thus,thereareatotaloftwelveseparategoldprospectinggamesthateachsubjectplays.Inparticular,thethreecompensationschemesare:IncentiveScheme1(FixedWage):\Youwillbepaidaxedwageof1gnperperiod."IncentiveScheme2(Pay-for-Performance):\Youwillbepaid10%ofthegoldrevenueduringthe1standthe2ndperiodoftheexperiment."IncentiveScheme3(Exploration):\Youwillbepaid10%ofthegoldrevenueinthe2ndperiodoftheexperiment."Theseincentivesystemsexactlymirrorthepaymentschemesinthelemonadestandex-perimentandgeneratethesamequalitativepredictions.Table2showstheoptimalchoicesforarisk-neutralagentineachtreatmentcondition.Alongthecolumnstherearedierentvaluesofpwhilethedierentincentiveschemesareshowninthedierentrows.AsisapparentfromTable2,aspincreasesthechoiceof\Mountain2"becomesmoreattractive.ThismakesittheoptimalchoiceofactionintherstperiodforhighvaluesofpinthePay-for-PerformanceandExplorationcontracttreatments.Ourmainhypothesis, 24Toaccountforpotentialordereects,foreachsubjectanyofthefourpconditionswasequallylikelyandindependentlychosenastherst,second,thirdorfourthtreatment.Furthermore,withineachptreatment,foreachsubjecteachcompensationschemewasequallylikelyandindependentlychosenastherst,secondorthirdcompensationscheme.27 OptimalStrategiesp=005p=035p=045punknown FixedWagePeriod1StayatHomeStayatHomeStayatHomen/aPeriod2after100gnStayatHomeStayatHomeStayatHomen/aPeriod2after0gnStayatHomeStayatHomeStayatHomen/a Pay-for-PerformancePeriod1Mountain1Mountain1Mountain2n/aPeriod2after100gnMountain1Mountain1Mountain2n/aPeriod2after0gnMountain1Mountain1Mountain1n/a ExplorationPeriod1StayatHomeMountain2Mountain2n/aPeriod2after100gnMountain1Mountain2Mountain2n/aPeriod2after0gnMountain1Mountain1Mountain1n/a Table2:Optimalstrategiesforarisk-neutralagentineachtreatmentcondition.however,concernstheextenttowhichthedierentpaymentschemesencouragethechoiceof\Mountain2"inperiod1ofthegoldprospectinggameforagivenlevelofp.Ingeneral,duetothetoleranceforearlyfailureandrewardforlong-termsuccess,subjectsundertheexplorationcontractshouldbemorelikelytochoose\Mountain2"inperiod1thansubjectsunderthexedwageandthepay-for-performancecontracts.Thecomparisonismostclearlyvisiblewhenp=035:inperiod1subjectsshouldchoose\StayatHome"underthexedwagecontract,\Mountain1"underthepay-for-performancecontract,and\Mountain2"undertheexplorationcontract.4.2.4ResultsWerstcomparestrategychoicesacrossthethreeincentivetreatmentsforthe4dierentlevelsofthesuccessprobabilityp.Eachofthe30subjectsmadechoicesunderallthe12dierenttreatmentswhichallowsustomakewithin-subjectcomparisons.Wefocusontheexplorationbehaviorofsubjectsacrossthedierenttreatmentconditions.Table3showstherst-periodchoicesmadebysubjectsinthedierentconditionsaswellastheassociatedtheoreticalprediction.Consistentwiththepredictionthattheexplorationcontractencouragesexploration,moresubjectschoose\Mountain2"underthisincentiveschemethanundertheothertwocontracts.Moreover,subjectsunderthexedwagecontractoverwhelminglychoosetheshirkingoption\StayatHome",whilesubjectsunderthepay-for-performancecontractchoosetoexploit\Mountain1."25InTables2and3weshowthatforp=035,thepredictedchoiceofactionforarisk-neutralagentintherstperiodisdierentbetweenthepay-for-performanceandtheexploration 25Asaresultofthisincreasedexplorationactivity,theexpectedprobabilityofnding100gninperiod2giventhesubjects'strategiesis60%undertheexplorationcontract,yetonly52.9%and8.2%underthepay-for-performanceandxed-wagecontracts.28 ChoicesinPeriod1p=005p=035p=045punknown FixedWageStayatHome23252426Mountain17233Mountain20331TheoreticalPredictionStayatHomeStayatHomeStayatHomen/a Pay-for-PerformanceStayatHome0111Mountain126201118Mountain2491811TheoreticalPredictionMountain1Mountain1Mountain2n/a ExplorationStayatHome17124Mountain111556Mountain22242320TheoreticalPredictionStayatHomeMountain2Mountain2n/a Table3:ChoicesinPeriod1foreachtreatmentcondition.contractwhileforp=045theoptimalchoicesforarisk-neutralagentunderthetwoincentivesystemscoincide.Finally,forp=005arisk-neutralagentundertheexplorationcontractisexpectedtochoose\StayatHome"sincethepayofromtheexplorationoption\Mountain2"istoolowandsubjectsdonotearnenoughintherstperiodtojustifythecostsofchoosing\Mountain2".Whenp=035only9subjectschose\Mountain2"inperiod1underthepay-for-performancecontract,but24subjectschosetodosoundertheexplorationcontract.Duetoourwithin-subjectdesignweobservethechoicesofsubjectsunderallthedierenttreat-mentconditions.AMcNemarmatchedsamplestestshowsthatthedierenceinthechoiceof\Mountain2"forp=035underthepay-for-performanceandtheexplorationcontractishighlystatisticallysignicant(p-value0.0007).Thestatisticalsignicanceofthisdierenceisalsoconrmedbyasubjectxed-eectslogitregressionwherewecomparetherst-periodchoicesforp=035underthepay-for-performanceandtheexplorationcontract.Thebinarydependentvariabletakesthevalue1ifasubjectchose\Mountain2"inperiod1andvalue0otherwiseandthedependentvariabletakesthevalue1ifthetreatmentwastheexplorationcontractand0ifitwaspay-for-performancecontract.Theassociatedp-valueofthecoecientofthedependentvariableis0.004indicatingagainthatthedierenceishighlystatisticallysignicant.AsisobviousfromtherawdatashowninTable3,thedierenceinbehaviorbetweentheexplorationandthexedwagecontractisalsostatisticallysignicant.Underthexedwagecontractsubjectsoverwhelminglychoosetheshirkingoption\StayatHome".Therelevantp-valuesfortheMcNemartestandthecoecientofthesubjectxed-eectslogitregressionare0.00001and0.001.Althoughwearenotabletoprovideprecisepredictionsforsubjectbehaviorinthetreat-29 mentconditionswherepisunknown,thissetoftreatmentconditionsisofparticularinterestasitsharesthefeatureoftheunknownsuccessprobabilitywiththemorerealisticlemon-adeexperiment.Table3showsthatsubjectbehaviorintherstperiodforunknownpissimilartowhenp=035andformalstatisticaltestsconrmthisimpression.Subjectsaresignicantlymorelikelytochoose\Mountain2"intherstperiodundertheexplorationcontractthanunderthepay-for-performanceincentivescheme(p-value0.032,xed-eectslogit).Thus,evenwhenthesuccessprobabilitypisunknowntheearlyfailuretoleranceoftheexplorationcontractrelativetothepay-for-performancecontractmotivatessubjectstochoosetheexplorationoption\Mountain2".Giventheobservedbehaviorofthesubjectstheexperimentalsoallowsustocalculatehowlargetheexpectedprotofaprincipalwouldbewhoimplementsdierentincentiveplans.Forp=005theprotsoftheprincipalarehighestwhenheoersapay-for-performancecontract(75.53)ratherthananexplorationcontract(59.47).Forhigherexpectedsuccessprobabilitiesoftheexploratoryoption,i.e.p=035,p=045orpunknowntothesubjects,theprincipalalwaysreapsahigherprotwhenoeringanexploration(91.23,103.13,103.33)ratherthanapay-for-performancecontract(87.23,94.2,96).Finally,whenp=035andexplorationisdesirable,theprincipalwouldneedtooersubjects27%ofthetotalgoldrevenueinsteadofjust10%toinducerisk-neutralsubjectstoexploreunderastandardpay-for-performancecontract.5ConclusionInthispaper,wearguedthatappropriatelydesignedmanagerialcompensationiseectiveinenablingentrepreneurshipandmotivatinginnovation.Inareal-eorttaskthatinvolvesinnovationthroughexperimentation,wendthatsubjectsunderapaymentschemethattoleratesearlyfailureandrewardslong-termsuccessexploremoreandarethusmorelikelytodiscoversuperiorstrategiesthansubjectsunderxed-wageorstandardpay-for-performanceincentiveschemes.Wealsondthatthethreatofterminationcanundermineinnovation,andthatthiseectismitigatedbythepresenceofacontractthatmimicsagoldenparachute.Byusingacontrolledlaboratoryexperimentinwhichindividualsarerandomlyassignedtodierentcontracts,weareabletoestablishacausalrelationbetweenparticularincentiveschemesandinnovationperformance.Ourresultscomplementothercontributionstotheliteratureonentrepreneurshipandinnovation,whichusenaturallyoccurringdatatoshowthattoleranceforfailureisassociatedwithinnovationactivity.Providedthatourresultsareexternallyvalid,themostdirectout-of-sampleimplicationofourresultsisthatinsituationsinwhichenablingentrepreneurshipandmotivatinginnovation30 areimportantconcerns,suchasthedesignofcompensationplansoftopexecutivesandmiddlemanagersinlargecorporationsaswellasofentrepreneursinstart-upcompanies,toleranceforearlyfailurecoupledwithrewardsforlong-termsuccessareeectiveinmotivatinginnovation.Inthecontextofexecutivecompensation,forexample,implementingtoleranceforfailureinvolvestheuseofpracticesthatareoftencriticizedsuchasoptionrepricing,managerialentrenchment,andgoldenparachutes.Ourresultssuggestthatrestrictingthesepracticescanhaveadverseeectsoninnovation.Theframeworkandmethodsproposedinthispapercanalsobeusefulinstudyingotherimportantproblemsintheentrepreneurshipliterature.Forexample,howdoesthechoiceofnancingaecttheentrepreneur'sattitudetowardsinnovation?Howcanwedesigncom-pensationpackagestoattractcreativeentrepreneurswhilekeepingshirkersandconventionalentrepreneursaway?Weleavethesequestionsforfutureresearch.31 AppendicesALemonadeStandExperimentA.1ExperimentalInstructionsInstructionsYouarenowtakingpartinaneconomicexperiment.Pleasereadthefollowinginstructionscarefully.Everythingthatyouneedtoknowinordertoparticipateinthisexperimentisexplainedbelow.Shouldyouhaveanydicultiesinunderstandingtheseinstructionspleasenotifyus.Wewillansweryourquestionsatyourcubicle.Duringthecourseoftheexperimentyoucanearnmoney.Theamountthatyouearnduringtheexperimentdependsonyourdecisions.Allthegainsthatyoumakeduringthecourseoftheexperimentwillbeexchangedintocashattheendoftheexperiment.Theexchangeratewillbe:100francs=$1Theexperimentisdividedinto20periods.Ineachperiodyouhavetomakedecisions,whichyouwillenteronacomputerscreen.Thedecisionsyoumakeandtheamountofmoneyyouearnwillnotbemadeknowntotheotherparticipants-onlyyouwillknowthem.Pleasenotethatcommunicationbetweenparticipantsisstrictlyprohibitedduringtheexperiment.Inadditionwewouldliketopointoutthatyoumayonlyusethecomputerfunctionswhicharerequiredfortheexperiment.Communicationbetweenparticipantsandunnecessaryinterferencewithcomputerswillleadtotheexclusionfromtheexperiment.Incaseyouhaveanyquestionsdon'thesitatetoaskus.ExperimentalProceduresInthisexperiment,youwilltakeontheroleofanindividualrunningalemonadestand.Therewillbe20periodsinwhichyouwillhavetomakedecisionsonhowtorunthebusiness.Thesedecisionswillinvolvethelocationofthestand,thesugarandlemoncontentandthelemonadecolorandprice.Thedecisionsyoumakeinoneperiod,willbethedefaultchoicesforthenextperiod.Attheendofeachperiod,youwilllearnwhatprotsyoumadeduringthatperiod.Youwillalsohearsomecustomerreactionsthatmayhelpyouwithyourchoicesinthefollowingperiods.PreviousManagerGuidelinesDearX,Ihaveenclosedthefollowingguidelinesthatyoumayndhelpfulinrunningyourlemonadestand.Theseguidelinesarebasedonmypreviousexperiencerunningthisstand.Whenrunningmybusiness,Ifollowedthesebasicguidelines:32 Location:BusinessDistrictSugarContent:3%LemonContent:7%LemonadeColor:GreenPrice:8.2francsWiththesechoices,Iwasabletomakeanaverageprotofabout90francsperperiod.Ihaveexperimentedwithalternativechoicesofsugarandlemoncontent,aswellaslemonadecolorandprice.TheabovechoicesweretheonesIfoundtobethebest.Ihavenotexperimentedwithalternativechoicesoflocationthough.Theymayrequireverydierentstrategies.Regards,PreviousManagerCompensation(Thefollowingparagraphisusedintheinstructionsforsubjectsinthetreatmentwiththexedwagecontract.)Youwillgetpaidaxedwageof50francsperperiodduringthe20periodsoftheexperiment.Yournalcompensationdoesnotdependonyourprotsfromthelemonadestand.(Thefollowingparagraphisusedintheinstructionsforsubjectsinthetreatmentwiththepay-for-performancecontract.)Yourcompensationwillbebasedontheprotsyoumakewithyourlemonadestand.Youwillgetpaid50%ofyourowntotallemonadestandprotsduringthe20periodsoftheexperiment.(Thefollowingparagraphisusedintheinstructionsforsubjectsinthetreatmentwiththeexplorationcontract.)Yourcompensationwillbebasedontheprotsyoumakewithyourlemonadestand.Youwillgetpaid50%ofyourownlemonadestandprotsinthelast10periodsoftheexperiment.(Thefollowingparagraphisusedintheinstructionsforsubjectsinthetreatmentwiththeterminationcontract.)Yourcompensationwillbebasedontheprotsyoumakewithyourlemonadestand.Youwillgetpaid50%oftheprotsyoumakeduringthelast10periodsoftheexperiment.However,iftheprotsyoumakeduringtherst10periodsoftheexperimentarebelow800francs,theexperimentwillendearly.(Thefollowingparagraphisusedintheinstructionsforsubjectsinthetreatmentwiththegoldenparachutecontract.)Youwillgetpaid50%oftheprotsyoumakeduringthelast10periodsoftheexperiment.Iftheprotsyoumakeduringtherst10periodsoftheexperimentarebelow800francs,theexperimentwillendearlyandyouwillreceiveapaymentof250francs.A.2ExperimentalDesignThesubjectswereabletomakethefollowingparameterchoices:33 Location=fBusinessDistrict,School,StadiumgSugarContent=f00102;:::;9910gLemonContent=f00102;:::;9910gLemonadeColor=fGreen,PinkgPrice=f00102;:::;9910gThetablebelowshowstheoptimalproductmixineachlocation. BusinessDistrictSchoolStadium Sugar1.5%9.5%5.5%Lemon7.5%1.5%5.5%LemonadeColorGreenPinkGreenPrice7.52.57.5 MaximumProt10020060 Inordertocalculatetheprotsineachlocationwhenthechoicesaredierentfromtheoptimalchoicesabove,weimplementedalinearpenaltyfunctionwitha\roorset0suchthatlossesinabsolutetermsforthesubjectwereimpossible.Ineachlocation,thepenaltyfactorsassociatedwithadeviationofoneunitforeachofthevariablesaregivenbythenexttable. BusinessDistrictSchoolStadium Sugar560.5Lemon560.5LemonadeColor20600.5Price560.5 BElicitingRiskandAmbiguityAversionWemeasuredthesubjects'riskandambiguityaversionbyobservingchoicesunderuncertaintyinanexperimentthattookplaceafterthebusinessgameexperiment.Aspartofthisstudy,thesubjectsparticipatedinaseriesoflotteriesofthefollowingform.RiskAversionLotteryA:Win$10withprobability1=2,orwin$2withprobability1=2.IfsubjectsrejectlotteryAtheyreceive$7.LotteryB:Win$10withprobability1=2,orwin$2withprobability1=2.IfsubjectsrejectlotteryBtheyreceive$6.34 LotteryC:Win$10withprobability1=2,orwin$2withprobability1=2.IfsubjectsrejectlotteryCtheyreceive$5.LotteryD:Win$10withprobability1=2,orwin$2withprobability1=2.IfsubjectsrejectlotteryDtheyreceive$4.LotteryE:Win$10withprobability1=2,orwin$2withprobability1=2.IfsubjectsrejectlotteryEtheyreceive$3.AmbiguityAversionIfaredballischosenyouwillwin$7,ifablueballischosenyouwillwin$2.CaseA:ChooseUrn1containing20ballsthatareeitherredorblueORchooseUrn2containing16redballsand4blueballs.CaseB:ChooseUrn1containing20ballsthatareeitherredorblueORchooseUrn2containing14redballsand6blueballs.CaseC:ChooseUrn1containing20ballsthatareeitherredorblueORchooseUrn2containing12redballsand8blueballs.CaseD:ChooseUrn1containing20ballsthatareeitherredorblueORchooseUrn2containing10redballsand10blueballs.CaseE:ChooseUrn1containing20ballsthatareeitherredorblueORchooseUrn2containing8redballsand12blueballs.CaseF:ChooseUrn1containing20ballsthatareeitherredorblueORchooseUrn2containing6redballsand14blueballs.CaseG:ChooseUrn1containing20ballsthatareeitherredorblueORchooseUrn2containing4redballsand16blueballs.Aftersubjectshadmadetheirchoicesforbothriskandambiguityaversiononelotterywaschosenatrandomandeachsubjectwascompensatedaccordingtohisorherchoice.Theabovelotteriesenableustoconstructindividualmeasuresofriskandambiguityaversion.Foreachmeasurewethenusedthemedianobservationtosplitthesampleintoamoreandalessrisk/ambiguity-aversegroup.CGoldProspectingExperimentInstructions35 Youarenowtakingpartinaseriesofeconomicexperiments.Pleasereadthefollowinginstructionscarefully.Everythingthatyouneedtoknowinordertoparticipateinthisseriesofexperimentsisexplainedbelow.Shouldyouhaveanydicultiesinunderstandingtheseinstructionspleasenotifyus.Wewillansweryourquestionsatyourcubicle.Atthebeginningoftheexperimentsyouwillreceiveaninitialendowmentof$5inadditiontoyourshow-upfee.Duringthecourseoftheexperimentsyoucanearnafurtheramountofmoneybygaininggoldnuggets.Theamountthatyougainduringtheexperimentsdependsonlyonyourowndecisions.Allthegainsthatyoumakeduringthecourseoftheexperimentswillbeexchangedintocashattheendoftheexperiments.Theexchangeratewillbe:3goldnuggets(gn)=$1Attheendoftheseriesofexperimentsyouwillreceivethemoneythatyouearnedduringtheexperiments.Duringeachexperimentyouhavetomakedecisions,whichyouwillenteronacomputerscreen.Thedecisionsyoumakeandtheamountofmoneyyougainwillnotbemadeknowntotheotherparticipants.Onlyyouwillknowthem.Pleasenotethatcommunicationbetweenparticipantsisstrictlyprohibitedduringtheexperiments.Inadditionwewouldliketopointoutthatyoumayonlyusethecomputerfunctionswhicharerequiredfortheexperiments.Communicationbetweenparticipantsandunnecessaryinterferencewithcomputerswillleadtotheimmediateexclusionfromtheexper-iments.Incaseyouhaveanyquestionspleasedon'thesitatetoaskus.OverviewoftheExperimentalProceduresInthisseriesofexperiments,youwillbeplayingseveralinstancesofagoldprospectinggame.Ineachexperimentthegamewillbethesame,butthepayosofyourchoicesandthewayyouarecompensatedwilldier.Inparticular,thechoicesoneachscreenconstituteanentirelynewgame.Eachgoldprospectinggameconsistsof2periods.Ineachperiodyoumaychoosetoprospectgoldinoneoftwomountainsortostayathome.Eachoftheseoptionshasanassociatedrevenueandcost:\Stayathome"alwaysyields0gnanditcostsyou0gntochoosethisoption.\Mountain1"isawell-exploredmountainthatisclosetoyourhome.Whenyouchoosetoprospectgoldatthismountain,ityields100gnwithaprobabilityof50%and0gnwithaprobabilityof50%.Thus,bychoosing\Mountain1"youcannotlearnwhetherityields0gnor100gn.Itcostsyou0.25gntotravelto\Mountain1"andtoprospectgoldthere.\Mountain2"isanunexploredmountainthatisfurtheraway.Withaprobabilityofp\Mountain2"isagold-richmountain,inwhichcaseitalwaysyields100gnwheneveryouchoosethisoption.Withaprobabilityof1 p,\Mountain2"isagold-poormountain,inwhichcaseitalwaysyields0gnwheneveryouchoosethisoption.Youonlylearnthe36 qualityof\Mountain2"ifyouchoosetoprospectgoldatthatmountain.Sinceitisfurtherawayitcostsyou0.75gntotravelto\Mountain2"andtoprospectgoldthere.Finally,notethateachgoldprospectinggameyouplayisindependent.Thus,youcanonlylearnaboutthequalityof\Mountain2"forthatparticularinstanceofthegoldprospectinggame.First,therewillbe3practicetrialswithdierentvaluesofpwhichdonotaectyourcompensationandwillhelpyougetacquaintedwiththeinterface.Afterthat,therewillbeatotalof4experiments,whichdierintermsoftheprobabilitypthat\Mountain2"isagold-richmountain.Atthebeginningofeachexperimentwewillannouncetheprobabilitypthatwillberelevantduringthatexperiment.Theprobabilitypcantakeonthreevalues:5%,35%,45%oritmaybeunknown.Ineachofthe4experiments,youwillbeplayingthegoldprospectinggameunder3dierentcompensationschemes.Thus,thereisatotalof43=12separategoldprospectinggameswhichareallindependent.Yourcompensationwilldependonyourgoldproductioninthetwoperiodsofeachgame.Inparticular,the3compensationschemesare:1.Youwillbepaidaxedwageof0gnperperiodplus10%ofthegoldrevenueduringthe1standthe2ndperiodoftheexperiment.2.Youwillbepaidaxedwageof0gnperperiodplus10%ofthegoldrevenueinthe2ndperiodoftheexperiment.3.Youwillbepaidaxedwageof1gnperperiod.Thedierentcompensationschemesandexperiments(12intotalafterthetrialphase)arecompletelyindependent.Thatis,yourchoicesineachgoldprospectinggamehavenoeectonanyotherchoicesintheexperiment.Whatyoulearnaboutthepayosisspecictoeachgame.Howisyourincomecalculated?Yourtotalincomeinthetwoperiodsiscalculatedinthefollowingway:totalincome=xedwagein1stperiod+xedwagein2ndperiod+revenuein1stperiod*revenuepercentagein1stperiod+revenuein2ndperiod*revenuepercentagein2ndperiod-costsin1stperiod-costsin2ndperiodYourincomeisthereforehigherwhenyourxedwage,yourrevenueandyourrevenuepercentagearehigherandyourcostsarelower.Yourxedwageineachperioddependsonthecompensationoftheexperimentyouarecurrentlyin.Itiseither0gnor1gn.Yourrevenueineachperioddependsonthechoiceyoumadeandthetypeofmountain.Therevenuepercentageis0%or10%anddependsonthecompensationoftheexperimentyouarecurrentlyin.Thatistosayyoumayreceivenoshareoftherevenueoryoureceive37 someoftherevenuedependingonthecompensationofthecurrentexperiment.Finally,asmentionedbeforethecostofchoosingoneofthetwomountainsis0.25gnor0.75gnwhereasstayingathomeiscostless.ExampleLet'sassumethefollowingscenario.Inthe1stperiodandthe2ndperiodyoureceiveaxedwageof0gnandthattherevenuepercentageinthe1stperiodis0%and10%inthe2ndperiod.Let'sassumeyouchoose\Mountain2"inthe1stperiod.Ifthemountainrevenuewas0gninthe1stperiod,theninthe2ndperiodyouchoose\Mountain1".Ifthemountainrevenuewas100gninthe1stperiodyouchoose\Mountain2".CaseA:Revenueof\Mountain2"is0gnin1stperiodYourincomeinthe1stperiodthereforeis0+0*0-0.75=-0.75gnandinthe2ndperiodyouchoose\Mountain1"whichmayyield0gnor100gn(andyouknowforsurethat\Mountain2"yieldsarevenueof0gninthe2ndperiod).Let'sassumethat\Mountain1"yieldsarevenueof100gninthe2ndperiod.Henceyourincomeinthe2ndperiodis0+100*0.1-0.25=9.75gnCaseB:Revenueof\Mountain2"is100gnin1stperiodYourincomeinthe1stperiodthereforeis0+100*0-0.75=-0.75gnandinthe2ndperiodyouchoose\Mountain2".Since\Mountain2"hadarevenueof100gninthe1stperiodyouknowforsurethat\Mountain2"yieldsarevenueof100gninthe2ndperiod.Henceyourincomeinthe2ndperiodis0+100*0.1-0.75=9.25gn38 ReferencesAcharya,V.V.,R.Baghai-Wadji,andK.Subramanian,2009,\LaborLawsandInnovation,"SSRNeLibraryAcharya,V.V.,andK.Subramanian,2009,\BankruptcyCodesandInnovation,"ReviewofFinancialStudies,22,4949{4988.Aghion,P.,J.V.Reenen,andL.Zingales,2008,\InnovationandInstitutionalOwnership,"WorkingPaperAghion,P.,andJ.Tirole,1994,\TheManagementofInnovation,"QuarterlyJournalofEconomics,109,1185{1209.Almeida,P.,andB.Kogut,1999,\LocalizationofKnowledgeandtheMobilityofEngineersinRegionalNetworks,"ManagementScience,45,905{917.Amabile,T.M.,1996,CreativityinContext.WestviewPress,Boulder,CO.Argote,L.,B.McEvily,andR.Reagans,2003,\Introductiontothespecialissueonmanagingknowledgeinorganizations:Creating,retaining,andtransferringknowledge,"ManagementScience,49(4),5{8.Ariely,D.,U.Gneezy,G.Loewenstein,andN.Mazar,2009,\Largestakesandbigmistakes,"ReviewofEconomicStudies,76(2),451{469.Arrow,K.J.,1969,\ClassicatoryNotesontheProductionandTransmissionofTechnologicalKnowledge,"TheAmericanEconomicReview,59(2),29{35.Atanassov,J.,2008,\QuietLifeorManagerialMyopia:IstheThreatofHostileTakeoversBenecialforTechnologicalInnovation?,"SSRNeLibraryAzoulay,P.,J.G.Zivin,andG.Manso,2011,\IncentivesandCreativity:EvidenceFromtheAcademicLifeSciences,"RandJournalofEconomics,42,527{554.Bandiera,O.,I.Barankay,andI.Rasul,2005,\Socialpreferencesandtheresponsetoincen-tives:Evidencefrompersonneldata,"QuarterlyJournalofEconomics,120(3),917.Banks,J.,M.Olson,andD.Porter,1997,\Anexperimentalanalysisofthebanditproblem,"EconomicTheory,10(1),55{77.Bebchuk,L.,andJ.Fried,2004,PayWithoutPerformance.HarvardUniversityPress,Cam-bridge,MA.Bull,C.,A.Schotter,andK.Weigelt,1987,\TournamentsandPieceRates:AnExperimentalStudy,"TheJournalofPoliticalEconomy,95(1),1{33.Chemmanur,T.,andX.Tian,2011,\DoAntitakeoverProvisionsSpurCorporateInnova-tion?,"WorkingPaperChemmanur,T.,X.Tian,andE.Loutskina,2011,\CorporateVentureCapital,ValueCre-ation,andInnovation,"WorkingPaperDickinson,D.L.,1999,\AnExperimentalExaminationofLaborSupplyandWorkIntensi-ties,"JournalofLaborEconomics,17(4),638{670.Fahr,R.,andB.Irlenbusch,2000,\Fairnessasaconstraintontrustinreciprocity:earnedpropertyrightsinareciprocalexchangeexperiment,"EconomicsLetters,66(3),275{282.Falk,A.,andA.Ichino,2006,\CleanEvidenceonPeerEects,"JournalofLaborEconomics24(1),39{57.Fallick,B.,C.A.Fleischman,andJ.B.Rebitzer,2006,\Job-HoppinginSiliconValley:SomeEvidenceConcerningtheMicrofoundationsofaHigh-TechnologyCluster,"TheReviewof39 EconomicsandStatistics,88(3),472{481.Fehr,E.,S.Gachter,andG.Kirchsteiger,1997,\ReciprocityasaContractEnforcementDevice:ExperimentalEvidence,"Econometrica,65(4),833{860.Fischbacher,U.,2007,\z-Tree:Zurichtoolboxforready-madeeconomicexperiments,"Ex-perimentalEconomics,10(2),171{178.Francis,B.,I.Hasan,andZ.Sharma,2010,\DoIncentivesCreateInnovation?EvidencefromCEOCompensationContracts,"WorkingPaperFrederick,S.,2005,\Cognitivere\rectionanddecisionmaking,"JournalofEconomicPer-spectives,19(4),25{42.Frey,B.,andR.Jegen,2001,\Motivationcrowdingtheory,"JournalofEconomicSurveys15(5),589{611.Gabaix,X.,D.Laibson,G.Moloche,andS.Weinberg,2006,\CostlyInformationAcquisi-tion:ExperimentalAnalysisofaBoundedlyRationalModel,"AmericanEconomicReview96(4),1043{1068.Gilson,R.J.,1999,\TheLegalInfrastructureofHighTechnologyIndustrialDistricts:SiliconValley,Route128,andCovenantsNottoCompete,"NewYorkUniversityLawReview,74,575{629.Gneezy,U.,S.Meier,andP.Rey-Biel,2011,\WhenandWhyIncentives(Don't)WorktoModifyBehavior,"JournalofEconomicPerspectives,25(4),191{209.Gneezy,U.,andA.Rustichini,2000a,\AFineIsaPrice,"JournalofLegalStudies,29,1. ,2000b,\Payenoughordon'tpayatall,"TheQuarterlyJournalofEconomics115(3),791{810.Hellmann,T.,2007,\Whendoemployeesbecomeentrepreneurs?,"ManagementScience,53,919{933.Hellmann,T.,andV.Thiele,2011,\IncentivesandInnovation:AMulti-taskingApproach,"AmericanEconomicJournal:Microeconomics,3,78{128.Holmstrom,B.,andP.Milgrom,1987,\AggregationandLinearityintheProvisionofIn-tertemporalIncentives,"Econometrica,55(2),303{328.Kohn,A.,1993,Punishedbyrewards:thetroublewithgoldstars,incentiveplans,A's,praise,andotherbribes.HoughtonMiinCo.,Boston.Lazear,E.P.,2000,\PerformancePayandProductivity,"TheAmericanEconomicReview90(5),1346{1361.Lerner,J.,andJ.Wulf,2007,\InnnovationandIncentives:EvidencefromCorporateR&D,"ReviewofEconomicsandStatistics,89,634{644.Liu,Q.,andK.P.Wong,2011,\IntellectualCapitalandFinancingDecisions:EvidencefromtheU.S.PatentData,"ManagementScience,57(10),1861{1878.Manso,G.,2011,\MotivatingInnovation,"JournalofFinance,66,1823{1860.March,J.G.,1991,\ExplorationandExploitationinOrganizationalLearning,"OrganizationScience,2(1,SpecialIssue:OrganizationalLearning:PapersinHonorof(andby)JamesG.March),71{87.Marx,M.,D.Strumsky,andL.Fleming,2009,\Mobility,Skills,andtheMichiganNon-CompeteExperiment,"ManagementScience,55(6),875{889.McCullers,J.C.,1978,\IssuesinLearningandMotivation,"inTheHiddenCostsofRewarded.byM.R.Lepper,andD.Greene.LawrenceErlbaumAssociates,Hillsdale,NJ.40 McGraw,K.O.,1978,\TheDetrimentalEectsofRewardonPerformance:ALiteratureReviewandaPredictionModel,"inTheHiddenCostsofReward,ed.byM.R.Lepper,andD.Greene.LawrenceErlbaumAssociates,Hillsdale,NJ.Meloso,D.,J.Copic,andP.Bossaerts,2009,\PromotingIntellectualDiscovery:PatentsVersusMarkets,"Science,323(5919),1335{1339.Meyer,R.J.,andY.Shi,1995,\SequentialChoiceunderAmbiguity:IntuitiveSolutionstotheArmed-BanditProblem,"ManagementScience,41(5),817{834.Nalbantian,H.R.,andA.Schotter,1997,\ProductivityUnderGroupIncentives:AnExper-imentalStudy,"TheAmericanEconomicReview,87(3),314{341.Rogerson,W.P.,1985,\RepeatedMoralHazard,"Econometrica,53(1),69{76.Sannikov,Y.,2008,\AContinuous-TimeVersionofthePrincipal-AgentProblem,"TheRe-viewofEconomicStudies,75,957{984.Sapra,H.,A.Subramanian,andK.Subramanian,2008,\CorporateGovernanceandInnova-tion:TheoryandEvidence,"SSRNeLibrarySauermann,H.,andW.M.Cohen,2010,\Whatmakesthemtick?EmployeeMotivesandIndustrialInnovation,"ManagementScience,56,2134{2153.Shearer,B.,2004,\PieceRates,FixedWagesandIncentives:EvidencefromaFieldExperi-ment,"TheReviewofEconomicStudies,71(2),513{534.Sillamaa,M.-A.,1999,\Howworkeortrespondstowagetaxation:Anexperimentaltestofazerotopmarginaltaxrate,"JournalofPublicEconomics,73(1),125{134.Sorenson,O.,andT.E.Stuart,2003,\Liquidityeventsandthegeographicdistributionofentrepreneurialactivity,"AdministrativeScienceQuarterly,48(175-201).Tian,X.,andT.Wang,2010,\ToleranceforFailureandCorporateInnovation,"WorkingPaperWeitzman,M.L.,1979,\OptimalSearchfortheBestAlternative,"Econometrica,47(3),641{654.Yanadori,Y.,andJ.H.Marler,2006,\Compensationstrategy:doesbusinessstrategyin-\ruencecompensationinhigh-technologyrms?,"StrategicManagementJournal,27(6),559{570.41