1SeeCrawfordandSobel1982Olszewski2004Mullainathanetal2008KamenicaandGentzkow2011Dziuda2011andCheetal2013amongmanyothersfortheoreticalresultsonwhenandhowsuchpersuasionispossible2S ID: 396050
Download Pdf The PPT/PDF document "1IntroductionPeopleoftenusecommunication..." 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.
PathwaysofPersuasionLucasComanyBostonCollegePaulNiehauszUCSanDiegoJuly1,2020AbstractWhileeconomictheoriesofpersuasionemphasizetheroleofself-interest,othersemphasizeother-regard.Tostudythesepathways,weintroduceasimpleexperimentalframeworkwheresellersusefree-formconversationtoconvincebuyerstoraisetheirvaluationsforobjects.Wendsellersmorefrequentlytargetbuyers'self-interest,andchangesinself-interestexplainmorevariationinpersuasionoverall.Additionally,sellers'gainsalongonepathwaycomeataconsiderablecostalongtheother.However,whensellerstargetother-regard,theyareattheirmostpersuasive. ThisresearchwasinpartfundedbytheOhioStateBehavioralDecisionMakinginitiative,theUCSanDiegoResearchCommittee,andaUCSanDiegoHellmanAward.WethankJohnBeshears,JasonBlevins,SanfordBraver,KatherineComan,StefanoDellaVigna,ChristineExley,PaulHealy,JohnKagel,IanKrajbich,JereyNaecker,MurielNiederle,AndresSantos,JoelSobel,andseminarparticipantsatUCBerkeleyHaasSchool,BrownUniversity,SITEExperimental,UCSanDiegoRadySchool,UCSanDiego,UCSanDiegoPsychology,UniversityofPittsburgh,StanfordUniversity,HarvardUniversity,UCSantaBarbara,Cornell,Michigan,theFlorenceWorkshopandESASantaCruzforhelpfulcomments.ThispaperbenetedgreatlyfromtheresearchassistanceofWilliamBiscarri,EricaNg,HaochiZhang(mainexperiments),NatalieMouzooni(codingexperiments),andJohnConlon(analysis).Anearlierversionofthepapercirculatedunderthetitle\InterpersonalIn uence."yDepartmentofEconomics,BostonCollege,140CommonwealthAve,ChestnutHill,MA02467.lu-cas.coman@bc.edu.zNBERandDepartmentofEconomics,UniversityofCaliforniaatSanDiego,9500GillmanDrive#0508,SanDiego,CA92093-0508.pniehaus@ucsd.edu.1 1IntroductionHowdopeoplepersuade?Considerasalespersoninteractingwithapotentialbuyer,forexample.Shecouldappealtothebuyer'sself-interest,describingappealingfeaturesoftheproduct.Alternatively,shecouldtrytocultivatethebuyer'sother-regard:beinglikeable,buildingrapport,andsoon.Thispaperexaminesasimplebutimportantquestion:whatrolesdothesetwopathwaysplayinpersuasion?1Wedeneanexperimentalframeworktoaccountquantitativelyfortherolesofself-interestandother-regardinpersuasion.Wefocusonasalessettinginwhichsubjectsplaytheroleofeithersellerorbuyer.2Buyersspeak(bytelephone)withsellers,whoreceiveacommissionthatincreasesinexpectationwiththebuyer'swillingnesstopayforanobject,suchasaniPad.3Buyersandsellerscommunicatehowevertheywish,givingsellersfreedomtochoosetheirownpersuasivestrategy.Wethenmeasurehowbuyers'valuationsdierasaresultoftheseconversations,whichforthepurposesofthepaperishowwedeneandmeasurepersuasion.Todecomposethesedierencesintocomponentsduetoself-interestandother-regard,weelicitbuyers'willingnesstopayundertwoalternativeassumptions:thatthesellerwillreceiveacommission,andthatshewillnot.Changesinthelattervaluationisolateeectsonself-interest,whilechangesintheformercapturethecombinedeectofbothpathways.Wealsobenchmarkresultsagainstthosefromaconditioninwhichbuyersdonotconversewithsellersbuthaveasimilaramountoftimetothinkontheirown.Thisapproachletsusquantifytheimportanceofeachpathwayandalsoexaminehowtheyinteract,e.g.asstrategiccomplementsorsubstitutes.Itis,however,purelyanaccountingdecomposition:itdoesnottellus,forexample,whatitisaboutbuyers'underlyingpreferencesthatcausesthemtotakesellers'payosintoaccountwhenmakingtheirdecisions.Despiteseller'scon ictofinterest,buyerschangetheirvaluationsbyan(absolute)averageof16%followingconversations,andincreasetheirvaluationsonaveragesignicantly,by5%,comparedtoonly1%withoutcommunication.Thesearesizeableeectsgiventhatsellerswereamateursandspokewithbuyersforjust3.5minutesonaverage.Bothself-interestandother-regardmatterforpersuasion,withself-interestaccountingfor 1Evenwithinasingledisciplinesuchaseconomics,analysesofpersuasionandofother-regardingpreferenceshaveremainedlargelysiloed.Asacrudeindicator,thevemodelsofother-regardingpreferenceswiththemostGoogleScholarcitations(Loewensteinetal.(1989),FehrandSchmidt(1999),Rabin(1993),BoltonandOckenfels(2000),andCharnessandRabin(2002))arecitedbyzeroeconomicsorpsychologypaperswith\persuasion"inthetitle.Thereisofcourseevidencethatcommunicationcanaectother-regardmorebroadly.SeeforexampleYamamorietal.(2008),AndreoniandRao(2011),andGreineretal.(2012)oncommunicationindictatorgames;CharnessandGneezy(2008)andBohnetandFrey(1999)on(removing)anonymity,Landryetal.(2006)onface-to-faceinteractionandattractiveness,Roth(1995),Dawesetal.(1977),andBuchanetal.(2006)onirrelevantversusstrategicconversation,andBrandtsetal.(forthcoming)onrapport-buildingconversation.Thereisalsorelatedworkonhowmaterialgiftsaectother-regardandsubsequentbehavior.Forexample,thegiftexchangeliteraturesurveyedbyCooperandKagel(2009)exploreshowemployeesreciprocateabove-marketwageswithincreasedproductivity;MalmendierandSchmidt(2012)showthatbribescanaectwillingnesstopurchasefromthebriber.2Asalessettingisconvenientbecausethe\receiver"'sdecision,howmuchtopay,isstraightforwardtomeasureandinterpret.Theconceptsofself-interestandother-regardseemaslikelytoapplytootherimportantpersuasioncontexts,however{whomtohire,forwhomtovote,etc.3Sellers'incentivesarecommonknowledge,sothatdisclosureisnotasalientchoice.SeeLoewensteinetal.(2011)forevidenceontheeectsofdisclosingcon ictsofinterest.2 roughlytwiceasmuchofthecross-sectionalvariation.Further,thebulkofsellers'nancialgainsareduetochangesinself-interest:theoverallmeanchangeinself-interestis$18,whilethemeanchangeinother-regardis$4.However,nurturingotherregardispossibleandeective.Foraminorityofsellers,targetingotherregarddoesseemtobeanintentionalandeectivestrategy.Inthe19%ofcaseswheresellers'targetedother-regard,theyincreasedbuyervaluationsbymorethantwicetheaverage.Themechanicsofpersuasionlikelyvarywithmanyelementsofcontext:howexperiencedtheparticipantsare,whatdecisionstheymustmake,throughwhatmediumtheycommunicate,andsoon.Ourexperimentalframeworkcanaccommodatevariationofanyoftheseelements.Inthispaperwetakeamodestrststep,examininghowresultsvarywiththenatureoftheobjectbeingsold.Wechoseobjectstovarythedegreeofinformationasymmetrybetweenbuyersandsellers:twoareconsumerelectronicsdeviceswithwhichsellershad(bydesign)morefamiliaritythanbuyers,whiletheothertwowerealotteryticketandapost-datedcheckaboutwhichbothpartiesweresymmetricallyinformed.Wendthatsellersaremoreeectiveatsellingtheconsumerelectronics,butdoalsondthattheysignicantlyincreasemeanbuyervaluationsforthelotteryandpost-datedcheck,withthelattergainsdierentiallyduetoincreasesinother-regard.Whileonlysuggestive,thisisconsistentwiththeideathatother-regardmayplayalargerroleinpersuasionwhereinformationislessasymmetric.Wealsoexaminehowpersuasionvarieswithsimilaritybetweenthebuyerandseller.Inprincipleonecouldimaginethatsimilarityfacilitatespersuasionforavarietyofreasons:itcouldenhancetrust,improvecommunication,orsimplyleadthebuyertofeelmorealtruistictowardstheseller.Consistentwiththisidea,wendevidenceofaformofhomophily:per-suasionissignicantlyhigherwhenthebuyerandsellerarethesamegender,orwhenthebuyerperceivesthesellerasmoresimilartohim.Persuasionisalsosubstantiallyhigherwhenbuyerandsellerareofthesamenationality,thoughthisdierentisnotstatisticallysignicant.Theeectsofsimilarityonpersuasionappeartoworkthroughbothself-interestandother-regard,butwithsomewhatstrongerevidenceforself-interest.Otherworkhashighlightedtheimportanceofsimilarity,e.g.electingforpreventivehealthmeasuresafterspeakingwitharace-matcheddoctor(Alsanetal.,2018),doingbetterinschoolwithrace-matchedteachers(Dee,2004)orgender-matchedteachers(MuralidharanandSheth,2016),andfemalestudentstakingmoreSTEMclassesafterhavingafemaleprofessor(BettingerandLong,2005;Carrelletal.,2010).Ourresultsproviderstevidencehowsimilaritypotentiatespersuasion.Next,weexamineinteractionsbetweenthetwopathways.Ouraccountingdecompositionofpersuasionintoself-interestandother-regardingcomponentsisalsoarststeptowardsunderstandinghowthetwoaecteachother.Manymechanismsseemplausible:buyersmightputmorefaithinadvicefromsellerstheylike,forexample,ordislikepeoplewhotrya\hardsell."Inthedata,changesinself-interestandother-regardaresignicantlynegativelycorrelatedwitheachother,bothunconditionally(-0.64)andafterprojectingthedataontocovariatestoremovepotentiallycorrelatedmeasurementerror(-0.70).Thisrelationshipdoesnotappeartore ectmechanicaltradeosbetweenuseofconversationtime,asitisroughlyinvarianttooverallconversationlength.Sellersthusappeartofaceatrade-o,withgainsalongonepathwaypartiallyosetbylossesalongtheother.Wealsoexaminedescriptivelytheroleoftrust,usingindependentcoderratingsofthe3 trustworthinessofeachsellerbasedontheirconversationtranscripts.Somesellersaresys-tematicallyperceivedasmoretrustworthythanothers,andperceivedtrustworthinessstronglypredictspersuasion:sellersperceivedasonestandarddeviationmoretrustworthyseeanin-creaseinpostvaluationof$11,ora79%increaserelativetothe$14meanincreaseinvalu-ationafterconversation.Interestingly,thiseectworksentirelythroughself-interestandnotthroughother-regard,suggestingthatbuyersdonotfeelmorealtruisticorreciprocaltowardsmoretrustworthysellers.Finally,weexaminewhatresearchersworkingonpersuasionbelievemakesaconversationpersuasive,andhowwelltheirmetricspredicttheresults.Specically,wegatherpredictionsfromaninterdisciplinarysetofresearchersworkingonpersuasiononwhatRA-codedfeaturesofconversationtranscriptswouldbestpredictpersuasion.Takentogether,thesepredictionsprovidearoughsnapshotofthestateofknowledgeaboutpersuasion.Theresearchersexpectbothappealstoother-regardandself-interesttoplayarole:21itemswerecodedbyRAsascapturingappealstoself-interestand15toother-regard.However,exactlywhattheresearcherswerecapturingbeyondthesetwopathwaysvariedgreatly.Congruenceamongresearcherswaslow,suggestingaliteratureyettoconverge:onlythreeitemsappearedatleasttwice,withonly`lengthoftheconversation'appearingthreetimes.Predictivepowerislowoverall,suggestivethereisroomtodevelopmorepredictivetheoriesofpersuasivecontent.Ourexperimentbuildsontwostrandsofpersuasionwork.Thersthasexaminedtheeectsofcommunicationinsettingswhereonlyoneofthetwopathwaysisoperative.Indictatorgames,forexample,thereisnouncertaintyaboutthebestself-interestedactionandyetpersuasionispossibleviaappealstoother-regard(e.g.Yamamorietal.(2008)).Insender-receivergameswherethesendercanonlysendmessagesaboutthereceiver'spayo,foranotherexample,communicationthroughself-interestispossible(e.g.Dickhautetal.(1995)).Collectivelythesestudiesshowthatbothpathwaysarepossible,andmotivateourinterestinsettingswherebothcanoperateatthesametime.Asecondstrandofworkhasexaminedaggregatepersuasioninsuchsettings(seeRoth(1995));ourcontributionistodecomposethisaggregate,testingforandquantifyingtheroleofeachpathwayandtheirinteractions.Thishelpsusunderstandnotonlywhetherpersuasionoccurs,butwhy.Wealsobuildonalargetheoreticalliteratureonpersuasionineconomicswhichhastyp-icallyfocusedonself-interest,4andonworkinpopularpsychologywhichhasmoreoftenemphasizedtheroleofother-regard.DaleCarnegie'sself-helpclassic\HowtoWinFriendsandIn uencePeople"promisedtoreveal\thesixwaysofmakingpeoplelikeyou"(Carnegie,1936),andCialdini(1993)arguesthat\peopleprefertosayyestoindividualstheyknow 4Forthepurposesofthispaper,wedenepersuasionbroadlyasthechangeinthereceiver'sactionthatisinducedbyinteractionwiththesender.Inthissense,relatedworkincludesmodelsofinformation-basedpersuasion(e.g.CrawfordandSobel(1982),PettyandCacioppo(1986),KamenicaandGentzkow(2011),Olszewski(2004),Dziuda(2011),andCheetal.(2013))aswellastaste-basedpersuasion(e.g.BeckerandMurphy(1993)).Ourexperimentaldesignisarguablylessdirectlyrelatedtothe\Bayesianpersuasion"(KamenicaandGentzkow,2011)approach,whichstudiescontextsinwhichsenderscancommittoconditioningmessageprobabilitiesonstatesoftheworld,aswedonotprovidesellersinourexperimentwithmeanstocommit.Otherapproachestoconceptualizingpersuasionincludemodelsthatincorporatebehavioralassumptions(e.g.MurphyandShleifer(2004),Mullainathanetal.(2008))andthatdistinguishbetweenbelief-basedandpreference-basedpersuasionDellaVignaandGentzkow(2009).4 andlike."Inasimilarvein,butalaboratorycontext,Brandtsetal.(forthcoming)ndthat\friendliness"infree-formconversationsiscorrelatedwithhigherearningsincontractrenego-tiation.Ourresultsalsoaddtoworkonother-regardingpreferences,whichhasbeguntodemon-stratetheroletheyplayinreal-worldsettings,fromworkplaceproductivity(BellemareandShearer,2009)toworkplacesatisfaction(Cardetal.,2012)tolaborrelations(KruegerandMas,2004)tocharitablegiving(AndreoniandPayne,2013).Persuasionseemsanotherimpor-tantcontextinwhichtounderstandtheroleofother-regardgivenitscentralroleineconomiclife,accountingforasmuchas25-30%oftheUSeconomy(McCloskeyandKlamer,1995;Antioch,2013).2ConceptualFrameworkConsiderabuyerbwithquasi-linearpreferencesgivenbyu(c;a;; )=c+v(a;)+w(at; )(1)Herecrepresentsthevalueofconsumptionofallothergoods,vthevalueofadiscretegoodthebuyermaypurchase,andwthebuyers'preferencesovertransferstoasellers.a2f0;1gindicateswhetherthepurchaseismadeatpricepinwhichcasethesellerreceivescommissiont.Thebuyer'sconsumptionisthusc=ybap.Theparametersand captureuncertaintyabouttheutilityofthegoodtothebuyerandaboutthebuyer'slevelofcareorconcernfortheseller,respectively.Forsimplicitywenormalizev(0;)andw(0; )tozero.Givenbeliefs,thebuyer'swillingnesstopayWTPcforthegoodisWTPc=E[v(a;)+w(t; )](2)Theexpressionontheleftistheincrementalutilityfromconsumingpc,whilethatontherightisthesumofthebenetsfrompurchasingthegood,includingboththeprivatebenetsandtheeectsthroughother-regard.Nowsupposeinsteadthebuyerisoeredtheopportunitytopurchasethegood,butistoldthesellerwillnotreceiveacommissionregardlessoftheoutcome.HiswillingnesstopaypncisdenedbyWTPnc=E[v(a;)](3)Noticethatifthebuyervaluestheseller'sincomethenthiswillbelessthanhiswillingnesstopaywithcommission,whileifhedoesnotitcouldbethesameorevengreater(ifhedislikestheseller).Wedenepersuasioninthissettingasthechangepcinthebuyers'valuationinducedbyinteractionwiththeseller.Mechanically,thisdecomposesintoachangeduetoself-interest5 andoneduetoother-regard:WTPc| {z }Persuasion=WTPnc| {z }Selfinterest+(WTPcWTPnc)| {z }Otherregard(4)=E[v(a;)]+E[w(t; )](5)Noticethatthisispurelyanaccountingdecomposition,andassuchallowsformanypossibleinteractionsbetweenthetwopathways.Wediscussexamples,andanalyzepotentialinter-actionsinsection5.5.Itisalsoabroaddistinctionthattakesnostandontheunderlyingreasonsforornatureofother-regardingpreferences.Abuyermightvaluetransferstothesellerbecausehecaresaboutequity,orbecausehecaresaboutaggregateeciency,orbecausehefeelssomesenseofreciprocitytowardssellerswhohavebeenhelpful,forexample,amongmanyotherpossibilities.Thedecompositionshouldbeunderstoodasarst,simplestepto-wardsamorenuancedanalysisoftheunderlyingpreferencesthatdriveresponsestopersuasivecommunication.InSection5.6wedoexaminetheroleofonespecicfactor,trust,thatcouldpotentiallyaectpersuasionthroughbothchannels.3ExperimentalDesignWeorganizetheexperimentasaseriesofinteractionsineachofwhichabuyersubmitsaninitialvaluationforagood,thebuyerisrandomlymatchedtoaseller,thebuyerandsellerhaveafree-formtelephoneconversationlastingatmost10minutes,andthebuyerthensubmitsanupdatedvaluation.5Itiscommonknowledgethattheseller's(expected)payoincreasesinthebuyers'willingnesstopay,creatinganincentiveforhertopersuadethebuyer.Sellerswerenotgivenanyinformationaboutbuyers,however,includingtheirexanteorexpostvaluations.Wemeasurepersuasionbyelicitingthehighestpricebuyersarewillingtopaybothwith(pc)andwithout(pnc)commission,bothbeforeandafterthephonecall,resultinginfourseparatevaluations.Identifyingpersuasionwithin-subjectinthiswaymaximizesourstatisticalpowertoanalyzeitsdeterminants.Specically,thebuyerrstmakesaseriesofincentivizeddecisions,choosingbetween$500ortheobjectplus$X.Astheinstructionsexplaintothesubjects,thepriceofthegoodisthus$500-Xforthatdecision.WevaryXfrom$0to$480inincrementsof$20.(SeeTableA-1,PanelA.)Wedenepncas$500lessthesmallestXsuchthatthebuyerchoosesthegoodplus$Xover$500.Thebuyerthenmakesasecondseriesofincentivizeddecisionsbetweeneither$500forherselfand$0fortheseller,ortheobjectand$Xforherselfand$500fortheseller.6(TableA-1,PanelB.)Theonlychangefromtherstelicitationisthatthesellernowreceives$500ifapurchaseismade,thuslettinguscapturepc.7 5Theparadigmismostsimilartothe\Oral"conditionfromMobiusandRosenblat(2006),whereonesubjectattemptstopersuadeanotheroverthecourseofave-minutefree-formtelephonecall.6Wesizedcommissionstocreatemeaningfulincentivesforsellerstoexerteort,knowingthatanygivenbuyerchoicehadalowchanceofbeingimplemented,ratherthantore ecttypicalreal-worldvalues.7ThisdesignissimilartoBremzenetal.(2015)whodisentangleproductvaluationandaltruismbyseparatelyelicitingavaluationforthegoodaswellasthatgoodwithanother-regardcomponent.6 Weincentivizevaluationdecisionsasfollows:Foreveryftybuyers,acomputerlotteryheldroughlyoneweeklaterrandomlychoosesoneandthenchoosesoneofherdecisionsacrosstheentireexperimentalsessiontoimplement.8Forexample,ifacommissionquestionischosenwhereshechoosesthegood,sheisgiventhegoodand$X,andthesellerispaid$500.Sinceanyquestionmightbeimplemented,thebuyer'sincentiveistoreporttruthfullyherpreferencestoanyonequestion.Further,sinceonlyonequestionisimplemented,eachquestioncanbeconsideredinisolation,thusnegatingmotiveslikehedging(Azrielietal.,forthcoming).Becausebothcommissionandno-commissiontreatmentsareequallylikelyatthetimeconversationstakeplace,incentivesdonotvaryacrossconversations.Atthesametime,theseller'smaterialincentiveistoincreasethenumberofquestionsinwhichthebuyerselectsthealternativeinwhichthesellerreceivesacommission,asahigherpcincreasesthelikelihoodthesellerispaidherbonus.Thisisexplainedtothesellerinherinstructions(allexperimentalinstructionsandquestionnairesavailableonbothauthors'websites).Thisexperimentaldesignisintentionally exible,withscopetovarymanyelementsofcontext.Wetakerststepsinthisdirection,varyingthesetofobjectsforsale.Eachsubjectparticipatesinfourbuyer-sellerinteractions,alwaysinthesamerole,withanewcounterpart,andwithadierentobjectforsale.Wechoseobjectstoinducevariationinthedegreeofinformationasymmetrybetweenbuyersandsellers.Twooftheobjectswerenew,populartabletcomputers:the\NewAppleiPad"andthe\MicrosoftSurface".Thesetwoproductshavefeaturesthatcreaterichcontextfordiscussionand(potentially)uncertaintyaboutval-uationforthebuyer,whomaynotknowaboutallofthem.Ameaningfulproportionofoursubjectpoolalreadyownedoneofthesetwoproducts,whichenabledustorecruitsellerswithmorefamiliaritythanthetypicalbuyer(seebelow).Theothertwoobjectswerea$500checkpost-datedforJanuary1stofthefollowingyear(8{11.5monthsinthefuture),andalotterythatpaid$1,000with50%chanceand$0otherwise.Incontrasttothetablets,therewasnouncertaintyabouttheobjectiveattributesofthesegoods.Buyervaluationsfortheseshouldintheorydependonlyontimeandriskpreferencesandhenceshouldberelativelydicultforsellerstoin uencethroughtheself-interestpathway.Theorderinwhichsubjectsdiscussedthesefourproductswasrandomizedatthesessionlevel,andsubjectsdidnotlearnwhichproducttheywouldbediscussingineachconversationuntilafterbuyer-sellerpairingshadbeenmade.BuyersandsellerswererecruitedtodierentbuildingsontheOhioStateUniversitycam-pus.Theywerenotmadeawareoftheotherroomfullofparticipantsuntiltheinstructionswereread.BuyerssatintheEconomicsexperimentallaboratoryandreadinstructionsonthecomputerwithonlyclarifyingquestionsanswered.Theirvaluationelicitationsweredoneinprivateattheircomputerterminaloncetheyconrmedtheinstructionswereclear.Theythendialedintocallswithsellersusingcordlesslandlines.Sellerssatinindividualrooms,preventingthemfromoverhearingtheconversationsothersellerswerehavingand(potentially)learningfromtheirapproaches.Sellersreadtheirinstructions,withonlyclarifyingquestionsanswered.Tomeasuresellerapproaches,justbeforeeachphonecall,throughanonlinesurvey,they 8Westratiedthedrawingsothatanequalnumberofcommissionandnon-commissionquestionswouldbechosen,whichletustellsellersthepreciseprobabilitiesofcommissionsbeingrelevant.7 providedfreeresponseanswerstotwoquestions:Whatareyourgoalsforthisconversation,andhowdoyouplantoachievethem?Theythendialedintocallswithbuyersusingtheircellphonesandheadsetsprovidedbytheresearchers.Callsweremadeandrecordedusingathirdpartyconferencecallservice.Callsendedattheseller'sdiscretionoraftertenminuteshadpassed.AftereachconversationandWTPelicitation,buyerslledoutanunincentivizedsurveyabouttheinteraction.3.1NoCommunicationConditionAfterconducting11experimentalsessionsusingthedesignabove,andbasedonearlyfeed-backonthedatafromtheserounds,wealsoconductedadditionalexperimentalsessionsinwhichweimplementedasecond\nocommunication"condition.Thepurposeofthisconditionwastomeasurefactorsotherthancommunicationthatcouldchangevaluations.Inthiscondi-tion,buyerswaitedforaboutthelengthofaphonecallbeforesubmittingexpostvaluations.Valuationscouldchangeinthisconditionasbuyersintrospectedontheirdecision,forexample.ComparingthevaluationchangesweobserveintheCommunicationandNoCommunicationconditionsletsusestimatetheeectsofcommunication,netofanyeectsoftimetore ect.3.2ResearcherPredictionsandConversationCodingBesidesvaluationsandsurveyresponses,messagesbetweenbuyersandsellersthemselvesarepotentiallyusefuldata.Recentworkhasfoundthatsuchcontentcanhavesurprisingpredictivepowerandyieldinsightsintothenatureofcommunication(Pennebaker,2011).Wethereforerecordedandtranscribedbuyer-sellerconversations.9Wethencodedasetofpredeterminedvariablesdescribingtheseconversations.Wedeter-minedvariablestocodebysolicitingproposalsfromotherresearchersworkingintheareaofpersuasion:specically,wecontactedeightresearchersfromacrosseconomicsandpsychology,gavethemtheinstructionsfromourexperiment,andaskedthemtodescribevevariableseachthatcouldbeobtainedfromtranscripts,usingeithercomputerorhumancoding,thatwouldbestpredictchangesinbuyervaluations.Weinformedthemthatwewouldestimatere-gressionmodelsinteractingtheirpredictorswithproductxedeects,identifythesubmissionwiththehighestR2,andpublishthenameofthewinnerwhilekeepingothersanonymous.10Wetookthisapproachtodisciplineouranalysiswhilealsoobtainingasnapshotviewofhowresearcherscurrentlythinkaboutpersuasion.Sincetheinterpretationofmessagesisofteninherentlysubjective,werecruitedcodersfromthesamesubjectpoolasthebuyersfromouroriginalexperimentinordertocaptureinter-pretationsthatare(statistically)thesameasthoseoftheoriginalaudience.Weincentivizedthoughtful,honestcodingusingthecoordinationgamestructureproposedbyHouserandXiao(2011):twocodersareanonymouslypairedwitheachother,readthesametranscript,answerthesamequestion(e.g.\Howlikeabledidyoundthisseller?"),andarepaidiftheiranswers 9Duetotechnicaldicultieswiththeteleconferencingservice,thephonesthemselves,andtheonlinerecord-ingandhostingservice,weeitherfailedtorecordorlosttherecordingsforsevenconversations,leaving257conversationsforthecontentanalysis.10Wecongratulatethewinner,JoelSobel,whosemeasurespredicted18%ofthevariationinpersuasion.8 match.Whentherewasdisagreementamongcoders,ourRAprovidedatie-breakingvotetoresolve.4RecruitmentandDataDescription4.1RecruitmentWerecruitedsubjectsviaanemailtotheOhioStateEconomicsexperimentsubjectpoolinvitingthemtotakeatwo-minuteonlineintakesurveywhichwouldqualifythemforourexperiment.Toidentifysubjectswithmoreorlessexperiencewiththetabletgoodsinourexperiment,weaskedsurveyrespondentsabouttheirownershipofandfamiliaritywiththeAppleiPadandtheMicrosoftSurface(amongother\ller"questionstohidethenatureoftheexperiment).WethenassignedthosewhoreportedowningeitheraniPadoraSurfacetobesellers,andthosewhoreportedowningneithertobebuyers.Wedidnottellsubjectswhatroletheywouldbeplaying,sothattheonlyperceptibledierencewasthelocationoncampustowhichsubjectswereinvited.Werstheld11experimentalsessionsduringwhichweobserved264buyer-sellerconversa-tions.Wepresentresultsfromthesesessionsbelowasourprimaryresults.Basedonfeedbackontheseresults,wesubsequentlyran4additionalexperimentalsessionsinwhichweincludedbothCommunicationandNoCommunicationarms,withbuyersexperiencingbotharmsinthosesessions.AppendixTableA-2summarizesparticipationbysession.11Overall,sixty-sixsellersparticipatedinfourconversationseach.Fifty-vebuyersparticipatedinsessionsinwhichtheyspokewithfoursellers,while28buyersparticipatedinsessionsinwhichtheywererandomlyassignedtothecommunicationconditiontwiceandthenocommunicationconditiontwice.12Subjectswerepaid$15forparticipationplustheirpaymentsfrombuyers'choicesandthelotterythefollowingweek.Nosessionlastedlongerthantwohours.4.2DataDescriptionAppendixTableA-3summarizesparticipantcharacteristics.Buyersandsellersaresimilardemographicallybutdier(bydesign)intheirproductownership.Buyersaresomewhatfamiliarwiththeseproducts,however,as86%and16%,reporthavingprevioususedaniPadoraSurface,respectively.FigureA-1displaysaverageinitialvaluationsforthefourproducts.13Themeaninitialvaluationacrossallproductsforthegooditselfis$279,whichincreasesby$14whenacom-missionisadded(p=0:06).Buyervaluationsofthetabletsareasexpectedgenerallybelowtheir$500marketpriceatthetime,withmeanvaluationsat$282fortheiPadand$236for 11Wefocusontherst11sessionsastherecouldpotentiallybeaninteractionbetweenthetwoarmsinthelattersessions.Theresultsbelowdonotqualitativelychange,therelationshiporeectsizesweestimatearemoderatelysmaller,whenconversationdatafromsessions12-15areincluded(SeeAppendixsubsectionA.1).12Samplesizewasdeterminedbythenumberoftabletownerswewereabletorecruitintotheexperiment.13Wedidnotimposemonotonicityonsubjectresponses:subjectscouldreportawillingnesstopayXbutnotX0(X).Of1,328valuationsweelicited,1,306(allbut2%)aremonotonicintheexpecteddirection.9 theSurface.Buyersareimpatient,valuinga$500checkninetotwelvemonthsfromnowat$390,andquiteriskaverse,valuinga50/50chanceofwinning$1,000at$211.4.3ExampleConversationsConversationsvaryenormously.Togivesomesense,weprovideexcerptsherethatcap-turesellerattemptstopersuadeviaappealtoself-interestandtoother-regard,respectively.AppendixCprovidesfurtherexamplesillustratinghowsomesellersgaveupquickly,othersmonopolizedtheconversation,otherssimplylied,etc.Anumberofsellersreportthattheyplantofocusontheattributesandvalueoftheobjectforsale{to\talkabouthowawesometheiPadis,"\tellthemthespecsandconvincethemtheSurfaceisworthfullretailprice,"\giveextrainformation,"and\sellbenets{USB,keyboardfornotes,fastinternet,touchscreencapabilities,nostylusrequiredtotakenotes."Thefollowingconversationillustratesthisapproach:Seller:So,I'dliketotalktoyoualittlebitmoreabouttheiPadasfaras,doyouownaniPadoranything?Er,doyouownaniPhone?Buyer:Uh,Ihaveonly,I'musinganApplerightnow.Seller:Okso,theiPadobviously,tome,IpersonallyhaveaniPad,andI,Iloveitcause,itseemsgreatfortakingnotesandeverythinglikethatandit'sveryhandycauseit'smuchlighterthanalaptop.Sothat'skindanice.Buyer:Uh,huh.Seller:Uh,Iwoulddefinitelyrecommendgettingone.Iknowtheyareonthepricierside,butasfarasIknowtheyaremuchbetteriPadthanlikeandroidtabletsthathavebrokenlikewithinaweek.So,Iwoulddefinitelyrecommendit.Asfarasqualitygoes,theyaredefinitelyworththeprice.Plus,whenyoubuyityouautomaticallygetApplecareandeverythinglikethat...Um,it'sgreatformailandeverythinglikethat.Yeah,andasfarasfacebook,anysocialmedialikethat.Andthengamestokilltimebetweenclassiskindanice....Generallyspeaking,sellersappealtoself-interestfrequently:72%ofconversationsincludeatleastonestatementabouttheobjectforsale,forexample.Otherconversationsarelargelytangential,however,andinsomecasesthisisevidentlystrategic.Beforethefollowingconver-sation,forexample,thesellerstatesthatherplanisto\chatwiththemandtrytogetthemtoconnectwithmesotheywillspendmore:"Seller:...Ohyeah,sorry,Idon'tknowaboutit,badreceptioninthebuildingorsomething.Yeah,sothewholethingofitisthat,willingtoseehowmuchyouwould{orhowmuchyouwouldbewillingtospendtobuya{basicallyafuturewhichisa\$500checkforJanuary1st2014.Buyer:Yeah,yeahandIknowalittlebitaboutthetimevalueofmoney,so.Seller:No,Iunderstand,sodoI.Doyouhaveajob?Buyer:Yeah,IworkparttimeforOhiohighschoolathleticassociation.Seller:Oh,reallythat'scool.Whatdoyoudoforthem?Buyer:I'mabasketballrefereeandbaseballumpire.Seller:That'scool.WhatdidyouplayinHighSchool?10 Buyer:Yeah.Idonow.Seller:Andnowwhatyou'redoinglikegoonBuyer:No.IwasIplayedinHighSchoolandtheycertifiedmehereatOhioStateandHoustonconnectionsthatIknewsomerefsandtheytoldmeIcouldjointhesquad.Seller:Soyou'vegottodoitlikehereinColumbusordoyoudoitbackhomelikeduringthesummer?...Herethesellerquicklychangestheconversationtopicfromthedecisionathandtowardspersonaldetailsinthehopesofestablishingabetterconnectionwiththebuyer.Thoughlesscommon,manyconversationsincludesuchappealstoother-regard.Sellersstateexantetheywouldattempttobuildrapportin19%ofpairings,andbuyersreportexposttheybelievethattobetheirseller'sstrategyin16%ofpairings.Theirspecicstrategiesvaryconsiderably,from\getthecustomeronmyside,"to\buildarapport,"andto\getonbuyer'sgoodside,"andsoon.Someevenmention,anduse,specicother-regardingtacticsfromtheliterature:forexample,\remembertheperson'sname!Peoplelovewhenothersremembertheirnames."5ResultsWeconducthypothesistestsnon-parametricallywherepossibleusingclusteredrank-sumtests.Whereweconditiononadditionalvariables,weestimateregressionsoftheformWTP1bsp=+WTP0bsp+ p+bsp(6)wherebindexesbuyers,sindexessellers,gindexesproducts,andp0bsp(p1bsp)isameasureofwillingnesstopayforaproductbefore(after)conversations.Productxedeectsaccountforanysystematicdierencesbetweenproducts.145.1DoesCommunicationMatter?Werstcheckwhethercommunicationaectsbehavioratall.Sellershaveamaterialcon ictofinterest,sothatinprinciplebuyersmightdisregardeverythingtheysayas\cheaptalk."Communicationdoesinfactaltervaluations,however.Figure1showsthemeansofabsolutechangesinvaluationswithcommissionforthecommunicationandnocommunicationconditions.Themeanabsolutechangeinvaluationundercommunicationis$49(forall15sessions)signicantlyandsubstantiallylargerthantheanalogous$24withoutcommunication(clusteredrank-sump0:01).Ontheextensivemargin,56%ofbuyerschangetheirvaluationinthecommunicationgroupversus31%inthenocommunicationarm.Communicationthusaectsthemajorityofvaluations,someofthemsubstantially.Wenexttestwhethersellers'payosarehigherinexpectationwhentheyhavetheoppor-tunitytocommunicate.Wendthattheyare:themeanchangeinbuyers'valuationwithcommissionsis$14undercommunication(inthefull15-sessionsample),signicantlydierentfromthemeanchangewithoutcommunicationof$3(clusteredrank-sump=0:01)(Figure 14Wedonotincludeordereects,asthesehavenoimpactonestimatesoforonpredictivepowerasmeasuredbyAdjustedR-squared.11 Figure1:Absolutechangesinvaluationdrivenbycommunication Plotstheabsolutechangeintotalpersuasion,self-interest,andother-regard(perEquation4)forboththenocommunicationandcommunicationconditions.Verticalbarsdepictstandarderrorsallowingforclusteringbybuyerandbyseller.Meanofabsolutevaluationchangeinwillingnesstopays,byexperimentalcondition.2,left-mostbars).Incontrast,thenocommunicationtreatmentproducessignicantabsolutechangesbutnosystematicincreaseordecreaseinmeanvaluations.15Onepotentialconcernisthatvaluationchangescouldre ectbuyerconfusionorinattention,generatinginconsistentanswersbeforeandafterthephonecall.Thiswouldnotexplainthefactsthatvaluationsincreaseonaverageanddierentiallysointhecommunicationarmrelativetothenocommunicationarm,however.Wealsoredoouranalysisremovingsubjectswho(i)self-reportconfusion,(ii)werecodedaspotentiallybeingconfusedbycodersreadingthetranscripts,or(iii)providenon-monotonicvaluations(e.g.willingtopay$100,butnotwillingtopay$80),andndthisdoesnotqualitativelyaltertheresult(availableonrequest).5.2WhichPathwaysMatter?Thedatasuggestthatbothself-interestandother-regardmatter.ThesecondandthirdsetsofbargraphsinFigure1illustratethis,plottingthemeansofabsolutechangesforthetwocomponentsofpersuasion.Themiddlebarsshowthemeanabsolutechangeinself-interestis$60comparedto$24withoutcommunication(clusteredrank-sump0:01),andasintheright-mostbars,themeanabsolutechangeinother-regardis$50comparedto$26innocommunication(clusteredrank-sump0:01).Toquantifymorepreciselyhowmuchself-interestandother-regardcontributetovariationinpersuasion,weaskwhatproportionofthe 15Itisalsoclearintherawdata(Figure5)thatcommunicationisadriverofadditionalchangesinvaluation.12 Persuasion Self-Interest Other-Regard 0 10 20 30 40 50 60 70 Communication No Communication Figure2:Communicationincreasesvaluations;changesinself-interestaremeanpositive Plotsthemeanchangeintotalpersuasion,self-interest,andother-regard(perEquation4)forboththenocommunicationandcommunicationconditions.Verticalbarsdepictstandarderrorsallowingforclusteringbybuyerandbyseller.overallvariationiseliminatedbyconditioningoneachcomponent.16Conditioningonself-interestreducesvariancemorethanconditioningonother-regard:16%ofthevariancecanbeexplainedbyother-regardalone,and29%canbeexplainedbyself-interestalone.Whilechangesinother-regardaresubstantial,theyalsotendtocanceleachotheroutonaverage.Figure2showstheaveragegainsbypathwayusingthefullsample(tofacilitatecomparisonswiththeNoCommunicationarm).Whiletheaveragechangeinperceivedself-interestis$18andsignicantlydierentfromzero,theaveragechangeinother-regardis-$4andinsignicant.Thesameistrueintherstelevensessions(withoutNoCommunication):changesinself-interestnettoameanchangeof$20,signicantlydierentfromzero(p0:01),comprising115%oftotalpersuasion,whilechangesinother-regardroughlycanceleachotherout,nettingameanchangeof-$2.64,insignicantlydierentthanzero(p=0:7).Oneinterpretationisthatother-regardisexogenous\whitenoise",whileself-interestisthemainpathwayforintentionalpersuasion.Forexample,buyersmightndsomesellersmorelikeablethanexpectedandotherslessso,withoutsellersmakinganyintentionaleorttomanipulatebuyers'regardforthem.Recallthatsellersreportedtheirstrategiesbeforetheconversationsbegan.Toexamine 16Specically,wecalculatetheproportion1E[V(WTPcjWTPnc)] V(WTPc)(7)oftotalvarianceattributablesolelytoself-interest,andtheproportion1E[V(WTPcj(WTPcWTPnc))] V(WTPc)(8)attributablesolelytoother-regard.13 Persuasion Self-Interest Other-Regard -10 0 10 20 Communication No Communication Table1:Sellerswhotargetother-regardoutperform PersuasionSelf-InterestOther-regardPersuasionSelf-InterestOther-regard TargetsOther-Regard23.771.525.5229.1210.8524.34(14.20)(14.42)(10.22)(21.93)(22.18)(10.85)TargetsSelf-Interest0.738.402.827.0012.2614.31(18.48)(20.37)(15.16)(14.30)(22.57)(17.46)HasAnyStrategy11.2110.906.142.923.087.77(19.42)(20.49)(11.91)(19.22)(25.75)(17.06) SellerFENoNoNoYesYesYesObservations220220220220220220R20.740.610.070.790.700.33AdjR20.730.600.040.720.590.07 Eachcolumnisaseparateregression.Outcomesarechangesintotalpersuasion,self-interest,andother-regardfrombeforetoaftercommunication.\Targetsother-regard",\Targetsself-interest",and\Anystrategy"areindicatorsequaltooneifatleasttwoofthreecodersagreedthattheseller'sself-reportedstrategyinvolvesmanipulatingother-regard,manipulatingself-interest,orin uencingthebuyerinanywayrespectively.Standarderrorsclusteredbybuyerandbysellerinparenthesis.Statisticalsignicanceisdenotedas:p0:10,p0:05,p0:01.howsuchstrategiesfared,wehadthreeundergraduateresearchassistantsindependentlycodeeachstrategyasreferringtoother-regard,referringtoself-interest,and/orcontaininganyplan.Coderssawonlystrategystatementsandsubjectidentiercodes.Wecodeastrategyashavingafeature(e.g.targetingother-regard)ifatleasttwoofthreecoderssaiditdid.Overallwecode19%ofstatementsastargetingother-regard,30%targetingself-interest,and55%containinganyplan.Thiscodingappearsconsistentwithourindependentcodingoftheconversationtranscriptsthemselves:forexample,sellerswhomwecodeastargetingother-regardusethebuyer'srstnamedirectionallymoreoften(p=0:27),havemorepersonalconversations(p=0:02),andhavemoretangentialconversationnotfocusedontheproduct(p=0:03).17Sellerswhouseother-regardingstrategiesperformsignicantlybetterthantheirpeers(Table1,Column1).Moreover,theestimatedeectsizerepresentsasizeable137%ofaveragepersuasioninourconversations.18Thedatadonotprovidestrongevidencethesegainsaretheresultofonepathwayoranother.Sellerswhousestrategiesthattargetother-regardaremodestly,butinsignicantlymoresuccessfulinraisingbuyers'perceivedself-interest(Column2)andother-regard(Column3).Ofcourse,itcouldbethatsellerswhoreportother-regardingstrategiesareintrinsicallymorelikeableandthatwemeasuretheeectsoftheirpersonalityasopposedtotheirchoices.Weexaminethisbyaddingsellerxedeectstothemodel(Columns4-6).Theestimatesarequalitativelysimilarandifanythinglargerthanthepooledestimates;however,theestimateonthenetgainsarelargebutinsignicant(Column4).Interestingly,thesamesellershave 17p-valuesfromOLSregressionscontrollingforproductxedeects,clusteringatsellerlevel.18Therelationshipbetweentargetingother-regardandpersuasionisnotsignicantbyconventionalthresholdswhenincludingdatafromthelastfour,mixedtreatmentsessions(seesubsectionsec:mixedsessions).14 alargereectonother-regardwhenusingstrategiesthattargetother-regardthanwhentheydonot(Column6).5.3HowDoesPersuasionVarywithContext?Wenextexaminehowpersuasionvarieswiththenatureofthegoodbeingsold.Sellersweremoresuccessfulonaverageatpersuadingbuyerstobuytabletcomputers.Figure3breaksdownaveragevaluationchanges,overallandforthetwopathways,forthetabletsandtheabstractgoods.Totalvaluationsincreaseby$25.1forthetabletscomparedto$9.6fortheabstractgoods(rank-sum,clusteredatbuyerandsellerlevel,p=0:03).Thisisconsistentwiththeviewthatinformationasymmetrycreatesgreaterscopeforpersuasion.Yetthevaluationincreaseforabstractgoodsisstatisticallysignicant:sellersbenetfrompersuasionevenwhen\selling"lotteryticketsandpost-datedchecks.19Thisresultishardtoreconcilewithmodelsofpersuasioninwhichbuyerpreferencesareself-interestedandxed.Perhapspersuasionisfeasibleherebecausesellerscanexploitother-regard?Consistentwiththisview,Figure3showsthatother-regardplaysalargerroleforabstractgoodsthanfortablets.Themeanchangeinself-interestissignicantlyhigherfortablets($34versus$6,rank-sumclusteredatbuyerandsellerlevel,p=0:05).Incontrast,other-regardactuallydecreasesonaveragefortablets,whileincreasinginsignicantlyforab-stractgoods($8.9versus$3:6,rank-sum,clusteredatbuyerandsellerlevel,p=0:22).Broadlyspeaking,theseresultsshowthatwhichpathwaysmatterdependonthegoodbeingsold(ormoregenerally,thedecisionbeingmade).Morespecically,forthegoodswecompre-hensivelydescribetoallsubjects{thelotteryandthecheck{theinformationchannelislesseectivewhileappealstoother-regardaremoreso.Thissuggestsself-interestplaysalargerrolewheninformationislesssymmetric.20Onecaveat,however,isthatwecannotrejectthehypothesisthatthesamepatternsholdinthenocommunicationcondition,duetoimpreciseestimates.Moregenerally,persuasionmaydieracrossgoodsforavarietyofreasons{self-ishpreferencesforabstractgoodsmightbein uencedbyframing,forexample{sothatanydierencesacrossproducttypesshouldbeinterpretedassuggestive.215.4SimiliartyandpersuasionWenextexaminehowpersuasionvarieswithsimilaritybetweenthebuyerandseller.Wemeasuresimilarityintwocomplementaryways.First,weusebuyers'self-reportsoftheirperceivedsimilaritywiththeseller.Thebenetofthismeasureisthatitiscomprehensive;thedrawbackisthatitissubjective,andbuyersmightfeelaneedtoanswerinawaythatrationalizedtheirvaluationresponses.Wethereforecomplementitbyalsotestingdirectlyforeectsofamatchontheobservablecharacteristicsofgenderandnationality.These 19AmoregenericF-testforequalityinmeanpersuasionacrossproductsismarginallyinsignicant(p=0:11).20Onepotentialcaveattothisinterpretationisthatwhilebothplayerswereprovidedwithidentical,com-prehensiveinformationaboutthelotteryandcheck,oneplayermightunderstandthisinformationbetterthantheother,creatingasymmetriesineectiveinformation.However,theseasymmetriesineectiveinformationshouldnotbesystematicallytiltedtowardsbuyersorsellersinthewaythatinformationaboutthetabletswasbydesign.Wethankoneoftherefereesforpointingoutthispossibility.21Wethankoneoftherefereesforsuggestingthispossibility.15 Figure3:Valuationsincreasemoreforconsumerelectronics;drivenbyappealstoself-interest Plotsthemeanchangeinself-interestandother-regardingvaluationcomponentsasdenedinEquation4aswellastotalpersuasion,fortabletsandabstractgoods.Verticalbarsdepictstandarderrorsallowingforclusteringbybuyerandbyseller.areincompletemeasuresofsimilarity,butobjective.Overallbothmeasuresappearsensible.Perceivedsimilarityisdistributedfairlyuniformlyfrom1(\notatall")to11(\verymuchso"),withnoscoreoccurringlessthan5%ofthetimeoranyscoreoccurringmorethan14%ofthetime.Buyerswhowerematchedongenderornationalitywiththeirsellerreportedsignicantlyhighersimilarityscores(1.0pointsand1.5points,respectively;rank-sump0:01inbothcases).Wendthatsimilaritystronglypredictspersuasion.Figure4illustratesthisforself-reportedsimilarity,whileTable2reportsregressioncoecientsforallthreemeasures.InColumn1weestimatea$5.7increaseinpersuasionforeveryonepointincreaseinsimilarityscore(p0:01).Forperspective,thisimpliesthataonestandarddeviationincreaseinsimi-laritywouldproducean$17.7increaseinpersuasion,morethandoublingaveragepersuasion($17.4).Thisresultisstrongerwhenwefocusthematchongenderspecically:whileneitherbuyerorsellergenderpredictspersuasionorchangesinself-interestorother-regard,theinter-actionoftheirgenderdoes.Thefourthcolumnshowsthatpairsthatarematchedongenderhaveincreasesof$22morethanthegender-mismatchedpairs(p=0:04).22Fornationalitythispatternisweaker;persuasionishigherby$45.4forsame-nationalitypairsbutthisisnotstatisticallysignicant.23Theeectsofsimilaritycouldplausiblyworkeitherthroughself-interestorthroughother- 22Weestimatesimilareectsofmale-maleandfemale-femalematches.23Whiletwonon-UScitizensarenotnecessarilywelldescribedas\matched,"anecdotallytheprobabilityishighthatbothareEastAsian,andinanycasesuchpairingsaretoorare(4%)tobedrivingourresults.16 Persuasion Self-Interest Other-Regard -10 0 10 20 30 40 Tablet Abstract Figure4:Perceivedsimilaritypredictsin uence Plotsthemeanchangeinpersuasionbyhowsimilarthesellerwastothebuyeraccordingtothebuyer,where11ismostsimilar.Verticalbarsdepictstandarderrorsallowingforclusteringbybuyerandbyseller.regard.Similaritycouldbegetaltruism,buildtrust,orgivethesellergreaterinsightintothebuyer'spreferences,forexample.Wendevidenceforbothchannels,butsomewhatstrongerevidenceforself-interest.Aonepointincreaseinreportedsimilarityincreaseschangesinself-interestby$4.2(p=0:06)andother-regardby$3.0(p=0:04).Withgender-matchedpairsthisgapisstarker:gender-matchedpairsseealarge$29.4increaseinvaluationsviaself-interest(p=0:02)butonlyaninsignicant$3.3increaseviaother-regard(p=0:6).Neitherchannelissignicantonitsownfornationalitymatch,however.2425Overall,wendstrongevidencethatsimilarityfacilitatespersuasion,anovelformofho-mophily,andsuggestiveevidencethatthisworksprimarilythroughsellers'abilitytoconvincebuyersaboutwhatitistheirownself-interest.5.5HowdoSelf-InterestandOther-RegardInteract?Inprinciple,theself-interestandother-regardpathwayscouldinteractwitheachotherinseveralways.Mechanicalinteractionscouldarisebecausesellersfaceconstraintsintermsoftime,buyerattention,etc.andmustchoosebetweenin uencingself-interestorother-regard.Causalinteractionscouldariseifagivensellerbehavioraectsbothself-interestandother- 24Wealsondthatsimilaritycontinuestopredictpersuasionevenafterweconditiononameasureoftrustcodedfromthetranscripts(notreported).25Concurrentwithourwork,Heetal.(2015)alsondthatgender-matchedpairscooperatemoreinsocialdilemmagamesthatinvolvecommunication.17 0 14 1 17 2 17 3 14 4 19 5 27 6 26 7 35 8 34 9 22 10 13 11 15 -60 -40 -20 0 20 40 60 80 100 Similarity Score Sample Size Table2:Buyer-sellersimilaritypredictspersuasionthroughbothpathways PersuasionSelf-InterestOther-RegardPersuasionSelf-InterestOther-RegardPersuasionSelf-InterestOther-Regard SimilarityScore5.714.223.00(1.53)(2.23)(1.41)GenderMatch22.0129.363.28(10.33)(12.41)(6.68)BuyerFemale1.53-5.16-3.46(12.46)(13.56)(8.79)SellerFemale-1.26-2.54-14.90(10.29)(13.02)(7.68)CitizenshipMatch5.390.4413.45(7.12)(20.42)(17.09)BuyerNon-American15.691.1319.79(9.46)(16.33)(19.97)SellerNon-American-11.61-30.1616.91(7.42)(21.93)(15.16)N210210210220220220216216216Adj.R20.740.600.060.730.610.050.730.610.05 Eachcolumnisaseparateregression.Eachregressionincludesproductxedeects.\Similarityscore"isthebuyer'sperceivedsimilaritywiththesellerona1-11scale.Standarderrorsclusteredbybuyerandbysellerinparenthesis.Statisticalsignicanceisdenotedas:p0:10,p0:05,p0:01. 18 Figure5:Communicationincreasesmorevaluations;netin uencefrequentlydrivenbychangesinself-interest Plotsthechangesinself-interestandother-regardingvaluationcomponentsasdenedinEquation4foreverybuyer-sellerpairing,forboththecommunication(leftgure)andnocommunication(rightgure)conditions.Eachvectorrepresentsonebuyer-sellerpairing,withinitialvaluationsnormalizedtotheorigin;vectorspointtovaluationchanges,withjitteradded.Thex-axisdenoteschangesinperceivedself-interest,andthey-axischangesinother-regard.Vectorsthatlienortheastofthey=-xline(lightbluevectors)denotepositivenetin uence;vectorslyingeastofthey-axisdenotepositivechangesinperceivedself-interest;vectorslyingnorthofthex-axisdenotepositivechangesinother-regard.regard.Theseeectscouldbecomplementaryorosetting.Forexample,spendingtimedescribingpositivefeaturesofaproductcouldcomeacrossashelpfulandengenderfeelingsofreciprocityinthebuyer,oraspushyandengenderresentment.Establishingthingsthatthepartieshaveincommoncouldaectbothother-regardandthecredibilityofadviceaboutthebuyer'sself-interest.Tovisualizethisinteraction,Figure5plotsthefulldistributionofvaluationchangesasvectorsin(self-interest,other-regard)space,andseparatelyforthecommunication(left)andnocommunicationcondition(right).Vectorslyingeast(west)ofx=0indicateanincrease(decrease)inself-interest,andvectorslyingnorth(south)ofy=0indicateanincrease(decrease)inother-regard.Ifthepathwaysofpersuasionwerepositivelycorrelated,mostvectorswouldpointtothenortheastorsouthwest.Instead,mostofthevectorslieinthenorthwestandsoutheastquadrants,indicatinganegativecorrelation.Indeed,whenweseenon-zerochangesinbothpathways,againinonepathwayiscoupledwithalossalongtheotherin79%ofcases.Overall,changesinself-interestandother-regardaresignicantlynegativelycorrelated(=0:64).Oneconcernisthatthiscorrelationcouldbebiasedbyerrorinourmeasurementofval-19 -200 0 200 400 -400 -200 0 200 400 Communication N = 264 Net Influence Positive: 36% Negative: 21% Zero: 44% -200 0 200 400 No Communication N = 68 Net Influence Positive: 13% Negative: 18% Zero: 69% D Perceived Self-Interest D Other-Regard uations,asthechangeinself-interestentersnegativelyintoourcalculationofthechangeinother-regard.Toremoveanyunsystematicmeasurementerror,weprojectchangesinself-interestandother-regardontothefullsetofconversationfeaturesrecommendedbyotherresearchers,26aswellasproductxedeectsandexaminethecorrelationbetweenthettedvalues.Thislimitstheanalysistothecomponentofpersuasionnotorthogonaltothesere-gressors,butalsoeliminatesanyspuriousnegativecorrelationduetoclassicalmeasurementerror.27Wendthatpredictedchangesinself-interestandother-regardareinfactevenmorenegativelycorrelated,=:70,andthisholdsforallfourproductsindividually(iPad=0:79;Surface=0:79;Check=0:68;Lottery=0:52).Thissuggestsasharptrade-obetweenthetwopathways.Thispatterndoesnotappeartobedrivenbymechanicalsubstitutionbetweentacticsinshortconversations.Noconversationshittheten-minutetimeconstraint,andresultsaresimilarregardlessofconversationlength,eveninthelongestconversations(inwhichonemightexpectthemechanicalconstraintstobelessbinding).Ifanythingweseeasharpertrade-oinlongerconversations:byquartileofconversationlength,thecorrelationsare=0:54,=0:76,=0:73,and=0:70respectively.Itthusseemsunlikelythatsellerssimplyrunoutoftime,andmorelikelythattheirbehaviorshaveosettingeects.Onequalicationtothispatternisthat,aswesawinTable1,sellerswhointentionallychosestrategiesthattargetedother-regardpaidonlyasmall(andstatisticallyinsignicant)penaltyintermsofself-interestfordoingso.Inotherwords,whileitistrueonaveragethatgainsalongonepathwaytendtocomeattheexpenseoftheother,thereareevidentlystrategiesusedbysomeofthemoresuccessfulsellersthatlargelymitigatethistradeo.Understandingthispositivedeviancecouldbeaninterestingdirectionforfuturework.5.6TrustAsdiscussedabove,ourdecompositionofvaluationchangesintoself-interestedandother-regardingcomponentsispotentiallyconsistentwithmanydierentunderlyingmodelsofpref-erences(anddistributionsofthesepreferencesinthesubjectpool).Whiletestingexhaustivelybetweentheseisbeyondthescopeofthispaper,wedoexamineinmoredepththeroleofonefactorspecically:trust.Trustisplausiblyrelatedtobothchannelsofpersuasion,asbuyersmaybemoreinclinedtobelieveclaimsmadebysellerstheytrust,andmoreinclinedtotrustsellerstheylike.Weexaminetheroleoftrustusingdatafromourpost-experimentanalysisofthecontentofconversations.Specically,weaskedcodersafterreadingeachtranscript\Onascaleof1to5,howmuchdoyoutrustthisseller?"28Thisgivesusameasureofhowtrustworthythesellerwasperceivedtobebyasubjectdrawnfromthesamepoolasthebuyer,lettingusexaminehowtheextenttowhichsellerswereperceivedastrustworthycorrelateswithpersuasionthrougheitherchannel. 26Afterremovingfeaturesrecommendedbymorethanoneresearcher,weareleftwith31uniquefeatures.Wedonotmakeanyfurtherselectionamongfeatures.27Formalderivationavailableonrequest.28Twocodersreadeachtranscript.Weusetheaverageoftheirscoresasourmeasureintheanalysis.20 Table3:Sellerswhotargetother-regardoutperform Dependentvariable: PersuasionSelf-InterestOther-Regard Trustseller15.5013.082.42(7.48)(7.91)(5.17) Observations212212212R20.740.640.03AdjustedR20.740.630.002 Eachcolumnisaseparateregression.\Trustseller"istheaverageoftwothirdparties'ratingsofhowtrustworthythesellerwasintheconversation(1-5,withhigherbeingmoretrustworthy).Standarderrorsclusteredbybuyerandbysellerinparenthesis.Statisticalsignicanceisdenotedas:p0:10,p0:05,p0:01.Perceivedtrustworthinessvariesmeaningfullyfromconversationtoconversation,withmostsellersrateda2(23%),3(52%),or4(21%)outofve.Moreover,somesellersweresystem-aticallyperceivedasmoretrustworthythanothers.Sellerxedeectsexplain14%ofthevariationinperceivedtrustworthiness,andthisrelationshipisstronglysignicant(p0:01,F-testofsellerxedeectsinOLSregressionpredictingperceivedtrustworthinessincludingproductdummies,clusteringatbuyerandsellerlevel).Trustworthinessalsostronglypredictschangesinvaluation,andthiseectworksentirelythroughchangesinperceivedself-interest.Column1ofTable3reportsresultsfromaregressionoftheforminEquation6thatincludessellertrustworthinessasapredictor.Sellersperceivedasonestandarddeviationmoretrustworthyseeanincreaseinpost-conversationvaluationof$11,ora79%increaserelativetothe$14meanincreaseinvaluationafterconversation.InColumns2and3weexaminetherelationshipbetweentrustworthinessandself-interestandother-regard,respectively.Sellersperceivedasonestandarddeviationmoretrustworthyseeasignicantincreaseinself-interestof$9.4,butasmallandinsignicantchangeinother-regard.Theseresultssuggestthatsellertrustworthinessisimportantforbuyers'willingnesstoaccepttheirstatementsaboutwhatisintheirownself-interest,butnotfortheirfeelingsofaltruismorreciprocitytowardssellers.Consistentwiththisidea,wealsondthatsellersarenotperceivedasmoretrustworthywhentheyintentionallyplantotargetother-regard(TableA-4,Column1),andthattheestimatedrelationshipsbetweenthesestrategiesandpersuasionoutcomesinTable1donotchangesubstantiallyifwecontrolfortrustworthiness(TableA-4,Columns2-3).Overall,thepatternswendsuggestthattrustandtrustworthinessmatterfortheself-interestchannelbutnotsubstantiallyforother-regard.5.7ResearcherpredictionsSevenoftheeightresearcherswhomweinvitedagreedtoparticipateinour\predictionchallenge."AppendixBliststhe35questionstheyproposed,whicharguablyprovideasnap-shotviewintohowsocialscientistscurrentlyunderstandpersuasion.Twofeaturesappearnoteworthy.21 First,congruenceacrossresearchersislow.Tothehumaneyetheyappeartoproposeverydierentmeasures,withonlythreeitemsappearingmorethanonce:thenumberofwordsintheconversation(3),thenumberoftimesthesellerusesthebuyer'sname(2),andhowlikeabletheselleris(2).Ofcourse,variablescouldappeardierentbutstillcapturesimilarstatisticalinformation.Wethereforecalculatehowwellagivenvariableproposedbyoneresearcherispredictedbythevariablesproposedbyasecondresearcher.Thismetricwouldbe100%ifallresearcherswereinfactproposingthesameunderlyingmodel,and2%inexpectationiftheirvariableswerecompletelyunrelated(duetochancecorrelation).Inourdatatheaveragevalueofthisstatisticis26%,implyingatmostmodestagreement.Thissuggeststousarelativelyyoungeldofresearchthathasyettoconvergeonaconsensusviewofhowpersuasionworks.Second,researcherscollectivelyexpectbothself-interestandother-regardtoplayaroleinpersuasion.Ofthethirty-veitems,twentyonewerecodedbyatleasttwoofthreeresearchassistantsascapturingappealstoperceivedself-interest,andfteenitemsascapturingappealstoother-regard.29Researcher3,forexample,focusedonself-interest,proposingaseriesofquestionsaboutproductattributes(e.g.\Howmanystatementsoftheform`youcanuseitforX'doesthesellermake?")andtheirinteractionwithameasureoftrustworthiness.Severalotherresearchers,meanwhile,askedquestionsaboutthelikeability(ordis-likeability)oftheseller,oraboutfeatureslikelanguagestylematchingwhicharethoughttopromoteliking.5.7.1Whatresearcheritemspredictpersuasion?Wecanalsousetheseresearcher-providedvariablestolearnwhatfacilitatespersuasioninoursetting.TablesA-5,A-6,andA-7showhoweachoftheseitemscorrelatewithourthreeoutcomes.Eachrowrepresentsaseparateregression:WerunthebaselineregressionfromEquation6whileincludingtheresearchervariablelistedasoneadditionalcovariate.Here,wehighlightafewoftheitemsthathelpexplainpersuasioninoursetting.WefocusonitemsthatincreaseAdjustedR-squaredfromourbaselineregressionby0.005ormoreandareatleastmarginallysignicant.Forpredictingpersuasionoverall,themostexplanatorypowerwasprovidedbyapurelysubjectivevariable:Doyouthinkthisincreasedthebuyer'svaluation?Thoughpersuasionishard,andhardtopredict,apparently,wedorecognizeitwhenweseeit(AdjR2=0.009,p0:01).Anothermeaningfulpredictorofpersuasionwas\Doyoutrusttheseller?"(AdjR2=0.007,p=0:04).Expost,theimportanceoftrustmayseemobvious;however,onlyoneresearcherincludeditintheirveitems.Perhapsthemostsurprisingpredictorofpersuasionwasthenumberoftimesthesellerusestheword\I"(AdjR2=0.007,p=0:09).Thiscouldpotentiallycapturethesellerspeakingoftheirownexperiencewiththeproduct,theirownperspective,orperhapsgettingpersonalwiththebuyer.Lookingatwhichresearcheritemspredictchangesinperceivedself-interest,itseemsthattalkingpositivelyaboutthegoodbeingsoldiseective.Thismaybetrueindependentofhowtruethestatementsareperceivedtobe.First,thenumberandstrengthoftheseller's 29Thecodingwasnotmutuallyexclusive,andtwoitemswerecodedasintendingtocapturebothpathways.Noitemwascodedasintendingtocaptureneitherchannel.22 arguments,asperceivedbyourcoders,predictstronggainsinproductvaluations(AdjR2=0.009,p0:01),consistentwithPettyandCacioppo(1986).Second,howmanyliesorexaggerationsthesellertoldisalsosignicantandpredictive(AdjR2=0.012,p=0:02).Strikingly,everyadditionallieorexaggerationourcoder'sperceivedcoincidedwitha$38.8increaseinproductvaluationonaverage.Wealsoonlyfoundtwovariablesthathelppredictvarianceinchangesinother-regard.Therstvariableisagainthesubjectivecodingof\Doyouthinkthisincreasedthebuyer'svaluation?"Interestingly,thisvariableisnotatallpredictiveofgainsalongperceivedself-interest.Thepersuasionthatourcoderscanseeinthetranscriptsishappeningthroughincreasedaltruismonly(AdjR2=0.018,p=0:03).Second,thenumberoftimesthesellersaid\no"alsocorrelatespositivelywithgainsinother-regard(AdjR2=0.017,p=0:05).Thisitemisnotjustoperatingthroughitscorrelationwithlengthofconversationasaseparateexpertpredictionmeasuresthatdirectly.Perhapsthistacticcouldbuildcredibilitywiththebuyerbyadmittingthelimitsoftheproducts(akintotheintuitioninDziuda(2011)).Finally,apopulariteminbooks(andseminars)isthenumberoftimesthesellerusesthebuyer'srstname.Thoughsellersattemptthisstrategy(meannumberoftimesnameismentioned=0.5),anditismarginallysignicant;itdoesnotincreaseourabilitytopredictvariancesubstantially(AdjR2=0.002).6ConclusionWedeneandapplyanexperimentalframeworkforstudyingwhetherandwhenpersua-sionworksthroughappealstoself-interestortoother-regard.Wendthatbothmechanismsmattertoadegree,withsubstantialvariationdependingoncontextandindividuals.The\average"sellerinourexperimentbenetsfromcommunicationprimarilybecausecommuni-cationconvincesbuyersthatitisintheirownbestinteresttopaymoreforobjects.Atthesametime,aminorityofsellersmakeastrategicchoicetofocusongettingbuyerstolikethem,andthesesellersoutperformtheirpeerssignicantlybothatcultivatingother-regardandatpersuasionoverall.Sellersgainsaredrivenmorebybuyerself-interestwhentheydiscusstan-gibleproductsaboutwhichsellerswerebetter-informed,butmorebyother-regardinsettingsofsymmetricinformation.Finally,sellersfaceatrade-o,withgainsononepathwaygenerallyosetinpartbylossesontheother.Futureworkcouldadapttheexperimentalframeworkweuseheretostudyanumberofadditionalquestions.Onenaturalextensionwouldbetocomparethemethodsandperfor-manceofamateurtoexpertpersuaders,suchasprofessionalsalespeople.Asecondwouldbetoexaminethereturnstodierentformsofcommunication{forexample,whetherinteractingface-to-faceyieldslargebenetstothepersuader.Athirdwouldbeexperimentallymanipu-latewhatispermissibleinaconversation{forexample,prohibitingsubjectsfromdiscussingtheproduct(ormorebroadly,thedecision)athandmightforcethemtorelyuponappealstoother-regard.Finally,itwouldbeinterestingtocontrastpersuasionaboutpurelyprivategoodssuchasthoseweuseheretopersuasionaboutpublicgoods.Persuadinganeighbortovoteinfavorofanenvironmentalreform,forexample,hastheinterestingfeaturesthat(i)23 itmayneverbeintheneighborsnarrowself-interesttodoso,and(ii)votingforthereformlikelybenetsthepersuaderifandonlyifitalsobenetstheneighbor,creatingadirectlinkbetweenself-interestandother-regard.Finally,futureworkcouldexaminewhatdrivesthetrade-oweseebetweenpathways,perhapsbyexperimentallyvaryingthestrategiessellersusethroughcoaching.24 ReferencesAlsan,Marcella,OwenGarrick,andGrantCGraziani,\Doesdiversitymatterforhealth?ExperimentalevidencefromOakland,"2018.Andreoni,JamesandA.AbigailPayne,\CharitableGiving,"inAlanAuerbach,RajChetty,MartinFeldstein,andEmmanuelSaez,eds.,HandbookofPublicEconomics,Volume5,2013. andJustinM.Rao,\Thepowerofasking:Howcommunicationaectsselshness,empathy,andaltruism,"JournalofPublicEconomics,August2011,95(7-8),513{520.Antioch,Gerry,\Persuasionisnow30PercentofUSGDP,"EconomicRoundup:TheTreasuryoftheAustralianGovernment,June2013,(1).Azrieli,Yaron,ChristopherP.Chambers,andPaulJ.Healy,\IncentivesinExperi-ments:ATheoreticalAnalysis,"JournalofPoliticalEconomy,forthcoming.Becker,GarySandKevinMMurphy,\ASimpleTheoryofAdvertisingasaGoodorBad,"TheQuarterlyJournalofEconomics,November1993,108(4),941{64.Bellemare,CharlesandBruceShearer,\Giftgivingandworkerproductivity:Evidencefromarm-levelexperiment,"GamesandEconomicBehavior,2009,67,233{244.Bettinger,EricPandBridgetTerryLong,\Dofacultyserveasrolemodels?Theimpactofinstructorgenderonfemalestudents,"AmericanEconomicReview,2005,95(2),152{157.Bohnet,IrisandBrunoSFrey,\TheSoundofSilenceinPrisoner'sDilemmaandDictatorGames,"JournalofEconomicBehaviorandOrganization,011999,38(1),43{57.Bolton,GaryEandAxelOckenfels,\ERC:Atheoryofequity,reciprocity,andcompe-tition,"AmericanEconomicReview,2000,pp.166{193.Brandts,Jordi,GaryCharness,andMatthewEllman,\Let'sTalk:HowCommunica-tionAectsContractDesign,"JournaloftheEuropeanEconomicAssociation,forthcoming.Bremzen,Andrei,ElenaKhokhlova,AntonSuvorov,andJeroenVandeVen,\Badnews:anexperimentalstudyontheinformationaleectsofrewards,"ReviewofEconomicsandStatistics,2015,97(1),55{70.Buchan,NancyR,EricJJohnson,andRachelTACroson,\Let'sgetpersonal:Aninternationalexaminationofthein uenceofcommunication,cultureandsocialdistanceonotherregardingpreferences,"JournalofEconomicBehaviorandOrganization,2006,60,373{398.Card,David,AlexandreMas,EnricoMoretti,andEmmanuelSaez,\InequalityatWork:TheEectofPeerSalariesonJobSatisfaction,"AmericanEconomicReview,2012,102(6),2981{3003.Carnegie,Dale,HowtoWinFriendsandIn uencePeople,SimonandSchuster,1936.Carrell,ScottE,MarianneEPage,andJamesEWest,\Sexandscience:Howprofessorgenderperpetuatesthegendergap,"TheQuarterlyJournalofEconomics,2010,125(3),1101{1144.25 Charness,GaryandMatthewRabin,\UnderstandingSocialPreferencesWithSimpleTests,"TheQuarterlyJournalofEconomics,August2002,117(3),817{869. andUriGneezy,\What'sinaname?AnonymityandSocialDistanceinDictatorandUltimatumGames,"JournalofEconomicBehaviorandOrganization,2008,68,29{35.Che,Yeon-Koo,WouterDessein,andNavinKartik,\PanderingtoPersuade,"Amer-icanEconomicReview,February2013,103(1),47{79.Cialdini,Robert,In uence:thePsychologyofPersuasion,WilliamMorrowandCompany,Inc.,1993.Cooper,DavidandJohnKagel,\Other-RegardingPreferences:ASelectiveSurveyofExperimentalResults,"inAlvinRothandJohnKagel,eds.,HandbookofExperimentalEconomics,Volume2,2009.Crawford,VincentPandJoelSobel,\StrategicInformationTransmission,"Economet-rica,November1982,50(6),1431{51.Dawes,RobynM,JeanneMcTavish,andHarrietShaklee,\Behavior,communication,andassumptionsaboutotherpeople'sbehaviorinacommonsdilemmasituation.,"JournalofPersonalityandSocialPsychology,1977,35(1),1.Dee,ThomasS.,\Teachers,Race,andStudentAchievementinaRandomizedExperiment,"ReviewofEconomicsandStatistics,2004,86(1),195{210.DellaVigna,StefanoandMatthewGentzkow,\Persuasion:EmpiricalEvidence,"Na-tionalBureauofEconomicResearchWorkingPaperNo.15298,2009.Dickhaut,JohnW,KevinAMcCabe,andArijitMukherji,\Anexperimentalstudyofstrategicinformationtransmission,"EconomicTheory,1995,6(3),389{403.Dziuda,Wioletta,\Strategicargumentation,"JournalofEconomicTheory,July2011,146(4),1362{1397.Fehr,ErnstandKlausMSchmidt,\Atheoryoffairness,competition,andcooperation,"QuarterlyjournalofEconomics,1999,pp.817{868.Greiner,Ben,WernerGuth,andRo'iZultan,\Socialcommunicationanddiscrimina-tion:avideoexperiment,"ExperimentalEconomics,2012,15(3),398{417.He,Simin,TheoOerman,andJeroenvandeVen,\TheSourcesoftheCommunicationGap,"SSRNWorkingPaperNo.2598441,2015.Houser,DanielandErteXiao,\ClassicationofNaturalLanguageMessagesUsingaCoordinationGame,"ExperimentalEconomics,2011,14(1),1{14.Kamenica,EmirandMatthewGentzkow,\BayesianPersuasion,"AmericanEconomicReview,October2011,101(6),2590{2615.Krueger,AlanBandAlexandreMas,\Strikes,scabs,andtreadseparations:Laborstrifeandtheproductionofdefectivebridgestone/restonetires,"JournalofPoliticalEconomy,2004,112(2),253{289.Landry,CraigE.,AndreasLange,JohnA.List,MichaelK.Price,andNicholasG.Rupp,\TowardanUnderstandingoftheEconomicsofCharity:EvidencefromaFieldExperiment,"QuarterlyJournalofEconomics,2006,121(2),747{782.26 Loewenstein,George,DaylianM.Cain,andSunitaSah,\TheLimitsofTransparency:PitfallsandPotentialofDisclosingCon ictsofInterest,"AmericanEconomicReview,May2011,101(3),423{28.Loewenstein,GeorgeF,LeighThompson,andMaxHBazerman,\Socialutilityanddecisionmakingininterpersonalcontexts.,"JournalofPersonalityandSocialpsychology,1989,57(3),426.Malmendier,UlrikeandKlausSchmidt,\YouOweMe,"NBERWorkingPaper18543,2012.McCloskey,DonaldandArjoKlamer,\OneQuarterofGDPisPersuasion,"AmericanEconomicReview,PapersandProceedings,May1995,85(2),191{195.Mobius,MarkusMandTanyaSRosenblat,\Whybeautymatters,"AmericanEconomicReview,2006,96(1),222{235.Mullainathan,Sendhil,JoshuaSchwartzstein,andAndreiShleifer,\CoarseThinkingandPersuasion,"TheQuarterlyJournalofEconomics,052008,123(2),577{619.Muralidharan,KarthikandKetkiSheth,\Bridgingeducationgendergapsindevelopingcountries:Theroleoffemaleteachers,"JournalofHumanResources,2016,51(2),269{297.Murphy,KevinMandAndreiShleifer,\PersuasioninPolitics,"AmericanEconomicReview,Papers&Proceedings,May2004,94(2),435{439.Olszewski,Wojciech,\Informalcommunication,"JournalofEconomicTheory,August2004,117(2),180{200.Pennebaker,JamesW.,TheSecretLifeofPronouns:WhatourWordsSayAboutUs,BloomsburyPress,2011.Petty,RichardE.andJohnT.Cacioppo,CommunicationandPersuasion:centralandperipheralroutestoattitudechange,Springer,1986.Rabin,Matthew,\Incorporatingfairnessintogametheoryandeconomics,"TheAmericaneconomicreview,1993,pp.1281{1302.Roth,AlvinE,\BargainingExperiments,"inAlvinRothandJohnKagel,eds.,HandbookofExperimentalEconomics,1995.Yamamori,Tetsuo,KazuhikoKato,ToshijiKawagoe,andAkihikoMatsui,\Voicemattersinadictatorgame,"ExperimentalEconomics,December2008,11(4),336{343.27 AAppendix,foronlinepublicationTableA-1:ElicitingWillingnesstoPay,FlatWagePanelA:FlatWage Pleaseselectwhichalternativeyouwouldprefer IwouldpreferthenewiPadIwouldprefer$500 Iwouldprefer$20andthenewiPadIwouldprefer$500 ...... Iwouldprefer$480andthenewiPadIwouldprefer$500 PanelB:WithCommission Pleaseselectwhichalternativeyouwouldprefer IwouldpreferthenewiPadformeIwouldprefer$500formeand$500fortheseller.and$0fortheseller. Iwouldprefer$20andthenewiPadformeIwouldprefer$500formeand$500fortheseller.and$0fortheseller. ...... Iwouldprefer$480andthenewiPadformeIwouldprefer$500formeand$500fortheseller.and$0fortheseller. Eachrowisaseparatequestion.Buyersansweredallquestionsbeforeproceedingwiththeexperiment.AllquestionsinPanelAweredisplayedonthesamepage,withPanelBonthesubsequentpage.28 FigureA-1:Initialvaluationsshowmoderatealtruism Plotsbuyers'meaninitialvaluations,indollars,foreachofthefourproductsstudied.Valuationshavebeendecomposedintothemeanwillingnesstopayabsentanycommissionfortheseller(\self-regarding")andthemeanincrementalwillingnesstopaywhencommissionsareadded(\other-regarding").29 Lottery Surface iPad Future 0 100 200 300 400 500 Lottery Surface iPad Future 0 100 200 300 400 500 Self-Interest Other-Regard FigureA-2:Distributionofparticipantxedeects Plotseachstudyparticipants'absolutemeanchangeinpersuasionacrossallfourconversations,sortedfromsmallesttolargestbeforeabsolutevaluetaken,foreachseller(gray)andeachbuyer(blue),fromallsessionswithoutthenocommunicationcondition(1-11).30 0.0 0.2 0.4 0.6 0.8 1.0 0 50 100 150 Percentile of distribution Absolute value Buyer FEs Seller FEs TableA-2:ParticipationbySession SessionBuyersSellersConversationsNoCommunication 14416026624035520045520057728064416075520085520095520010552001144160121062020137382014624201554128Totals837026468 TableA-3:Participantcharacteristics BuyersSellers VariableMeanSDMeanSD Age21.241.8621.211.80Gender=Male0.550.500.560.50WeeklyExpenditure67.3442.7486.9582.93Voted0.560.500.700.46OwnsaniPad0.000.000.800.40UsedaniPad0.880.320.940.24OwnsaWindowsTablet0.000.000.240.43UsedaWindowsTablet0.180.390.560.50OwnsaniPhone0.540.520.680.47OwnsaniPod0.760.460.800.40OwnsanAndroidTablet0.060.290.140.34 Reportsmeansandstandardvariationsofage(toprow)andindicatorvariablesforwhetheragivenstatementistrue(allotherrows).Bysampleconstruction,nobuyersownedaniPadoraWindowstablet.31 TableA-4:Sellerstrategiesandperceivedtrustworthiness:correlationsandmediations Dependentvariable: TrustsellerPersuasion Trustseller15.16(7.17)TargetsOther-Regard0.0122.5122.56(0.13)(14.39)(14.28)TargetsSelf-Interest0.040.831.49(0.15)(19.47)(19.94)HasAnyStrategy0.119.758.05(0.16)(19.18)(18.65) Observations212212212R20.010.740.75AdjustedR20.020.730.74 Column1reportsanOLSregressionpredictingperceivedsellertrustworthinessusingseller'spre-conversationreportedstrategies.Columns2and3regress(OLS)totalpersuasiononseller'sreportedstrategies(Column2)aswellasperceivedsellertrustworthiness(Column3).Standarderrorsclusteredbybuyerandbysellerinparenthesis.Statisticalsignicanceisdenotedas:p0:10,p0:05,p0:01.32 TableA-5:Experts'Items'EectsonPersuasion Row#VariableCodingMeanBetaSEAdj.R2Adj.R2y 1Baseline0.860.050.7302Increasedvaluation1{5;5=alot2.2618.236.380.7390.0093Trustseller1{5;5=alot2.9415.507.480.7370.0074Likeseller1{5;5=alot2.948.406.170.7330.0035Buyeraman0=No;1=Yes0.490.3212.840.729-0.0016Buyerman,Sellerwoman0=No;1=Yes0.2415.2512.210.7310.0017Num.wordsspokenExactvalue517.890.010.010.7320.0028Num.\I"s,sellerExactvalue10.981.090.640.7370.0079FirstnameuseExactvalue0.503.293.070.7320.00210Num.\no"sExactvalue0.766.216.640.7340.00411InterruptionsExactvalue8.760.020.280.7310.00112Statements,experienceExactvalue0.188.129.980.7320.00213Statements,others'experienceExactvalue0.0624.6715.470.7330.00314Statements,useExactvalue0.700.574.120.7310.00115Statements,notincentivesExactvalue0.0190.4325.520.7340.0041615(12+13+14)Exactvalue0.0232.6213.370.7340.00417Convaboutobject0=prosandcons;1=tangential0.181.5611.880.7310.00118Downsidesofobject0=No,1=Yes0.2115.6617.870.7330.00319%ofconvbybuyerExactvalue0.3311.6150.100.7310.00120QuestionsaboutbuyingExactvalue2.660.062.260.7340.00421Name,you,oryour,sellerExactvalue1.900.974.470.7310.00122Strengthofarguments1{5;5=manystrongarguments2.289.494.880.7360.00623Warm,likeable1{5;5=verylikeable2.7711.676.240.7340.00424Competent,expert1{5;5=verycompetentorexpert2.707.025.450.7320.00225Lie,exaggerateExactvalue0.206.1811.910.7310.00126Panhandlerplea0=No;1=Yes0.0642.7346.950.7360.00627Detailsofproducts0=No;1=Yes0.564.4916.330.684-0.04628Actuarialvalues0=No;1=Yes029Sellerexplainincentives0=No;1=Yes0.518.3710.150.7320.00230LanguagestylematchingExactvalue6.070.552.840.7310.001pluspositiveemotionwords31Initialvaluation(guess)Exactvalue170.070.020.030.7320.00232Howpersonal1{5;5=verypersonal2.151.664.980.7310.00133Initialinterest1{5;5=veryinterested2.2310.307.650.7330.003 Note:p0.1;p0.05;p0.01yDenotesdierenceinadjustedR2comparedtobaselineregressioninrstrow. 33 TableA-6:Experts'Items'EectsonPerceivedSelf-Interest Row#VariableCodingMeanBetaSEAdj.R2Adj.R2y 1Baseline0.780.060.612Increasedvaluation1{5;5=alot2.260.976.420.602-0.0043Trustseller1{5;5=alot2.9414.158.170.6070.0014Likeseller1{5;5=alot2.9411.236.960.605-0.0015Buyeraman0=No;1=Yes0.497.5913.990.604-0.0026Buyerman,Sellerwoman0=No;1=Yes0.2416.1112.980.6060.0007Num.wordsspokenExactvalue517.890.030.020.6080.0028Num."I"s,sellerExactvalue10.980.980.700.6070.0019FirstnameuseExactvalue0.500.246.570.602-0.00410Num."no"sExactvalue0.765.675.600.605-0.00111InterruptionsExactvalue8.760.650.450.605-0.00112Statements,experienceExactvalue0.189.1111.660.603-0.00313Statements,others'experienceExactvalue0.0625.7215.300.604-0.00214Statements,useExactvalue0.707.144.850.605-0.00115Statements,notincentivesExactvalue0.0169.4325.800.604-0.0021615(12+13+14)Exactvalue0.0228.319.010.604-0.00217Convaboutobject0=prosandcons;1=tangential0.185.2117.000.602-0.00418Downsidesofobject0=No,1=Yes0.214.3118.220.602-0.00419%ofconvbybuyerExactvalue0.3366.6552.440.6070.00120QuestionsaboutbuyingExactvalue2.660.892.190.599-0.00721Name,you,oryour,sellerExactvalue1.900.718.080.602-0.00422Strengthofarguments1{5;5=manystrongarguments2.2814.934.890.6150.00923Warm,likeable1{5;5=verylikeable2.7713.519.520.6070.00124Competent,expert1{5;5=verycompetentorexpert2.704.047.080.603-0.00325Lie,exaggerateExactvalue0.2038.8416.400.6180.01226Panhandlerplea0=No;1=Yes0.0636.4336.670.6060.00027Detailsofproducts0=No;1=Yes0.560.7122.250.460-0.14628Actuarialvalues0=No;1=Yes029Sellerexplainincentives0=No;1=Yes0.5116.3213.650.605-0.00130LanguagestylematchingExactvalue6.073.793.110.604-0.002pluspositiveemotionwords31Initialvaluation(guess)Exactvalue170.070.040.040.604-0.00232Howpersonal1{5;5=verypersonal2.151.837.680.602-0.00433Initialinterest1{5;5=veryinterested2.2311.289.590.604-0.002 Note:p0.1;p0.05;p0.01yDenotesdierenceinadjustedR2comparedtobaselineregressioninrstrow. 34 TableA-7:Experts'Items'EectsonOther-Regard Row#VariableCodingMeanBetaSEAdj.R2Adj.R2y 1Baseline0.230.080.052Increasedvaluation1{5;5=alot2.2614.596.600.0690.0183Trustseller1{5;5=alot2.942.375.110.047-0.0044Likeseller1{5;5=alot2.943.215.650.048-0.0035Buyeraman0=No;1=Yes0.492.408.410.047-0.0046Buyerman,Sellerwoman0=No;1=Yes0.2410.1511.930.050-0.0017Num.wordsspokenExactvalue517.890.010.010.048-0.0038Num.\I"s,sellerExactvalue10.980.190.450.047-0.0049FirstnameuseExactvalue0.506.053.430.0530.00210Num.\no"sExactvalue0.767.693.940.0680.01711InterruptionsExactvalue8.760.410.430.0520.00112Statements,experienceExactvalue0.184.238.880.047-0.00413Statements,others'experienceExactvalue0.062.607.380.047-0.00414Statements,useExactvalue0.704.994.280.0520.00115Statements,notincentivesExactvalue0.0129.7021.440.048-0.0031615(12+13+14)Exactvalue0.029.459.020.048-0.00317Convaboutobject0=prosandcons;1=tangential0.186.4311.240.048-0.00318Downsidesofobject0=No,1=Yes0.2115.0712.430.0540.00319%ofconvbybuyerExactvalue0.330.0524.890.047-0.00420QuestionsaboutbuyingExactvalue2.660.221.870.047-0.00421Name,you,oryour,sellerExactvalue1.900.225.320.047-0.00422Strengthofarguments1{5;5=manystrongarguments2.282.784.380.048-0.00323Warm,likeable1{5;5=verylikeable2.774.885.240.049-0.00224Competent,expert1{5;5=verycompetentorexpert2.704.275.590.049-0.00225Lie,exaggerateExactvalue0.2011.4915.390.0520.00126Panhandlerplea0=No;1=Yes0.066.9028.410.047-0.00427Detailsofproducts0=No;1=Yes0.5615.4016.550.005-0.04628Actuarialvalues0=No;1=Yes029Sellerexplainincentives0=No;1=Yes0.5110.958.950.0520.00130LanguagestylematchingExactvalue6.073.971.970.0560.005pluspositiveemotionwords31Initialvaluation(guess)Exactvalue170.070.040.040.0560.00532Howpersonal1{5;5=verypersonal2.151.384.320.047-0.00433Initialinterest1{5;5=veryinterested2.235.296.840.048-0.003 Note:p0.1;p0.05;p0.01yDenotesdierenceinadjustedR2comparedtobaselineregressioninrstrow. 35 A.1AnalysisincludingmixedsessiondataInthissectionwereplicatetheanalysisfromthemaintextbutalsoincludingdatafromtheCommunicationarminsessions12-15.Asareminder,thesesessionswereconductedafterourprimarydatacollectionnishedinresponsetofeedbackonthoseresults.Weconsideredthesesessionsasexpostandseparate;hadweplannedtoincludedatafromtheirinourmainanalysiswewouldhaveinterspersedCommunication-onlysessions(aswasdoneintherst11sessions)tocontrolforsubjectpooleects,seasonality,andsoonwhilesimultaneouslymeasuringpotentialspilloversfromrunningtheCommunicationarmandNo-Communicationarmwithin-subjectinmixedsessions.Forthesakeoftransparency,however,wealsopresenthereanalysispoolingdatafromtheCommunicationarmdatafromallsessions.Generallyspeakingtheresultsarequalitativelysimilar,thoughtherelationshipswedescribearegenerallymoderatelysmaller{perhapsduetospilloversacrossarms,perhapsduetosubjectpooleects,orperhapsbychance.TableA-8:AveragesdisplayedinFigure3includingmixedsessions Dependentvariable: PersuasionSelf-InterestOther-Regard(1)(2)(3) Tablet12.2726.0613.79(9.67)(9.56)(9.18)Constant7.735.002.73(6.21)(7.23)(7.85) Observations264264264R20.010.020.01AdjustedR20.000.010.00 Note:p0.1;p0.05;p0.0136 TableA-9:Table1includingmixedsessions Dependentvariable: PersuasionSelf-InterestOther-regardPersuasionSelf-InterestOther-regard(1)(2)(3)(4)(5)(6) TargetsOther-Regard17.300.721.7319.4511.1516.27(13.05)(13.19)(8.75)(21.35)(23.16)(10.76)TargetsSelf-Interest5.637.231.057.746.238.18(17.88)(18.71)(13.40)(14.41)(21.12)(16.57)HasAnyStrategy6.108.834.834.871.804.12(19.04)(19.11)(11.11)(19.10)(23.53)(15.53) SellerFENoNoNoYesYesYesObservations264264264264264264R20.730.610.070.790.700.32AdjustedR20.720.600.050.710.600.06 Note:p0.1;p0.05;p0.0137 TableA-10:Table2includingmixedsessions Dependentvariable: PersuasionSelf-InterestOther-regardPersuasionSelf-InterestOther-regardPersuasionSelf-InterestOther-regard(1)(2)(3)(4)(5)(6)(7)(8)(9) SimilarityScore5.113.262.47(1.38)(2.04)(1.28)GenderMatch21.7425.222.59(8.99)(10.84)(5.90)BuyerFemale0.273.964.68(10.97)(12.23)(7.67)SellerFemale1.142.6111.23(8.61)(11.28)(6.760)CitizenshipMatch7.091.5213.91(7.63)(18.82)(15.20)BuyerNon-American12.125.2019.14(8.51)(15.25)(17.99)SellerNon-American5.3416.9815.58(7.57)(20.42)(13.60)Constant51.8377.120.9667.6590.4737.5966.2497.040.77(19.45)(25.68)(12.32)(31.13)(35.42)(17.72)(19.55)(32.64)(17.58) Observations253253253264264264260260260R20.740.610.080.730.620.080.730.620.08AdjustedR20.730.600.060.730.610.050.730.610.06 Note:p0.1;p0.05;p0.01 38 BResearcherPredictionsforCorrelatesofPersuasionHere,wepresentthepredictionsfromthepersuasionresearcherswhoparticipatedinourPredictionChallenge.Thegoalofthecontestwastochoosevefeaturesoftheconversationsthatwouldbestpredictvarianceinnetin uence.Wepresenttheminnoparticularorder.Researcher11.Onascaleof1to5,howmuchdoyouthinkthisconversationincreasedthebuyer'svaluationfortheproduct?(1=notatall,5=alot)2.Onascaleof1to5,howmuchdoyoutrustthisseller?(1=notatall,5=alot)3.Onascaleof1to5,howmuchdoyoulikethisseller?(1=notatall,5=alot)4.Isthebuyeraman?5.Isthebuyeramanandthesellerawoman?Researcher21.Thelength(inwords)oftheconversation.2.Thenumberoftimesthatthesellersays\I"(rstpersonsingular).3.Thenumberoftimesthesellercallsthebuyerbyrstname.4.Thetotalnumberoftimes\no"isused.5.Thenumberof\overlaps/interruptions"(butpeopletalkingatthesametime).Researcher31.Howmanystatementsoftheform\I'veuseditandreallylikeX"doessellermake?2.Howmanystatementsoftheform\someoneIknowhasuseditandreallylikedX"doesthesellermake?3.Howmanystatementsoftheform\youcanuseitforX"doesthesellermake?4.Howmanystatementsoftheform\I'mnottellingyouthisbecauseofmyincentives"doesthesellermake?5.Theinteraction(4)x[(1)+(2)+(3)]Researcher41.Wastheconversationaboutprosandconsoftheobject/thedecisionofbuying,orwasitwastangential?2.Didthesellerappearimpartialbygivingalsoconsofbuyingtheobject?3.Theduration(inwords)oftheconversation(thoughnotcertaininwhichdirectionthiswillgo)4.Thefraction(ofwords)oftheconversationduringthebuyerspoke.5.HowmanyquestionsRELATEDtothebuyingdecision(notquestionsaboutjobsorweather)didthesellerask?39 Researcher51.Lookforthemerenumberofwordsusedforbuyerandseller2.Howmanytimesdoesthesellersaythebuyer'sname,theword\you",ortheword\your"?3.Onascalefrom1to5,howmanyandhowstrongdoyouthinkthesellersargumentsare?(1=noargumentsoronlyveryweakarguments,5=manystrongarguments)4.Onascalefrom1to5,howwarmorlikeabledoyoundtheparticipants(1=notlikeableatall,5=verylikeable)5.Onascalefrom1to5,howcompetentorexpertdoyoundtheparticipants(1=notatallcompetentorexpert,,5=verycompetentorexpert)Researcher61.Howmanytimesdidthesellerlieaboutorexaggeratetheproductattributes?2.Doesthesellermakinga`panhandler'kindofpleaforhelp?3.Doesthesellerdescribedetailsabouttheproducts(forthetwoproducts)?4.Doesthesellerprovidetheactuarialvalueoftheitems(forthecheckandlottery)?5.Doesthesellerexplaintheincentiveschemetothebuyer?Researcher71.Wordcount(themorewordsintheconversation,butespeciallythebuyer)2.Languagestylematchingpluspositiveemotionwords(LSM{thisusesLIWCandcal-culatestherelativeuseofthefollowingfunctionwords:personalpronouns,impersonalpronouns,articles,prepositions,auxiliaryverbs,negations,adverbs,conjunctions).Inotherwords,highLSMandhighpositiveemotionwordswillresultinthehighestamount.3.Whatdidthebuyersaytheywerewillingtopayatthebeginningoftheconversation{orif(s)hedidnotsay,whatdoyouthinks(he)waswillingtopayatthebeginningoftheconversation?4.Onascalefrom1to5,howpersonaldoesthebuyerget?(Personalisdenedbythebuyersayingsomethingabouthisbackground,emotions,orpersonalconcerns.Also,morelikelytosayhis/hername.)(1=notatallpersonal,5=verypersonal)5.Oneascalefrom1to5,howwouldyouratethebuyer'sinitialdesireorinterestintheproduct?Itemsaddedbyauthors1.Dothebuyerandsellermakeasidedeal(agreedtopaymentsoutsideoftheexperiment)and/oragreetomeetafterwards?2.Onascalefrom1to3,howwelldoesthebuyerappeartounderstandtherulesofthegame?(1=notatall,2=somewhat,3=verywell)3.Onascalefrom1to3,howwelldoesthesellerappeartounderstandtherulesofthegame?(1=notatall,2=somewhat,3=verywell)40 CAdditionalconversationsamplesWeprovidehereafewfurtherexamplestoillustratetherangeoftheconversationsbetweenbuyersandsellers.Thefulldatasetoftranscribedconversations(withidentiersremoved)isavailableonrequest.Someofouramateursellerscapitulatequickly:Seller:Yeah,ifIwere--yeahifIwereinyourposition,Iwouldprobablythinklikeonlygetlike,200,\$300profit.Thatwouldbeaverynicedealalready.Buyer:Yeah,definitely.Seller:[longpause]Well...[longpause]Idon'tknow.I'mnotagoodsalesperson.Buyer:It'sokay.Seller:Wecantalkaboutwhatever.Buyer:Sowhatyearareyou?...Othersseemfarmorecomfortable,tothepointofmonopolizingtheconversation:Seller:AnduhIknowjudgingbyyourlackofgamblingexperienceandnervousnesswiththat,Ithinkyouwouldprobablyagreewithmethere.Butatthispoint,Iwould---Iwouldsayyoumightaswellgoforit.Iwoulddoit.Causeyoumadefifteenbuckshere.So,you'repositive.You'renotgoingtoloseanymoney.I'dsay-I'dsayit'snotasaferisk,butfifty-fiftyshotonwinninga\$1000inadditiontothebigonefivethatyou'repullinginrightnowasyour[inaudible].Idon't--Idon't--Idon'tthinkIcouldturnthatdown.IthinkIwouldhavetogoforit.Buyer:Alright.Well,I'llkeepthatinmind....Somesellerswerenotabovelying:inthisexample,theseller(successfully)misleadsthebuyeraboutakeydetailoftheexperimentaldesign:Seller:SoIwasthinkingwecouldfindapricetomakesurethatwebothwin...Buyer:Mhmm,okay.Seller:So,what'syour,basically,yourbottomline,like,howmuchwouldyoubewillingtopayforthis?Buyer:Well,howmuchdoesit,Iguess,retailfor?Seller:Itretailsfor500.Buyer:Okay,500dollars?Yeah.I'mthinking,probablythen,mybottomwouldbe[longpause]ImeanIwouldbewillingtopay250dollarsforit.Seller:Well,yeah,but,[inaudible]Buyer:Ok,well,thehighestI'dbewillingtogo,wouldprobablybeabout325.Seller:Ok,well,seethere'sanotherthing.Ifwesellitfor,wecan'tmakeanythingifwesellitforlessthan360....41