The intrinsic motivation of bureaucrats is the focus of this paper and three primary results are shown First bureaucrats should be biased Second sometimes this bias takes the form of advocating for their clients more than would their principal while ID: 40372
Download Pdf The PPT/PDF document "The Motivation and Bias of Bureaucrats B..." 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.
givingbeneÞtstotheundeserving.Forinstance,convictingtheinnocentisperceivedaworseerrorthanlettingtheguiltygofree.Foreitherofthesereasons,oversighttendstobebiasedinfavorofclients.Client-favoredoversightim-pliesthattheseclientsbeneÞtfromformedagents,astheycan(atleastpartially)correctmistakesthatharmthembutleaveun-changedthosethatbeneÞtthem.Thiseffectleadstotheprincipalpreferringhostilebureaucrats.CombiningthiswiththefactthatthebeneÞtismorevaluablewhentrulyde-servedÑwhichleadstoademandforadvo-catesÑshowsthatthedesirabilityofclientadvocatesdependsonthreefactors.First,howmuchmoredoestheconsumervaluethegoodwhentheprincipalbelievesittobeneeded?Second,howcapableareclientsatpointingouterrorsthatharmthem?Third,howeasyisittocorrecttheseerrors?Theseobservationsmerelyillustratethatcer-tainkindsofemployeesaredesired.Unfortu-nately,theseworkersmaynotendupbeingemployed.Ifagentpreferencesareeasilyiden-tiÞed,hiringandÞringdecisionscanbebasedonthese.Thisseemsimprobableinmanyoccu-pations;instead,possibleemployeeslikelyhaveprivateinformationontheirpreferences.Ac-cordingly,theissueofwhetherbureaucratswiththeappropriatebiascanbeinducedtoself-selectisconsidered.Itisshownthatwithsuf-Þcientdispersionofpreferencesoverdesiredoutcomes,selectiontobureaucraticpositionsbecomesbifurcated,inthatitconsistsof(a)thosewhoarepreferredbytheprincipal,and(b)thosewhoarepreferredbytheprincipal.Inotherwords,thepeoplewhoarelikelytobedeterredfromapplyingforbureau-craticpositionsarethoseofintermediatepref-erences;thosewithextremepreferencesbecomeSectionIbeginsbyconstructingasimplemodelofbureaucraticeffortexertion.SectionIIillustrateshowtherelationshipbetweenclientandprincipalpreferencesleadstoademandforandSectionIIIconsidersthesourcesofclient-favoredoversight.ThemodelingworkisextendedinSectionIVtodealwithself-selection,andSectionVconsidersapairofrelevantoccupations.SectionVIconcludeswithabriefdiscussion.I.TheModelAnallocationmustbemadetoaclient,cantakeonavalueof0or1.Thisallocationaffectsthewelfareof(a)theclient,and(b)theremainingmembersofsociety.In-dividualsinsocietyvaryinhowtheyweightthese,andtheobjectiveoftheprincipalistomaximizethewelfareofthemedianmemberofTheutilityoftheallocationdependsonameasuringthestateofnature,whichalsotakesonavalueof0or1.TherearethreeactorsinthemodelÑtheclient,theprin-cipal,andthebureaucratÑwhoseincentivesandactionsaredescribedinturn.TheClient.)bethebeneÞtob-tainedbytheclientifhistypeistrulyandtheallocationis.Fornotationalconvenience,let(0,0)(1,0)bethebeneÞtattainedbytheclientwhencorrectlygivenallocation0relativetoincorrectlybeinggivenallocation1,andlet(1,1)(0,1)bethebeneÞtattainedbytheclientwhencorrectlygivenal-location1relativetoincorrectlybeinggivenallocation0.Withoutlossofgenerality,let0.Thus,inapolicesetting,isthebeneÞtofbeingsetfreewheninnocentcomparedtobeingarrested,whileisthebeneÞtofbeingarrestedwhenguiltycomparedtobeingsetfree.ThePrincipal.ÑAssumethatthedirectutil-itytoeachmemberoftherestofsocietyisgiven).However,theyalsovaluetheeffectoftheallocationontheclient:forinstance,societyvaluesthatitspoorbefed.Membersofsocietyvaryinhowtheyweightthese,whereutilityfromtheallocationforapersonoftypeisgivenby),wherethemeasureshowthatpersonweightsbeneÞtstotheclientrelativetotherestofsoci-ety.AssumethatthepoliticalprocessresultsinThisdemandforbiasdiffersfromrecentworkbyPhilippeAghionandTirole(1997),whobuildamodelofdelegatingauthoritybasedonthecongruenceofpreferencesbetweenagentandprincipal.TheresultsareunchangediftheprincipalÕsobjectivealsopartlyweightstheutilityoftheclient,aswouldbethecaseif,say,therewerearandomchanceofeachmemberofthepopulationbeingtheclient.182THEAMERICANECONOMICREVIEWMARCH2007 themaximizationoftheexpectedutilityofthemedianmemberofsociety,isthemedianvalueof.Theoptimalallocation(fromtheperspectiveoftheprinci-pal)involvesmatchingtheallocationtothestateofnature(),soitisassumedthatalloca-tion1ispreferredinstate1,andallocation0instate0.Forsimplicity,beginbyassumingthatthesurplusfromtheallocationissymmetric:1if0otherwise.Thus,socialsurplusispositiveonlyifproperlymatchedtotheunderlyingenviron-,andthecostoferrorisindependentofwhicherrorismade:themarginalcostis1;thecaseofasymmetricerrorcostsisconsideredbelow.Thetruevalueofisunknownandtakesvalues1or0withequalTheAgentorBureaucrat.ÑTheprincipalchoosesanagenttohelpmaketheassignment.Theagentexertsunobservedeffortwhichaffectstheprecisionofherestimateofthecorrectassignment,.Sheobservesasignal{0,1},whichiscorrectwithprobabilitywhereshechooseseffortatacost),where0,and.TheagentÕssignalisinformationandisob-servedbyallparties.Therearetypicallytwowaystoinduceagentstoexerteffort:(a)throughtheuseofmonetaryincentives,and(b)throughwhoishired.Inmostbureaucracies,monetaryincentivesplayaverylimitedrole(seeJohnBrehmandScottGates1997;AvinashDixit2002;andRonaldJohnsonandGaryLibecap1989).Inordertofocusonselectionconcerns,itisassumedherethatmonetaryincentivesarenonexistentbyas-sumingthattheprincipalhirestheagentatan(initially)exogenouswageTheprincipalchooseswhichbureaucrattohireandmaynotchoosethemedianvoter.NetofaÞxedwage,,herpreferencesdependontheallocationvia(1),andhowmucheffortsheexerts.Herutility(netofthewage)isgivenby),whichcanberewrittenasIassumethatthealternativewageofallpos-sibleagentsisindependentofandnormal-izedto0.Thepopulationdistributionofgivenby,withdensitythatrangesfrom.Iassumeinitiallythatisobservedbyallparties.ÑBureaucraciesusuallyhavesomemechanismavailableforcorrectingmistakes.Accordingly,assumethatwhentheagentÕssig-nalinstateiswrong,theprincipalreceivesconvincingevidenceofthiswithprobability).Thus,forinstance,inthepolicingexam-(1)((0))isthelikelihoodofamistakebythepoliceofÞcerbeingdetectedwhenthesuspectisinnocent(guilty).Assumeinitiallythattheseprobabilitiesoferrorcorrectionareexogenous,anissuethatisreturnedtoinSectionIII.LetoversightbeandletoversightbeThus,unbiasedoversightmeansthatmis-takesareequallylikelytobedetected,whilebiasedoversightmeansthatmistakeswheretheclienttrulyshouldbegiventhebeneÞt1)aremorelikelytobedetectedthanwhenheshouldbedenied(0).InSectionIII,Iexploretworeasonswhyoversightmaybebiased:(a)theabilityofclientstocom-plain,and(b)asymmetriesintheprincipalÕsÑThetimingofthegameisasfol-lows.First,theprincipalofferstoanagent.Ifaccepted,heisemployed;otherwise,thegoodisallocatedwithoutanagent.Second,natureas-Thisimportantassumption,discussedintheconclusion,allowsmetoconcentrateonhowbiasaffectseffortdecisionswithoutworryingaboutagentslyingabouttheirsignals.Seetheconclusionforhowtheresultsarechangedwheninformationissoft.VOL.97NO.1PRENDERGAST:THEMOTIVATIONANDBIASOFBUREAUCRATS wheretheprincipalchoosestheagentwhoex-ertsmosteffort,asthisincreasesexpectedsur-plusforthemedianmemberofsociety.Proposition1immediatelyfollows.PROPOSITION1:effortincreasesinclientadvocacythenadvocacydecreaseseffort.then,ifoversightisbiased,clientadvocatesexertlesseffort.Ifoversightisunbiased,allagentsexertthelowestlevelofeffort,andtheprincipalisindifferentoverwhomtohire.then,ifoversightisunbiased,effortisincreasinginaltruism.Ifover-sightisbiased,effortisincreasinginaltruismonlyif 010 Considereachofthesecasesinturn.WhenClientandPrincipalPreferences0.ÑInsomesettings,theinterestsoftheclientandprincipalcoincide.Agoodexampleofthiswouldincludeapatientbeingevaluatedforheartsurgery:onlyiftheclienttrulyneedsthesurgeryishelikelytowantit.Thisimpliesthat0and0.Then(6)alwaysholdsforanyoversightlevels,andtheuniquelyoptimalchoiceofagentisthemostaltruistictotheclient.Theideaistrivial;whenclientshavethesamerankingofout-comesastheprincipal,theyvaluemoreac-curateestimatesoftheirneeds,andtheagentwhocaresmostabouttheclientexertsthemosteffort.WhenClientPreferencesAreOpposedtoThoseofthePrincipal:0.ÑAwiderangeofbureaucraticenvironmentsinvolveclientswhohavepreferencesthatrelatelittletothoseoftheprincipal.Toseehowthisaffectsdesiredbias,Þrstconsiderthecasewhereclientsdesirethegoodatleastasmuchwhentheydonotdeserveitaswhentheydo,where(Forexample,thosewhohavecheatedontheirtaxeswishtoavoidanauditatleastasmuchasthehonest.)Then,iftheinequalityisstrict,,theclientbeneÞtsfromalessinformedagent,independentofwhetheroversightisbiasedornot.Thiscanbeseenfrom(6)andsimplyreßectsthefactthattheclientcaresmoreaboutavoidinginvestigationwhenguiltythanavoidinginvestigationwheninnocent.Next,considerthecasewhereForinstance,thiscouldbeatrafÞcÞneoraprisonsentence,whichhasasimilarcosttotheclientwhetherheisguiltyorinnocent.Beginwiththecasewhereoversightisunbi-asedso(6)becomes0.When,alltheclientcaresaboutistheaverageprobabilityofreceivingthegood,asheweightseacherrorequally.If(1),theprobabilityofreceivingthebeneÞtis,in-dependentofeffort,andsothebureaucrat.Consequently,agentat-tributeshavenoeffectoneffortandthechoiceofbureaucratisirrelevant.Butwhentheclient,onaverage,beneÞtsfromoversight(1)),(6)isalwaysviolatedif,andincentivesaremaximizedbyhiringagentswhoaremosthostiletotheclientÕswishes.Heretheclientstrictlyprefersinac-curatedecisions,becausehecan(atleastpar-tially)correctthemistakesthatharmhim.Hence,theoptimalresponsebytheprincipalistohiretheagentmosthostiletotheclient.Finally,considerthecasewheretheclientwantsthebeneÞt(0),butvaluesitmorewhenheisdeserving:0.ThiscaselikelycoversmanybeneÞtsofferedbybureaucrats.Forexample,thosewhoaretrulyillwantmedicaltestsmorethanthosewhoareusingthemasaprecautionarymeasure,orthosewhoaretrulydestituteneedwelfareassistancemorethanthosewhoseneedsarelessgreat.Inthecaseofsymmetricoversight,thereasonforaltruisticbureaucratsissimplythemirrorimageofthecasewhere0:heretheagentcaresmoreaboutreceivingthebeneÞtwhenwarrantedthanwhennot0,andso,onaverage,valuesmoreaccuracy.Withbiasedoversight,thedecisiononwhethertohireadvocatesdependsnotsimplyonthedifferenceintheÑwhichwouldleadtoadvocatesÑbutalsoontheÕs,wherebiasedoversightleadstoademandforhostileagents.Thelogicforthisissimple:asclientscancorrectVOL.97NO.1PRENDERGAST:THEMOTIVATIONANDBIASOFBUREAUCRATS mistakesthatharmthemandbeneÞtfromthosethatworkintheirfavor,theypreferlessaccuracy.Asaresult,thechoiceofagentdependsonwhicheffectisstronger,anissuethatisaddressedinSectionIII,whenover-sightisendogenized.Thusfar,allthathasbeenshownisthatcertaintypesofagentsworkharderthanothers.Asithasbeenassumedthatwagesaresetexogenouslyandsupplyofthedesiredagentisforthcoming,theprincipalhirestheagentwhoexertsthemosteffort.Aplausiblealternativeassumption,however,wouldbethattheprincipalcanchoosethewage,andwilldosotomatchtheagentÕsreservationInthiscase,theprincipalbecomestheresid-ualclaimantontheutilityofthebureaucrat,aswagessimplychangetoofferthebureaucratherreservationutility.Thisextensionstrengthensthevalueofhiringthehardestworkingagentbecause,holdingallelseconstant,theagentwhoexertsthemosteffortgainsthemostutilityfromthejob.Asaresult,hecanbeofferedalowerwagethansomeonewhoexertslessef-Thisdoesnotchangetheresultabove,asillustratedinLemma1,whoseproofisintheLEMMA1:Assumethattheprincipalcanchooseagentwagesandbinds.Then,ifeffortisincreasing)inthere-quiredwagetoemployaworkerisdecreasingAsaresult,theprincipalhirestheagentwhoexertsthemosteffort.Inotherwords,becauseagentsderiveutil-ityfromaffectingclients,andthosewhocaremostaboutthisarewillingtoexertmoreeffort,theyacceptlowerwagesfortheoppor-tunitytodoso.Hence,theresultsgeneralizetoamoretraditionalassumptiononwageIII.OversightOversightmattersonlyinthe(important)casewhereclientsbeneÞtfromreceivingtheallocation,butatleastasmuchwhendeserved:0.Toaddressthisfurther,Inowendogenizetheoversightdecisionwiththespe-ciÞcpurposeofofferingtworeasonswhyover-sightislikelytobebiased.From(7),itshouldbeclearthattwoissuesaffecthowmuchtheprincipalintervenestocorrectmistakes:(a)howmuchsheknowsaboutmistakesbeingmade,and(b)howmuchshecaresaboutthesemis-takes.Iaddresseachinturn.ClientInterestsPrincipalstypicallyrelyonrelevantinformationtohelpthemintervene.Aprimarysourceofsuchinformationiscomplaintsbyclients.Informationthatmistakeshavebeenmadetypicallycomesfrominterestedparties.SpeciÞcally,thosewhohavebeenwrongedwilloftenvoicetheircon-Tomodelthis,considerasimpleexten-sionofthemodelabovewheretheprincipalmayobservesomeusefulinformationbeforedecidingwhethertoinvestigate.AssumethatintheabsenceoftheprincipalreceivingdisconÞrmingevidencefromthecli-ent,theprincipalreceivesconvincingevidenceofanerrorwithprobability*.Inaddition,however,nowassumethatwhenanerrorhasbeenmade,theclientobserveshardevidenceofthaterrorwithprobability.Inallotherin-stances,theclientobservesnomorethantheagent.Forsimplicity,assumethattheprinci-palÕsandagentÕssignalsareconditionallyinde-pendent.ThesignalreceivedbytheclientisNotethatthisisonlytruethosewithmoredesiredpreferencesexertmoreeffort.Theagentvaluestheutilityoftheclient,independentofwhethersheworks:ifeffortwerenohigherthanthatofanalternative,shewouldgetasmuchutilityviawithoutbeingemployed.BydeÞnition,wheneverapoordecisionhasbeenmadebyabureaucrat,thereissomeonewhohasbeenwrongedwhocouldconceivablyrevealthisinformation.However,acentralissuethatarisesisthediffusionorconcentrationoflosers.Forinstance,takethecaseofasocialworkerwhomakesanincorrectdecisioninassigningbeneÞtstoanapplicant.IftheapplicantisincorrectlygiventhesebeneÞts,societyatlargeloses.Buttheloserstypicallydonotknow,asitisintheinterestofnoindividualtaxpayertomonitorthesecases.Bycontrast,ifanapplicantisincor-rectlydeniedbeneÞts,thisismorelikelytocometolightastheclienthasaconcentratedinterestinrevealingthismis-taketotheauthorities.Asaresult,thereisalikelyasym-metryinmistakescomingtolight.186THEAMERICANECONOMICREVIEWMARCH2007 Toseehowthisaffectstheresults,ignoreclientfeedback,sothattheprincipalreceivesinformationonallerrorswithprobabilityHowever,returntothemoregeneralspeciÞca-tionoftheprincipalÕspreferencesin(1)and(1,1)(1,0)and(0,1)asthecostsoferrorinthetwostates.Intheprevioussection,thesecostswereboth1.Inkeepingwiththediscussionabove,mostoftheseinstancesconcerncaseswherethecostsofdenyingdeservedbeneÞtstoclientsaregreaterthanthecostsofincorrectlyassigningthosebeneÞts.Consequently,assumethatthecostofmistakenlydenyingthebeneÞt(1)ishigherthanthatofmistakenlygivingthebeneÞt:Droppingthesymmetryassumptionhastwoeffectsonthemodel.First,theagentÕssignalmaynolongeraffecttheallocation,inwhichcasethereisnoneedtohireanagent.Forinstance,ifthecostofmistakenlydenyingbeneÞtstoaclientislargeenough(comparedtothecostofmistakenlygiv-ingthebeneÞt)theprincipalwillgivethebeneÞteveniftheagentÕssignalsaysotherwise.Inthiscase,theallocationproblemisnolongerinterest-ing,asthereisnorolefortheagent.Forthisnottobethecase,itisnecessarythat(1).Therefore,assumethatthisisthecase.Second,thereturntocorrectingbureaucraticmistakesthatharmtheclientexceedsthatfromcorrectingtheonesthatbeneÞthim.Howdoesthisaffectbureaucraticbias?Asthemodeliscur-rentlyconstrued,itmakesnodifferencebecausechangingthedecisioniscostless.However,anat-uralextensionofthemodelwouldbetoallowcostsofoverturningdecisionsmadebythebu-reaucrat,asrealisticallyaninvestigationisoftenrequiredbytheprincipal.Assume,accordingly,thattheprincipalfacesarandomcostofÒchangingÓtheallocationfromthatrecommendedbytheagentÕssignal.LetthedistributionofcostsbegivenbyInthiscase,theprincipalwilloverturndeci-sionswhen(a)hereceivesinformationthatanerrorhasbeenmade,and(b)thecostofover-turningthedecisionislessthanitsbeneÞt.Asaresult,itissimpletoshowthatsothat(0).Thepreferredchoiceofbureaucratisstillgivenby(6),butoversightisnowweightedinfavorofinvestigatingin-stanceswherethebeneÞtisdeniedtotheclientforthesimplereasonthattheprincipalcaresmoreaboutthesemistakes.Asaresult,altruisticbureaucratsarenowpreferredifFortherelevantcasewhere0,thisconditionismoredifÞculttosatisfythan(6),andsothosewhoarebiasedagainstclientsaremorelikelytobepreferred.IV.TheSelf-SelectionofBureaucratsTheprevioussectionidentiÞesademandforbiasedbureaucrats.Butthebureaucracywillexhibitthisbiasonlyifappropriatesupplyisforthcoming.Itseemsoftentimesunrealistictoassumethatpreferenceswillbeapparentinap-plicants;instead,theprincipalmaybereliantonothermechanisms.Thissectiondealswiththecasewhereindividualsholdprivateinformationontheirpreferencesandapplyforthepositionbasedonthosepreferences.Forinstance,itislikelythatchild-careworkershavemoreinfor-mationonhowmuchtheylikechildrenthandopotentialemployers.First,notethatthereisasimplereasonwhythemostdesiredagentsarelikelytoapplyfortheposition:byexertingmoreeffortthanwouldanalternative,theyaffectclientsinthewaydesiredbythemselvesandbytheprincipal.Despitethis,ifpreferencesaresufÞcientlydisperse,thebureau-cracybecomescomprisedofboththemostdesiredagentsandthosewithpreferencesThereasonisintuitive:whilethosewhohavethemostdesirablepreferencesaffectclientsinthewaydesiredbytheprincipal,thoseagentswhohavetheleastdesirablepreferencesaffectout-comesinthewaythatdesire,eventhoughitisdiscordantwiththeprincipalÕsdesire.Notethateffortexertionalsochangesfrom(5)whensurplusisasymmetric,butthisdoesnotchangetheselectionconditionfrom(6).Forexample,manycoulddisagreewithapolicythatrestrictshealthorwelfarebeneÞtstoillegalimmigrants.On188THEAMERICANECONOMICREVIEWMARCH2007 pointwhere).)Straightforwardcom-putationsshowthatthisarisesforallwages,wherewhere( e)d2(e )(e )(1*)]/(ee).Proposition2follows.PROPOSITION2:Ifthewageexceedsww( e)d2(e )(e )(1*)]/(ee),allagentsapplyforthepositionandonlythoseöexerthigheffort.Otherwise,thosewhoselecttobecomebureaucratsarethosewithwhoexerteffortofeexerteffortofeInotherwords,forlowenoughwages,bureau-cratsaredrawndisproportionatelyfrombothendsofthedistribution,solongaspreferencesareintenseenough:WillSelectionbeBifurcated?ÑTheprincipalcanaffectwhichequilibriumarisesthroughthechoiceofwage.Thereislittlethatisunintuitivehere.TheprincipalÕsobjectivefunctionislinearintheexpectedeffortoftheagentandisgivengiven*](1 )w.Ifwagesareset,allapplyfortheposition.Butofferingawagethishighhastwoproblems.First,itiscostly.Second,itmaybethatalowerwageoffershigheraverageeffort,ashigherwagesinducemorehigh-effortandlow-effortapplicants.Asaresult,theprincipalmaychooseawagesuchthattheparticipationconstraintbindsforsomeworkers,andsothereisnonrandomselection.Giventhis,whichbureaucraciesaremostlikelytoexhibitthistendency?First,itdependsonbureaucratscaringsufÞcientlyabouttheac-tivity.Asaresult,itwouldnotbesurprisingtoseethatthosewhoassignvisasorwhoprovidesocialwelfarebeneÞtscouldcomefrombothendsofthealtruismdistribution,butitwouldbealessplausibleexplanationforselectionatadepartmentofmotorvehicles,whereitishardtoseethestrengthofpreferencesbeinggreatenough.Intermsofthemodel,istoosmall.AnotherimportantissueistheextenttowhichtheagentispivotalÑ(1)inthemodel.Again,from(19)and(21),notethatas*tendstozero,theonlyfactorthataffectssupplyiswagesratherthan.Animportantcomponentoftheinßuenceofbureaucratsistheextenttowhichtheyareallowedtoexercisediscretion.Inmanybureaucracies,employeesdolittlemorethanrubberstampdocumentsprovidedbytheclient;again,theDMVwouldbeagoodexample.Onlyininstanceswherebureaucraciesmustdelegatedecisionmakingtoagentsdotheseissuesbecomeimportant:rele-vantexampleshereincludepoliceofÞcers,im-migrationofÞcials,andsocialwelfareworkersatadepartmentofchildrenandfamilyservices,whomustusejudgmentindecidingwhethertotakeachildfromafamily.V.ExamplesAsmentionedintheintroduction,thelitera-tureonachievingefÞciencyinpublicbureau-cracieslargelyconsidersmonetaryincentivesasinfeasibleorcounterproductive,andinsteadem-phasizestheempathy(orlackofempathy)thatbureaucratsfeeltowardtheirclients.Thissec-tionoffersexamplesfromtwooccupations.SocialWorkersResearchersinterestedinunderstandingwhysocialworkersperformastheydoroutinelyemphasizetheirempathyandaltruismtowardclients.Forinstance,inasurveyontheprefer-encesofsocialworkers,RobertPeabody(1964)notesthatÒbyfarthemostdominantorganiza-tionalgoalperceivedasimportant...isservicetoclienteleÓ(66),where83percentofsurveyrespondentsviewsuchserviceasimportant,comparedtoonly9percentwhoseeÒobligationtotaxpayersÓorÒassistancetothepublicingeneralÓasimportantconcernsaffectingtheirdecisions.SimilarevidenceisfoundinRichardWeatherlyetal.(1980),whonotethatÒafeelingofwantingtohelpothersÓwasthemainreasoncitedbyworkersintheirjobsatisfaction.BrehmandGates(1997)surveysocialworkersandÞndthatÒsocialworkersrankedtheircustomersasthemostimportantinßuenceoverhowtheyspendtheirtime,ÓwithÒhelpingothersÓasthesinglemostcommonreasonforlikingtheirjobs.Goodsell(1981)describesthealtruismofsocialworkerstowardtheirclientsindocument-ingthatthegreatestcomplaintthatsocialwork-ershaveisÒascarcityofresourceswithwhichtohelpclients.ÓAlongtheselines,Kadushin(1985,233)notestheabsenceofsuchaltruismasasignoftroubleinasocialworker.Thisissupportedbyaseriesofstudiespredictingtheturnoverbehaviorofsocialworkers,suchasVOL.97NO.1PRENDERGAST:THEMOTIVATIONANDBIASOFBUREAUCRATS study,andtoofferatheoryofhowsuchat-tributesaffecttheefÞciencyofbureaucracies.Aswithanymodel,theplausibilityoftheassumptionsuseddeterminesitscredibility.Withthisinmind,considertheinterpretationandplausibilityofsomeofthese.BiasagainstClients.ÑSofar,thepossibilitythatoversightisbiasedagainstclientsÑ (1)Ñhasbeenignored.Thisislargelybecause itsempiricalrelevanceseemslimited.Intheinterestsofcompleteness,considertheoutcomewhenoversightisbiasedagainstclients.Theselargelyareamirrorimageofthoseabove.Spe-ciÞcally,whenincentivesarenotsomisaligned,0,altruisticagentsarenowalwayspreferredÑunlikeinProposition1whereitdependedonparametervalues.Thisisbecause,here,thebiasinoversightagainstclientsexacer-batesthedesireforadvocates.Bycontrast,whenpreferencesarehighlymisaligned,where,thedesiredbureaucratnowdependsonparametervalues,unlikeinProposition1.Thisisbe-causetheclientÕsdesireforalessinformedbureaucratistradedagainstthefactthatoversightharmshimonaverage.Assuch,theresultsaresimplythemirrorimageofthoseabove.RareEvents.ÑInthemodelingsection,thetwostateshavebeenassumedtobeequallylikely.Thisimpliesthatwithoutoversightbytheprincipal,eachallocationisequallylikely,independentofeffort.Butthisisnottrueifonestateismorelikelythantheother.Suppose,forinstance,thattheprobabilityof1is.Theniftheagentexertseffortoflessthanhersignalislessvaluablethantheprior,andso1isimplementedunlesstheprincipalin-vestigates.Asaresult,foreffortlevelsbelowthedistributionoftheoptimalallocationbasedontheagentÕssignalis1.Foreffortlevelshigherthan,theagentÕssignaldominates,andtheexantedistributionoftherecommendedstateis1withprobability.Thus,theexanteexpectedoutcomedependsontheeffortlevel.Thisimpliesthatinequilibriumnoagentexertseffortbetween,becauseitdoesnotchangetheallocation.ThischangestheresultsabovebyaddinganotherfactortotheprincipalÕschoice,namely,bygivingareasontohireanagentwhovaluesthechangeintheclientÕswelfareintherarestate.Thus,ifthelikelihoodofarrestingasuspectislow,theprincipalpreferspoliceofÞcerswhocareespe-ciallyabouterrorswhentheguiltyarenotar-rested,eventhoughrare.ÑThroughoutthepaper,ithasbeenassumedthattheagentÕsinformationisinthesensethatherevidencecanbeobservedbytheprincipal.Thisassumptionisimportant,asitmeansthattheagentcannotlieabouthersignalby,forinstance,suggestingthataclientshouldbegiventhebeneÞtwhenhersignalsaysotherwise.Inapreviousversionofthepaper,allowingthisinformationtobesuchthatitcouldnotbeveriÞedÑwasconsid-ered.IftheagentÕsinformationissoft,theprincipalhastorelyontheagentrevealingittruthfullyratherthanlying.Thisleadstoapos-sibleneedtolimitthebiasoftheagentcom-paredtotheresultsaboveinordertoguaranteetruth-telling,sothatthosewhoarehiredarelessbiasedthanoutlinedhere.Formally,itaddsatruth-tellingconstraint,wherethereturntolyinginoneofthetwostatesisincreasingastheagentmovesfartherfromthemedianvoter.Instate,thisconditionisgivenAstheagentcaresaboutsurplusinadditiontoInthissense,thepapersharessomesimilaritywithDewatripontandTirole(1999),whoshowthatofferingagentsobjectivesthatdifferfromthatoftheprincipalcanincreaseeffortexerted.OnecasewhereitmayhavesomerelevanceiswherethebureaucracyincurscostsfromtreatingclientsandwishestoavoidthesecostsmorethanitvaluestheclientÕswelfare:forinstance,acostlymedicaltreatmentthatthehospitalwouldliketoavoidevenifsociallybeneÞcial,andsoisreluctanttooverturndecisionsthatsavethemmoney.Thereasonissimple.Byexertingloweffort,theclientismorelikelytoexperiencetheexante,morecommon,statethanwithhighereffort,becausetheagentÕsinforma-tionisnotenoughtooutweightheprior.IftheagentvaluestheclientreceivingthebeneÞtofthemorecommonstate,thisbecomesanotherreasontoexertnoeffort.Tocounter-actthis,theprincipaldesiresanagentwhoespeciallycaresabouterrorsintherarestate,i.e.,lettingtheguiltygofree.VOL.97NO.1PRENDERGAST:THEMOTIVATIONANDBIASOFBUREAUCRATS willingnesstohelpinexplainingrequirementsmaybeasubstantialpricetopayforhostilebureaucratsexertingmoreeffort.Ontheotherhand,thealtruismthatmaybedesiredbysocialworkersintermsofinducingeffortmaymakethemunwillingtocheckuponitemsthatmaymakeclientsineligibleforbeneÞts.Asaresult,thebeneÞtsdescribedaboveshouldbeseenasoneinputtothedecisiononwhomtohirewhenthereareaspectsofbureaucratsÕjobsotherthaneffortexertion,asabove.Finally,itisworthwhiletorelatetheseresultsbacktothetwoeconomicsliteratures.First,theliteratureoncompensatingdifferentialshaslongrecognizedthatemployeeshavejob-relatedpreferences.Butunliketheusualmodelofcompensatingdifferentials,thepremisehereisthatevenaftercontrollingforwages,employerscareaboutwhotakesthejob.Second,considertherelationshiptotheliteratureonincentiveprovision.Mostorganizationsusecompensa-tionpartlyforthepurposeofselectingworkerswithappropriatecharacteristics,andmuchoftherecentempiricalworkonincentiveshasemphasizedtheseselectioneffects(seePrender-gast1999fordetails).Perhapsthemostimpor-tantÞndingofthispaperisthatselectingonintrinsicmotivationiscomplexandneednotleadtoefÞcientoutcomes.Assuch,thispaperoffersanotherreasonwhytheefÞciencyofbu-reaucraciesislikelytobelimited.ROOFSOFPROOFOFLEMMA1:*betheexpectedutilitytotheagentiftheagentwhoisofferedthepositionturnsitdown.Asthegoodisassignedwithoutanagent,thisisequivalenttoanagentwhoexertseffort.(Allowingthealternativemecha-nism,wheretheclientproposesanallocation,whichispreferredwheninterestsarealigned,doesnotchangetheresult.)Giventhis,thewage,),offeredtoanagentoftypeischosensuchthatherutility),asthisconstitutesherreserva-tionutility.TheutilityoftheagentisgivenbysothatThen,notethat dede v(vvm) dede vd(e)de Butif(5)binds, de(vvm) so(24)simpliÞesto)betheexpectedwelfareoftheclientiftheagentexertseffortof.Then),asnoagentisequivalenttoonewhoexertseffortof.Asaresult, Firstconsiderthecasewhere0.Inthisinstance,theprincipalgenerateshighereffortfrommorealtruisticagents:butsoalsoisclientwelfarehigherwithmoreeffortandso0.Hencetheprincipalhiresthemostaltruisticagent.Nextconsiderthecase0andtheeffortishighestfortheleastaltruistic.Then0.Hencetheprincipalhirestheleastaltruisticagent.Acemoglu,Daron,andThierryVerdier.ÒTheChoicebetweenMarketFailuresandAmericanEconomicReview90(1):194Ð211.Inthissense,itfollowsDaronAcemogluandThierryVerdier(2000)andPrendergast(2003)initsfocusontheparticulardifÞcultiesofattainingbureaucraticefÞciency.VOL.97NO.1PRENDERGAST:THEMOTIVATIONANDBIASOFBUREAUCRATS