BIASINPOLITICALBELIEFREVISION
1
Thehearttrumpsthehead:Desirabilitybiasinpoliticalbeliefrevision
RunningHead:BIASINPOLITICALBELIEFREVISION
BenM.Tappina,1,*,LeslievanderLeerb,2,&RyanT.McKaya,3
a
ARCCentreofExcellenceinCognitionanditsDisorders,DepartmentofPsychology,Royal
Holloway,UniversityofLondon,Egham,SurreyTW200EX,UK
Institutionaltelephonenumber:+44(0)1784434455
b
Regent’sSchoolofPsychotherapyandPsychology,Regent’sUniversityLondon,London
NW14NS,UK
Institutionaltelephonenumber:+44(0)2074877700
Email:[email protected];[email protected];[email protected]
*Correspondingauthor
ARTICLETYPE:Article
Wordcount:3823
InpressatJournalofExperimentalPsychology:General
©2017,AmericanPsychologicalAssociation.Thispaperisnotthecopyofrecordandmay
notexactlyreplicatethefinal,authoritativeversionofthearticle.Pleasedonotcopyor
citewithoutauthorspermission.Thefinalarticlewillbeavailable,uponpublication,viaits
DOI:10.1037/xge0000298
Acknowledgements:ThisresearchwassupportedbyagrantfromtheEconomicandSocial
ResearchCouncilawardedtothefirstauthor,referencenumberES/J500148/1.
Dataavailability:ForaccesstotherawdatasupportingthisarticlepleasecontactResearch
DataManagementatRoyalHolloway,UniversityofLondon:[email protected].
BIASINPOLITICALBELIEFREVISION
2
Abstract
Understandinghowindividualsrevisetheirpoliticalbeliefshasimportantimplicationsfor
society.Inapre-registeredstudy(N=900)weexperimentallyseparatedthepredictionsof
twoleadingtheoriesofhumanbeliefrevision—desirabilitybiasandconfirmationbias—in
thecontextofthe2016USpresidentialelection.Participantsindicatedwhotheydesiredto
win,andwhotheybelievedwouldwin,theelection.Followingconfrontationwithevidence
thatwaseitherconsistentorinconsistentwiththeirdesiresorbeliefs,theyagainindicated
whotheybelievedwouldwin.Weobservedarobustdesirabilitybias—individualsupdated
theirbeliefsmoreiftheevidencewasconsistent(versusinconsistent)withtheirdesired
outcome.Thisbiaswasindependentofwhethertheevidencewasconsistentorinconsistent
withtheirpriorbeliefs.Incontrast,wefindlimitedevidenceofanindependentconfirmation
biasinbeliefupdating.Theseresultshaveimplicationsfortherelevantpsychological
theoriesandforpoliticalbeliefrevisioninpractice.
Keywords:Confirmationbias;desirabilitybias;motivatedcognition;beliefupdating;politics
BIASINPOLITICALBELIEFREVISION
3
Introduction
Peopleareroutinelyexposedtoabewilderingarrayofinformationrelevanttotheir
politicalbeliefs.Whetherandhowtheyincorporatethisinformationhasprofound
consequencesforsociety.Thebeliefthatvaccineshaveharmfulsideeffects(Moritz,2011),
orthatclimatechangeisahoax(Lewandowskyetal.,2013),canreducepeople’sintentions
tovaccinate(Gangarosaetal.,1998;Jolley&Douglas,2014a;Horneetal.,2015),orto
minimizetheircarbonfootprint(Douglas&Sutton,2015;Jolley&Douglas,2014b).Even
simpleinfographicsdisplayedduringlivetelevisedelectiondebatescanmeaningfullyshape
beliefsaboutdebateoutcome,potentiallyinfluencingthevotingintentionsofmillionsof
viewers(Davisetal.,2011).Aclearunderstandingofhowpeopleincorporateinformation
intotheirpoliticalbeliefsisthusofconsiderablepracticalimportance.
Twoprominenttheoriesoffersimilaryetdistinctpredictionsregardingwhenand
howpeopleincorporatenewinformationintotheirbeliefs.Onetheorycontendsthat
individualsassigngreaterweighttoinformationthatisdesirableversusundesirable—i.e.,a
desirabilitybias.Thisbiasisreportedtounderlieanasymmetrywherebypeopleupdate
theirpriorbeliefstoincorporatenewanddesirableinformationmorethannewbut
undesirableinformation(Sharot&Garrett,2016).Theothertheory,confirmationbias,
contendsthatpeoplepreferentiallysearchfor,evaluateandincorporatenewinformation
thatconfirmstheirpriorbeliefs(Nickerson,1998).Thisbiasisreportedtounderliean
asymmetrywherebypeopleupdatetheirpriorbeliefstoincorporatenewandconfirming
informationmorethannewbutdisconfirminginformation—eveniftheyreceiveabalanced
setofbothtypesofinformation(Lordetal.,1979;Taber&Lodge,2006;Taberetal.,2009).
BIASINPOLITICALBELIEFREVISION
4
Unfortunately,thepredictionsofdesirabilitybiasandconfirmationbiasareoften
conflated.Inthedomainofself-belief,thetendencyforpeopletobelievedesirablethings
aboutthemselvesandtheirfutures(Sedikides&Strube,1997;Weinstein,1980)meansthat
newinformationistypicallyeitherconfirminganddesirable,ordisconfirmingand
undesirable(Eil&Rao,2011).Inthedomainofpoliticalbelief,rigorousseparationof
desirableandconfirminginformationissimilarlydifficult.Ofthefewexperimentsthatare
appropriatelydesignedtodisentanglethem,groupidentityistakenasaproxyforthe
desirabilityofinformation—thatis,whethertheinformationisconsistent(i.e.,desirable)or
inconsistent(undesirable)withthepositionofanindividual’sculturalgroup—andbelief
updatingisnotthetargetoutcomemeasure(e.g.,seeKahan,2016a;2016b).
Hereweexperimentallyseparateddesirabilitybiasandconfirmationbiasinpolitical
beliefupdating.Todoso,wecapitalizedonthepoliticalcontextpriortothe2016US
presidentialelection.Toillustratetheadvantageofthiscontext,considerthatmany
supportersofcandidateDonaldTrumpmayhavebelievedHillaryClintonwouldwinthe
election—owingtoherestablishmentsupport(Green&Kapur,2016)or,more
conspiratorially,ariggedballot(Graham,2016).Insuchcircumstances,newinformation
mayhavebeensimultaneouslyconfirmingbutundesirable—forinstance,pollsindicatinga
Clintonwin—ordisconfirmingbutdesirable—pollsindicatingaTrumpwin:causing
desirabilitybiasandconfirmationbiastoyielddivergentpredictionsforbeliefupdating.
Weexploitedtheprofusionofclosepollingresults1tocrediblysuggesttoindividuals
thateitherDonaldTrumporHillaryClintonwouldbecomethenextpresident,and
1
th
Atthetimeofstudy(datacollectioncommenced26 September2016);see
http://www.realclearpolitics.com/epolls/2016/president/us/general_election_trump_vs_clinton-5491.html
BIASINPOLITICALBELIEFREVISION
5
measuredhowindividualswithcongruent(i.e.,samecandidate)desire-beliefprofiles,and
incongruent(differentcandidate)desire-beliefprofilesupdatedtheirbeliefsfollowing
receiptofthisinformation.Wethusindependentlymanipulatedwhetherinformationwas
consistentorinconsistentwith(a)whoindividualsdesiredtowintheelection,or(b)who
theybelievedwouldwintheelection.
Methods
Participants
Wecollecteddatafrom900participantsonlineviaAmazon’sMechanicalTurk(59%
female;Mage=37.89SD=12.91).ParticipantswereUSresidentsasdeterminedbyIP
address(IPaddresseslocatedoutsideoftheUSwereblockedpriortothestartofthe
experiment).Werequired779participantstoattaingreaterthan80%power(α=.05)to
detectasmalleffectofpartialetasquared(ηp²)=.01inourprimaryanalysesofcovariance.
Weaddedapproximately15%tothisnumbertoguardagainstpowerlossduetoplanned
dataexclusions.Followingthesedataexclusions,weretained811participantsforanalyses.
Thestudyhypotheses,design,datacollection,andanalysisplanwerepre-registered(see
https://aspredicted.org/idxgj.pdf).
Procedure&design
Atthebeginningofthesurvey,participantscompletedabriefscreening
questionnairedesignedtodeterminewhothey(a)desiredtowin,and(b)believedwould
winthe2016USpresidentialelection.Responsesto(a)wereprovidedinanominalchoice
format:Participantsselected“DonaldTrump”,“HillaryClinton”,or“neither”.Responsesto
(b)wereprovidedonabipolarslidingscalefrom0-100with“HillaryClinton”(0)atoneend,
BIASINPOLITICALBELIEFREVISION
6
and“DonaldTrump”(100)attheother(thenumericalvalueswerehiddenfrom
participants).Participantswereinstructedthatthemoreconfidenttheywerethata
candidatewouldwin,theclosertheyshouldslidethepointertothatcandidate’sname.
Thosewhorespondedwithscoresgreaterthan50werecategorizedasbelievingTrump
wouldwin,andscoreslessthan50asbelievingClintonwouldwin.Participantsselecting
“neither”for(a),orexactly50for(b),weredirectedtoanend-of-surveymessageandwere
unabletocontinue.Thisyieldedtwoquasi-experimentalgroups;thosewhosedesire-believe
candidateswerecongruent,andthosewhosedesire-believecandidateswereincongruent.
Webalancedtheseconditionassignmentstoobtainapproximately450ineachquasiexperimentalcondition(finalconditionsamplesafterdataexclusions:congruentdesirebelief:n=406[desireTrump/believeTrump:n=127,desireClinton/believeClinton:n=279];incongruent
desire-belief:n=405[desireClinton/believeTrump:n=91,desireTrump/believeClinton:n=314])2.
Participantsinbothconditionsthencompletedafillertask(the16-itemBalanced
InventoryofDesirableResponding;Hartetal.,2015)beforebeingrandomlypresentedwith
evidenceeitherconsistent,orinconsistent,withwhotheybelievedwouldwintheelection.
Specifically,participantsreadashortpassageaboutnationwidepollingresults,which
emphasizedeitherthatHillaryClintonorDonaldTrumpwaslikelytowintheupcoming
election.Participantswerealsopresentedwithabargraphfigureillustratingsuchan
outcome(studymaterialsareavailableintheMaterialsSupplement).Evidencepresentation
wasbalancedwithineachspecificcandidatethatparticipantsinitiallybelievedwouldwin
theelection.Forexample,ofthoseparticipantswhoinitiallybelievedTrumpwouldwin,half
2
Thesubstantialvarianceinconditionsizespercandidatereflectsthefactthatapproximatelythree-quarters
ofoursampleinitiallybelievedClintonwouldwin(seeFigure1intheresultssection)—explainingthesmaller
numberofindividualsinthe“believeTrump”condition(s).
BIASINPOLITICALBELIEFREVISION
7
receivedthepollingmanipulationsuggestingClintonwouldwin,andhalfreceivedthe
pollingmanipulationsuggestingTrumpwouldwin(likewiseforthosewhoinitiallybelieved
Clintonwouldwin).Thus,collapsingoverspecificcandidates,thisyieldedfourbetweensubjectsconditionsina2x2design:Evidenceconsistentorinconsistentwithwhothe
participantinitiallybelievedwouldwin(Confirmation:ConfirmatoryorDisconfirmatory)and
consistentorinconsistentwithwhotheydesiredtowin(Desirability:Desirableor
Undesirable).Followingtheevidencepresentationparticipantsrespondedtoseveralfiller
questionsaboutpollingdata—e.g.,“Towhatextenthaveyoubeenfollowingthepolling
datafortheupcomingUSpresidentialelection?”—beforeagainindicatingwhotheybelieved
wouldwintheelection,onthesamebipolarscaleusedinitially.
Beliefupdating
Wecalculatedhowmuchparticipantsupdatedtheirconfidenceinwhotheybelieved
wouldwintheelectioninthefollowingsteps.First,weconvertedboththeparticipants’
initialconfidence(T1),andtheirsubsequentconfidence(T2),ontoacomparablescale
indicatingtheabsoluteconfidencetheyhadinthecandidatetheyinitiallybelievedwasmost
likelytowin.Thus,forthosewhoinitiallybelievedTrumpwouldwinwesubtracted50from
T1andT2scores,whereasforthosewhoinitiallybelievedClintonwouldwinwesubtracted
T1andT2scoresfrom50.Next,wecomputedtheabsolutedifferencebetweenthesenewly
convertedT1andT2scoresforeachparticipant.Finally,wemultipliedthisdifferenceby
either1(iftheparticipantupdatedtowardsthepresentedevidence)or-1(iftheparticipant
updatedawayfromthepresentedevidence);meaningthathighernumbersrepresented
greaterbeliefupdatingtowardsthepresentedevidence.
Results
BIASINPOLITICALBELIEFREVISION
8
Dataexclusions
Participantswereexcludedfromallanalysesforfulfillingoneormoreofthepre-
registeredcriteria:Failinganattentioncheckembeddedinthefillertask(n=22,2.44%of
sample),answering“yes”toaquestionaskingthemiftheyrespondeddishonestlyor
mistakenlyduringthesurvey(n=48,5.33%),orrecordingabeliefupdatescoreofgreater
thanthemean±3SDintheirrespectivecondition(n=26,2.89%).Weexcluded1(0.11%)
furtherparticipantfortakingthesurveymorethanonce(identifiedviatheiruniqueAmazon
MechanicalTurkID).Followingtheseexclusions,811participantswereretainedfor
analyses.
Descriptives
Figure1displaystheproportionofparticipantsreportingwhothey(a)desiredtowin
and(b)initiallybelievedwouldwintheelection(fortheseresultssplitbygender,agegroup
andethnicity,seefiguresS1-S3intheAnalysisSupplement).
BIASINPOLITICALBELIEFREVISION
Figure1.Percentageofparticipantsreportingwhichcandidatethey(a)desiredtowinand
(b)initiallybelievedwouldwinthe2016USpresidentialelection.N=811.
Pre-registeredanalyses
Weconductedananalysisofcovariance(ANCOVA)toinvestigatetheeffectof
DesirabilityandConfirmationfactorsonbeliefupdating,adjustingforabsoluteT1
confidencescores3(Figure2displaystheadjustedmeanupdateineachcondition4).There
wasamaineffectofDesirability:F(1,806)=32.81,p<.001,ηp²=.0490%CI[0.02,0.06],
suchthatparticipantsupdatedmoretowardstheevidencewhenitwasconsistent(versus
inconsistent)withthecandidatetheydesiredtowin.Therewasalsoamaineffectof
Confirmation:F(1,806)=76.63,p<.001,ηp²=.09[0.06,0.12],butinthiscaseparticipants
updatedmoretowardstheevidencewhenitwasinconsistent(versusconsistent)withthe
candidatetheyinitiallybelievedwouldwin.Inotherwords,weobservedadisconfirmation
bias.Finally,weobservedasmallinteractionbetweenDesirabilityandConfirmation:F(1,
806)=7.15,p=.008,ηp²=.01[0.00,0.02].Todecomposethisinteraction,weconducted
plannedANCOVAscomparingupdatingineachcondition—whileadjustingforabsoluteT1
confidencescores.
3
4
Thispreventsregressiontothemeanspuriouslyaffectingbeliefupdating.
TherawmeansanddistributionsofupdatescoresarereportedintheAnalysisSupplement.
9
BIASINPOLITICALBELIEFREVISION
10
Figure2.Meanupdatebycondition.Errorbarsandparenthesesdenotestandarderrorofthe
mean.Note:MeansareadjustedforabsoluteT1confidenceandbasedonthe2x2ANCOVA
model.Oneunitofupdatecorrespondstoa1%adjustmentonthebipolarscaleusedto
measurebelief.N=811.
Forthoseparticipantsreceivingdisconfirminginformation,updatingwasgreaterif
thatinformationwasdesirable(versusundesirable):F(1,407)=36.58,p<.001,ηp²=.08
90%CI[0.04,0.13].Thispatternwasthesameforthosereceivingconfirminginformation,
albeitlesspronounced:F(1,398)=20.62,p<.001,ηp²=.05[0.02,0.09].Nextweexamined
thoseparticipantswhoreceivedundesirableinformation—updatingwasgreaterfor
disconfirming(versusconfirming)information:F(1,406)=23.76,p<.001,ηp²=.06[0.02,
0.09].Thisdisconfirmationpatternwasthesame,yetmorepronounced,forthosereceiving
BIASINPOLITICALBELIEFREVISION
11
desirableinformation:F(1,399)=47.72,p<.001,ηp²=.11[0.06,0.16].Finally,directly
comparingtheuniqueeffectofdesirableinformation(disconfirming-desirablecondition)
againsttheuniqueeffectofconfirminginformation(confirming-undesirablecondition)
revealedthatupdatingwasgreaterfortheformer:F(1,402)=75.26,p<.001,ηp²=.16
[0.11,0.21].
Inthefollowingsections,wereportaseriesofexploratoryanalysestoexamine(a)
therobustnessofourresults,and(b)extantdebatesinthefieldofpoliticallymotivated
cognition.
Robustness
Priorexposure.Itislikelythatparticipantshaddifferentamountsofpriorexposure
totheelectionpolls.Examinationofthedistributionofoneofourfillerquestions—“To
whatextenthaveyoubeenfollowingthepollingdatafortheupcomingUSpresidential
election?”—suggestedthiswasthecase(seeFigureS4intheAnalysisSupplement).Itis
possiblethisaffectedourmanipulationandsubsequentresults.WethusrepeatedourpreregisteredANCOVAwiththeadditionofthisvariableasacovariate.However,thepatternof
resultsremainedthesame.
Initialconfidence.Participants’initial(T1)confidencescoreswerenegatively
skewed—inparticular,asubstantialnumberreportedcomplete(orstrong)confidencein
theirinitialbeliefregardingwhichcandidatewouldwin(seeFigure3).Thisconstrainsbelief
updatingforthosereceivingconfirminginformationbecausetheyareunabletoupdate
towardsthenewinformation(i.e.,increasetheirconfidence).Incontrast,thosereceiving
BIASINPOLITICALBELIEFREVISION
12
disconfirminginformationcanupdatetowardsthenewinformation(i.e.,decreasetheir
confidence).Thismayaccountforthedisconfirmationbiasweobserved5.
Figure3.DistributionofabsoluteT1confidenceinbeliefaboutwhichcandidatewouldwin
theelection.Note:Thedashedlinedenotesthemedian.N=811.
Toexplorethispossibility,weselectedasubsetofparticipants(N=370)—excluding
thosewithhighinitialconfidence(absoluteT1confidencescores>25,Nexcluded=441)—and
recomputedthemeanupdateineachcondition(Figure4displaystheresults).Thepattern
ofmeansinthistruncatedsampleindicatedadiminisheddisconfirmationbias,butan
enduringdesirabilitybias.ToconfirmthisstatisticallyweconductedseparateKruskal-Wallis
5
Wearegratefultotwoanonymousreviewersforemphasizingthispoint.
BIASINPOLITICALBELIEFREVISION
13
testsonthedistributionofbeliefupdatingintheConfirmationandDesirabilityconditions,
respectively6.Assuspected,therewasnowonlyatrivialdifferenceinupdatingfor
participantswhoreceiveddisconfirmatory(Median=2.01,IQR=11.69)versusconfirmatory
(Median=1.78,IQR=6.58)information:χ²(1,N=370)=2.01,p=.156.Incontrast,
participantsreceivingdesirableinformationupdatedmore(Median=3.08,IQR=11.61)
thanthosereceivingundesirableinformation(Median=0.71,IQR=6.16):χ²(1,N=370)=
25.84,p<.001.
Figure4.Meanupdatebyconditionfollowingsampletruncation.Errorbarsandparentheses
denotestandarderrorofthemean.Note:Meansareunadjusted.Oneunitofupdate
correspondstoa1%adjustmentonthebipolarscaleusedtomeasurebelief.N=370.
6
ParametricanalyseswereinappropriateasthecellN’sacrossconditionswereunequalfollowingsample
truncation.
BIASINPOLITICALBELIEFREVISION
14
Tosupplementthisanalysis,wealsospecificallyexaminedupdatingamongthose
withweakconfidenceintheirinitialbelief.Thisisworthwhilebecauseparticipantswith
particularlylowconfidencemayhavebeen(a)lessconstrainedbytheupperlimitofthe
confidencescale,or(b)simplymorereceptivetoconfirminginformation,comparedtotheir
higherconfidencecounterparts.Thus,weselectedthoseparticipantswithlowconfidence
(absoluteT1confidencescores≤12.5,Nexcluded=622)andagainrecomputedthemean
updateineachcondition.BecausetheresultantNwassmall(Nlow-confidence=189)and
unevenlydistributedacrossconditions,wesimulatedbeliefupdatingscoresusingthe
parametersfromthelowconfidencesample.Specifically,foreachofthefourconditions,we
drew500scoresfromarandomnormaldistributioncentredontherespectivecondition
mean,andthepooledSD(i.e.,computedacrossthefourconditions)(thesimulationscript
anddataareavailableintheSimulationSupplement).
Thissimulatedsampleconferredgreaterthan99%powertodetectsmalleffects(ηp²
=0.01,a=0.05).ConductinganANOVAonthisdatarevealedamaineffectofDesirability,F
(1,1996)=53.73,p<.001,ηp²=.0390%CI[0.02,0.04],similarinsizeandequivalentin
directiontothatobservedintheprecedingempiricalanalyses.Themaineffectof
Confirmationwastrivialinsize,F(1,1996)=2.06,p=.151,ηp²=.001[0.000,0.005],aswas
theinteractionbetweenthetwofactors,F(1,1996)=1.54,p=.215,ηp²=.001[0.000,
0.004].
IdeologicalAsymmetryHypothesis
Thereisongoingdebateoverwhethermotivatedcognitionismorepronounced
amongindividualsonthepoliticalrightthanthepoliticalleft(Jostetal.,2003;Kahan,
2016b).WethusexploredwhethersupportersofDonaldTrumpdemonstratedgreater
BIASINPOLITICALBELIEFREVISION
15
desirabilitybiasthansupportersofHillaryClinton.WeconductedanANCOVA(adjustingfor
absoluteT1confidenceasbefore)withtwofactors:Desirability,andadummycoded
variabledenotingwhichcandidatetheparticipantdesiredtowin(“Supporter”).Therewasa
smallDesirabilitybySupporterinteraction,F(1,806)=8.58,p=.004,ηp²=.0190%CI[0.00,
0.03].SeparateANCOVAmodelsrevealedastrongerdesirabilitybiasamongsupportersof
DonaldTrump,F(1,438)=34.07,p<.001,ηp²=.07[0.04,0.11],thansupportersofHillary
Clinton,F(1,367)=2.54,p=.112,ηp²=.01[0.00,0.03].
Furtherexploration,however,revealedthisasymmetrywasduetothepreviously
identifiedceilingeffectininitial(T1)confidence.First,alargenumberofparticipants
supportedClintonandalsobelievedshewouldwin(n=279)—whereasfewerthanhalfthis
numbersupportedTrumpwhilealsobelievinghewouldwin(n=127).Second,these
participants(i.e.,thosewithcongruentdesireandpriorbelief)hadstrongnegativeskewin
theirinitialconfidence,withmanybelievingthattheirdesiredcandidatewascertaintowin
(seeFigureS5intheAnalysisSupplement).Takingthesefactstogetherimpliesthat
supportersofClintonweremorenumerousamongthosewhoreceiveddesirable
informationbutwereconstrained(byvirtueoftheirextremeinitialconfidence)inupdating
theirbelieftowardsthisinformation.
Thiswasconfirmedbyexaminingparticipantswho(i)hadacongruentdesire-belief
profile,(ii)receiveddesirableinformation,and(iii)reportedextremeinitialconfidence
(absoluteT1confidence>45).Oftheseparticipants(n=69),67%supportedClinton(n=46)
and33%supportedTrump(n=23).Thisdiscrepancymayhavedisproportionately
suppresseddesirabilitybiasamongClintonsupporters.Indeed,truncatingthesampleto
excludethosewithextremeinitialconfidence(absoluteT1confidence>45,Nexcluded=192,
BIASINPOLITICALBELIEFREVISION
16
Nincluded=619),andrepeatingtheANCOVAanalysis,diminishedthesizeoftheDesirabilityby
Supporterinteraction,F(1,614)=1.08,p=.298,ηp²=.00290%CI[0.000,0.012].Supporters
ofDonaldTrumpandsupportersofHillaryClintondemonstratedsimilardesirabilitybiasin
thissample:F(1,348)=27.19,p<.001,ηp²=.07[0.03,0.12]andF(1,265)=10.55,p=.001,
ηp²=.04[0.01,0.08],respectively.
Discussion
Understandinghowpeoplerevisetheirpoliticalbeliefshasimportantimplications
forsociety.Inthecontextofthe2016USpresidentialelection,weobservedarobust
desirabilitybias:individualsincorporatedinformationmoreifitwasconsistent(versus
inconsistent)withtheirdesiredoutcome.Thisbiaswasindependentofwhetherthe
informationwasconsistentorinconsistentwithindividuals’priorbeliefs.Incontrast,we
foundlimitedevidenceofanindependentconfirmationbiasinbeliefupdating.Theseresults
haveimplicationsfortheunderlyingpsychologicaltheoriesandforpoliticalbeliefrevisionin
practice.
Asubstantialbodyofworkspanningneuroscience,economics,andclinical
psychologyreportsanasymmetryintheupdatingofself-beliefswherebydesirable
informationisincorporatedmorethanundesirableinformation.Thisasymmetryhasbeen
observedwhenindividualsreceiveinformationabouttheirpersonalitytraits(Kornetal.,
2016;Kornetal.,2012),abilitiesandattractiveness(Eil&Rao,2011;Mobiusetal.,2011),or
riskofexperiencingfuturenegativelifeevents(Moutsianaetal.,2013;Sharotetal.,2011;
butseeShahetal.,2016;Garret&Sharot,inpress).Asimilaryetdistinctasymmetryhas
beenreportedintheupdatingofpoliticalbeliefswherebyindividualsbecomemore
confidentintheirpriorbeliefsdespitereceivingabalancedsetofconfirmingand
BIASINPOLITICALBELIEFREVISION
17
disconfirminginformation.Whentwoindividualswithconflictingpriorbeliefsarethus
exposedtothesamestreamofinformation,polarizationofpoliticalbeliefsisanoftobservedoutcome(e.g.,Lordetal.,1979;Taber&Lodge,2006;Taberetal.,2009).
Thepresentstudyadvancesthisworktwofold.First,wefindarobustasymmetryin
politicalbeliefupdatingthatisconsistentwithdesirabilitybias,independentofindividuals’
priorbeliefs.Incontrast,wefindlittleindependenteffectofpriorbeliefsonbeliefupdating.
Thissuggeststhatthebeliefpolarizationreportedinpreviousstudiesmaybedueto
individuals’conflictingdesires,nottheirpriorbeliefsperse.Second,whereaspast
investigationsofpoliticalbeliefupdatinghavemainlyfocusedonpoliticalattitudes(e.g.,
supportfororagainstapolicy),hereweexaminedbeliefupdatingaboutpoliticalreality—
specifically,individuals’beliefaboutwhichpresidentialcandidatewasgoingtobeelected.
Thoughonemightexpectbiasedbeliefupdatingintheformercase—afterall,attitudesare
guidedbypreferencesanddesires—itissomewhatmoresurprisingtofindthatindividuals’
desiresbiasedtheirbeliefupdatingoveraquestionoffact(Kahan,2016a).
Arecentstudyreportedthatindividualsupdatedtheirbeliefsaboutthefactsof
globalwarmingasymmetrically,butthatthespecificpatterndependeduponwhetherthey
wereweakorstrongbelieversinanthropogenicclimatechange(Sunsteinetal.,2016).
Particularly,whenconfrontedwithnewinformationregardingglobaltemperatureincrease,
strongbelieversupdatedtheirbeliefsmoreuponreceiptofostensiblyundesirable
information(i.e.,afastertemperatureincreasethanexpected),whereasweakbelievers
updatedtheirbeliefsmoreuponreceiptofostensiblydesirableinformation(aslower
increasethanexpected).Thoughthispatternappearsconsistentwithanindependent
confirmationbias,suchanoutcomemayemergewhenindividualsarepersonallyinvestedin
BIASINPOLITICALBELIEFREVISION
18
“beingright”—indeed,formanyclimatechangeactivistsabeliefthattheworldiswarming
constitutesacorepartoftheiridentity(Sternetal.,1999).Forsuchpeople,objectively
undesirable(butconfirming)informationabouttherateofglobalwarmingmaybe
subjectivelydesirable:vindicatingtheircommitmenttocombattingclimatechange
(Sunsteinetal.,2016)andaffirmingtheirculturalgroupidentity(Kahanetal.,2012).
Itisunlikelythatourdesigninadvertentlyconflatedconfirmationwithdesirabilityin
thisway.Aheadofanelection,itisdifficulttoimagineanindividualbeingpersonally
investedinthebeliefthattheirdesiredcandidatewouldnotgetintooffice.Indeed,inthe
domainofself-beliefupdating,rigorousseparationofconfirminganddesirableinformation
yieldsidenticalresultstothosereportedinthepresentstudy—namely,arobustdesirability
biasbutlimitedevidenceofconfirmationbias(Eil&Rao,2011).Wenotetheimportant
distinction,however,between(alackof)confirmationbiasobservedinbeliefupdatingas
measuredhere,andconfirmationbiasobservedinmeasuresofinformationsearchand
evaluation(e.g.,Ditto&Lopez,1992).Wedidnotdirectlyexaminethelatter,whichmayyet
manifestindependentofinformationdesirability.Additionalexplorationofourowndata
lentsupporttothisdistinction(see“informationalvalueofpolls”availableintheAnalysis
Supplement).
Finally,ourresultsofferamechanisticexplanationforwhyimpassionedpolitical
disagreementsintheUS,suchasthoseoverguncontrolorimmigration,appearincreasingly
polarizedandintractable(PewResearchCenter,2016).Insofarasindividualshavestrong
preferencesconcerningtheseissues(Kolevaetal.,2012),ourfindingssuggestthey
selectivelyincorporatenewevidenceintowhattheybelievetobetrueregardingthe
relevantfacts—provideditisconsistentwithwhattheydesiretobetrue.Polarizationover
BIASINPOLITICALBELIEFREVISION
factualbeliefsisinimicaltotheeffectivefunctioningofdemocraticsociety(Kahanetal.,
2012);itisthusaprioritytocontinueexploringwhichinterventionsamelioratethe
motivatedintegrationofevidence(Lewandowsky&Oberauer,2016).
19
BIASINPOLITICALBELIEFREVISION
20
References
Davis,C.J.,Bowers,J.S.,&Memon,A.(2011).Socialinfluenceintelevisedelectiondebates:
Apotentialdistortionofdemocracy.PloSOne,6,e18154.Doi:
10.1371/journal.pone.0018154
Ditto,P.H.,&Lopez,D.F.(1992).Motivatedskepticism:Useofdifferentialdecisioncriteria
forpreferredandnonpreferredconclusions.JournalofPersonalityandSocial
Psychology,63,568-584.Doi:10.1037/0022-3514.63.4.568
Douglas,K.M.,&Sutton,R.M.(2015).Climatechange:Whytheconspiracytheoriesare
dangerous.BulletinoftheAtomicScientists,71,98-106.Doi:
10.1177/0096340215571908
Eil,D.,&Rao,J.M.(2011).Thegoodnews-badnewseffect:asymmetricprocessingof
objectiveinformationaboutyourself.AmericanEconomicJournal:
Microeconomics,3,114-138.Doi:10.1257/mic.3.2.114
Gangarosa,E.J.,Galazka,A.M.,Wolfe,C.R.,Phillips,L.M.,Gangarosa,R.E.,Miller,E.,&
Chen,R.T.(1998).Impactofanti-vaccinemovementsonpertussiscontrol:The
untoldstory.TheLancet,351,356–361.Doi:10.1016/S0140-6736(97)04334-1
Garret,N.&Sharot,T.(inpress).Optimisticupdatebiasholdsfirm:Threetestsof
robustnessfollowingShahetal.ConsciousnessandCognition.Doi:
10.1016/j.concog.2016.10.013
Graham,D.A.(2016,17October).Republicanshavebeen‘rigging’electionsforyears.
Retrievedfromhttp://www.theatlantic.com/politics/archive/2016/10/trumpelection-rigged/504347/
Green,J.&Kapur,S.(2016,22June).NearlyhalfofSanderssupporterswon’tsupport
Clinton:Poll.Retrievedfromhttp://www.bloomberg.com/politics/articles/2016-0622/nearly-half-of-sanders-supporters-won-t-support-clinton
Hart,C.M.,Ritchie,T.D.,Hepper,E.G.,&Gebauer,J.E.(2015).TheBalancedInventoryof
DesirableRespondingShortForm(BIDR-16).SAGEOpen,5,1-9.Doi:
10.1177/2158244015621113
Horne,Z.,Powell,D.,Hummel,J.E.,&Holyoak,K.J.(2015).Counteringantivaccination
attitudes.ProceedingsoftheNationalAcademyofSciences,112,10321-10324.Doi:
10.1073/pnas.1504019112
Jolley,D.&Douglas,K.M.(2014a).Theeffectsofanti-vaccineconspiracytheorieson
vaccinationintentions.PLoSOne,9,e89177.Doi:10.1371/journal.pone.0089177
Jolley,D.&Douglas,K.M.(2014b).Thesocialconsequencesofconspiracism:Exposureto
conspiracytheoriesdecreasesintentionstoengageinpoliticsandtoreduceone’s
carbonfootprint.BritishJournalofPsychology,105,35–56.Doi:10.1111/bjop.12018
Jost,J.T.,Glaser,J.,Kruglanski,A.W.,&Sulloway,F.J.(2003).Politicalconservatismas
motivatedsocialcognition.PsychologicalBulletin,129,339-375.Doi:10.1037/00332909.129.3.339
BIASINPOLITICALBELIEFREVISION
21
Kahan,D.M.(2016a).ThePoliticallyMotivatedReasoningParadigm,Part1:Whatpolitically
motivatedreasoningisandhowtomeasureit.InR.Scott&S.Kosslyn(Eds.),
EmergingTrendsinSocialandBehavioralSciences(pp.1-16).Doi:
10.1002/9781118900772.etrds0417
Kahan,D.M.(2016b).ThePoliticallyMotivatedReasoningParadigm,Part2:Unanswered
questions.InR.Scott&S.Kosslyn(Eds.),EmergingTrendsinSocialandBehavioral
Sciences(pp.1-15).Doi:10.1002/9781118900772.etrds0418
Kahan,D.M.,Peters,E.,Wittlin,M.,Slovic,P.,Ouellette,L.L.,Braman,D.,&Mandel,G.
(2012).Thepolarizingimpactofscienceliteracyandnumeracyonperceivedclimate
changerisks.NatureClimateChange,2,732-735.Doi:10.1038/nclimate1547
Koleva,S.P.,Graham,J.,Iyer,R.,Ditto,P.H.,&Haidt,J.(2012).Tracingthethreads:How
fivemoralconcerns(especiallyPurity)helpexplainculturewarattitudes.Journalof
ResearchinPersonality,46,184-194.Doi:10.1016/j.jrp.2012.01.006
Korn,C.W.,LaRosée,L.,Heekeren,H.R.,&Roepke,S.(2016).Socialfeedbackprocessingin
borderlinepersonalitydisorder.PsychologicalMedicine,46,575-587.Doi:
10.1017/S003329171500207X
Korn,C.W.,Prehn,K.,Park,S.Q.,Walter,H.,&Heekeren,H.R.(2012).Positivelybiased
processingofself-relevantsocialfeedback.TheJournalofNeuroscience,32,1683216844.Doi:10.1523/JNEUROSCI.3016-12.2012
Lewandowsky,S.,&Oberauer,K.(2016).MotivatedRejectionofScience.CurrentDirections
inPsychologicalScience,25,217-222.Doi:10.1177/0963721416654436
Lewandowsky,S.,Oberauer,K.&Gignac,G.E.(2013).NASAFakedtheMoonLanding-Therefore,(Climate)ScienceIsaHoax:AnAnatomyoftheMotivatedRejectionof
Science.PsychologicalScience,24,622–633.Doi:10.1177/0956797612457686
Lord,C.G.,Ross,L.,&Lepper,M.R.(1979).Biasedassimilationandattitudepolarization:
Theeffectsofpriortheoriesonsubsequentlyconsideredevidence.Journalof
PersonalityandSocialPsychology,37,2098-2109.Doi:10.1037/00223514.37.11.2098
Mobius,M.M.,Niederle,M.,Niehaus,P.,&Rosenblat,T.S.(2011,May).Managingselfconfidence:Theoryandexperimentalevidence.NationalBureauofEconomic
ResearchWorkingPaperNo.17014.Doi:10.3386/w17014
Moritz,A.(2011).Vaccine-nation:Poisoningthepopulation,oneshotatatime.Enerchi.com.
Moutsiana,C.,Garrett,N.,Clarke,R.C.,Lotto,R.B.,Blakemore,S.J.,&Sharot,T.(2013).
Humandevelopmentoftheabilitytolearnfrombadnews.Proceedingsofthe
NationalAcademyofSciences,110,16396-16401.Doi:10.1073/pnas.1305631110
Nickerson,R.S.(1998).Confirmationbias:Aubiquitousphenomenoninmany
guises.ReviewofGeneralPsychology,2,175-220.Doi:10.1037/1089-2680.2.2.175
BIASINPOLITICALBELIEFREVISION
22
PewResearchCenter(2016,26August).GunRightsvs.GunControl.Retrievedfrom
http://www.people-press.org/2016/08/26/gun-rights-vs-gun-control/#party
Sedikides,C.,&Strube,M.J.(1997).Self-evaluation:Tothineownselfbegood,tothine
ownselfbesure,tothineownselfbetrue,andtothineownselfbebetter.Advances
inExperimentalSocialPsychology,29,209-269.Doi:10.1016/S0065-2601(08)600180
Shah,P.,Harris,A.J.L.,Bird,G.,Catmur,C.,&Hahn,U.(2016).Apessimisticviewof
optimisticbeliefupdating.CognitivePsychology,90,71-127.Doi:
10.1016/j.cogpsych.2016.05.004
Sharot,T.&Garrett,N.(2016).Formingbeliefs:whyvalencematters.TrendsinCognitive
Sciences,20,25-33.Doi:10.1016/j.tics.2015.11.002
Sharot,T.,Korn,C.W.,&Dolan,R.J.(2011).Howunrealisticoptimismismaintainedinthe
faceofreality.NatureNeuroscience,14,1475-1479.Doi:10.1038/nn.2949
Stern,P.C.,Dietz,T.,Abel,T.D.,Guagnano,G.A.,&Kalof,L.(1999).Avalue-belief-norm
theoryofsupportforsocialmovements:Thecaseofenvironmentalism.Human
EcologyReview,6,81-97.Availableat:https://works.bepress.com/troy_abel/3/
Sunstein,C.R.,Bobadilla-Suarez,S.,Lazzaro,S.C.,&Sharot,T.(2016).HowPeopleUpdate
BeliefsaboutClimateChange:GoodNewsandBadNews.AvailableatSSRN:
10.2139/ssrn.2821919
Taber,C.S.,Cann,D.,&Kucsova,S.(2009).Themotivatedprocessingofpolitical
arguments.PoliticalBehavior,31,137-155.Doi:10.1007/s11109-008-9075-8
Taber,C.S.,&Lodge,M.(2006).Motivatedskepticismintheevaluationofpolitical
beliefs.AmericanJournalofPoliticalScience,50,755-769.Doi:10.1111/j.15405907.2006.00214.x
Weinstein,N.D.(1980).Unrealisticoptimismaboutfuturelifeevents.JournalofPersonality
andSocialPsychology,39,806-820.Doi:10.1037/0022-3514.39.5.806