Unconscious Affective Reactions to Masked Happy Versus Angry Faces

15 Pages·2004·0.34 MB·English

ARTICLE
1PW0Ei.nR1k1Si7Oe7lNm/A0a1Ln4I 6eT1tY 6a A7l.2 N/0D 4A2 FS7FO1E3CC0I9TAILV PES PYRCIHMOINLOGGY BULLETIN
Unconscious Affective Reactions to
Masked Happy Versus Angry Faces Influence
Consumption Behavior and Judgments of Value
Piotr Winkielman
University of California–San Diego
Kent C. Berridge
University of Michigan
Julia L. Wilbarger
University of Wisconsin–Madison
The authors explored three properties of basic, unconsciously beverage? Second, how do basic affective reactions in-
triggeredaffective reactions: Theycan influence consequential teract with motivation? Third, can affective reactions
behavior,theyworkwithoutelicitingconsciousfeelings,andthey influence behavior without being accessible to con-
interactwithmotivation.Theauthorsinvestigatedtheseproper- scious awareness? In investigating these questions, we
tiesbytestingtheinfluenceofsubliminallypresentedhappyver- integratesocialpsychologicalapproachestoaffectwith
susangryfacesonpouringandconsumptionofbeverage(Study contemporary biopsychological models of incentive
1),perceptionofbeveragevalue(Study2),andreportsofcon- motivation. We suggest that such integration can ad-
sciousfeelings(bothstudies).Consistentwithincentivemotiva- vance emotion theory as well as social psychological
tiontheory,theimpactofaffectiveprimesonbeveragevalueand research on attitudes, judgment, and persuasion
consumptionwasstrongestforthirstyparticipants.Subliminal (Cacioppo & Gardner, 1999; Niedenthal & Kitayama,
smiles caused thirsty participants to pour and consume more 1994).
beverage(Study1)andincreasedtheirwillingnesstopayand To elicit basic affective reactions, researchers often
theirwantingmorebeverage(Study2).Subliminalfrownshad employ subliminal presentations of emotional facial
the opposite effect. No feeling changes were observed, even in expressions.Emotionalfacialexpressionsareimportant
thirstyparticipants.Theresultssuggestthatbasicaffectivereac- social stimuli, and their processing appears to involve
tions can be unconscious and interact with incentive motiva- biological affect programs (Ekman, 1984). Subliminal
tiontoinfluenceassessmentofvalueandbehaviortowardval- presentationsminimizecontributionsofcomplexcogni-
enced objects. tiveprocessesandreducestrategicresponding(Bargh,
Authors’Note:Fortheirgenerousassistanceatvariousstagesofthe
Keywords: affect; emotion; priming; motivation; consciousness; project, we are grateful to Hart Blanton, Gerald Clore, Phoebe
introspection Ellsworth, Kent Harber, Jan Keenan, Daniel McIntosh, Yuko
Munakata,RichardNisbett,BrucePennington,DollyRenick,Oliver
C
Schultheiss, Jeanne Shinskey, Norbert Schwarz, Diederick Stapel,
omplex emotions can result from an extended se-
ValerieStone,ElliotValenstein,andRobertZajonc.Theworkonthis
quence of cognitive appraisals (Ellsworth & Scherer, projectwassupportedbytheNationalScienceFoundationGrantBCS-
2003). However, basic affective reactions may only 0217294toPiotrWinkielman.Correspondenceshouldbeaddressedto
involveminimalprocessingandbeelicitedbysubliminal PiotrWinkielman,DepartmentofPsychology,DepartmentofPsychol-
ogy,UniversityofCalifornia,SanDiego,9500GilmanDr.,LaJolla,CA
stimuli (Zajonc,2000).Inthisarticle, we address three
92093-0109; e-mail: [email protected].
questionsaboutsuchbasicaffectivereactions.First,can
PSPB,Vol.31No.1,January2005 121-135
theyinfluenceconsequentialbehaviortowardahedonic
DOI:10.1177/0146167204271309
stimulus,asinthecaseofconsumptionofanunfamiliar ©2005bytheSocietyforPersonalityandSocialPsychology,Inc.
121
122 PERSONALITY AND SOCIAL PSYCHOLOGY BULLETIN
1996).Severallinesofresearchsuggestthatsubliminal avoidancemovements(Chen&Bargh,1999).However,
facial expressions elicit basic affective reactions with there is less evidence that affective stimuli influence
both judgmental and physiological consequences. more complex behaviors, especially when such stimuli
Niedenthal(1990)documentedthatsubliminalexpres- are presented subliminally. However, subliminal cogni-
sions influenced preference ratings for cartoon draw- tivestimulicaninfluencerelativelycomplex behaviors,
ings.MurphyandZajonc(1993)exposedparticipantsto suchasselectionofagamestrategy(Neuberg,1988)or
subliminal or supraliminal expressions that varied in decision to interrupt a conversation (Bargh, Chen, &
either valence (happy vs. angry) or gender (male vs. Burrows, 1996). Note, however, that cognitive stimuli
female).Undersubliminalpresentations,valenceofthe exert this influence by biasing the interpretation of an
expression influenced judgments of subsequent ideo- ambiguoustargetsituation.Thus,theinfluenceofacog-
graphs,butundersupraliminalpresentationsonlygen- nitiveprimerequires asemanticconnectiontothetar-
der information influenced judgments. Using facial get (Higgins, 1996). In contrast, as we explain shortly,
electromyography,Rotteveel,deGroot,Geutskens,and theinfluenceofanaffectiveprimemayonlyrequirethat
Phaf (2001) found more frowning to ideographs pre- thetargetengagesaffectiveandmotivationalprocesses,
cededbyangrythanhappyfaces,butonlyaftersublimi- even if the prime and target do not share a semantic
nal, not supraliminal, presentations. Neuroimaging connection.
studies suggest that subliminal angry and fearful faces Inthecurrentstudies,weexaminedthispossibilityby
activatetheamygdalaandrelatedlimbicstructures,pre- testingtheimpactofsubliminalpresentationsofhappy
sumablyviaadirectpathwayfromthevisualthalamusto andangryfacesontheactionsofpouringandconsum-
theamygdala(Morris,Öhman,&Dolan,1999;Whalen ingabeverage.Weselecteddrinkingforseveralreasons.
etal.,1998).Interestingly,supraliminalexpressionsacti- First, basic affective and motivational processes should
vate limbic structures more under implicit processing havethemostinfluenceonabehaviortowardahedonic
conditions, when participants classify faces on gender, stimulus—a stimulus with some initial value. In fact,
ratherthanexplicitprocessingconditions,whenpartici- drinking is often used in social and biological psychol-
pantsclassifyfacesonexpression(Critchleyetal.,2000). ogy to explore such basic processes (Edwards, 1990;
Littleisknownaboutthenatureofaffectivereactions Laeng, Berridge, & Butter, 1993; Strahan, Spencer, &
elicited with different types of subliminal expressions, Zanna,2003).Second,itisimportanttoexplorewhether
butitappearsthatsuchreactionsareorganizedmainly theimpactofaffectiveprimingcangobeyondratingsof
on a positive-negative dimension (Zajonc, 2000). simple drawings and influence a consequential behav-
Accordingly, effects of happy versus angry subliminal ior.Byconsuminganunfamiliarbeverage,apersonvol-
expressions on judgments combine additively with untarilyingestsasubstancethatispotentiallybeneficial
effects of other basic affect inductions, such as mere orharmful.Third,documentingapossibleinfluenceof
exposure(Murphy,Monahan,&Zajonc,1995).Further- affective primes on drinking is particularly informative
more,changesinglobalpositive-negativejudgmentsof becausethevalueofabeverageisintrinsicallypredeter-
stimuli can be obtained with different negative facial minedbyitstasteandothersensoryproperties.Ifaffec-
expressions,suchasangerordisgust(Murphy&Zajonc, tive primes can overcome such predetermination, it
1993; Niedenthal, 1990). Finally, under brief presenta- would suggest a robust underlying affective reaction.
tion conditions, participants have trouble extracting Finally,drinkingallowsonetoassesstheimpactofaffec-
morethangeneralnegativityfromdifferentexpressions, tiveprimingwith“real-world”unitssuchasvolumeand
such as anger, fear, disgust, or sadness (Murphy, 2001). price.
In sum, the extant research on effects of subliminal
Affect and Motivation
expressions suggests that they elicit basic positive-
negativeaffectivereactionsandthatthesereactionscan Previous research has focused on the role of basic
have immediate judgmental and physiological conse- affect in preference judgments (Zajonc, 2000). How-
quences.Buildingonthisresearch,thecurrentstudies ever, recent advances in biopsychology highlight the
employed subliminal expressions to address several close connection between affect and motivation. Tra-
questions about the nature of basic affective reactions. ditional theories of motivation postulated that hedonic
behavior (behavior toward stimuli with affective value)
Affect and Behavior
wasdrivenbyneedstates(Hull,1951).Forexample,peo-
Facilitation of adaptive behavior has long been con- pledrinktoreducetheunpleasantstateofthirst.Incon-
sidered an important function of the affect system trast,contemporarybiopsychologicaltheoriesofmotiva-
(Frijda, 1999). There is evidence that affective stimuli tionpositthathedonicbehaviorislargelydeterminedby
influencestartlereflexes(Lang,1993),aswellassimple the stimulus’ incentive value: whether the stimulus trig-
instrumental actions, such as immediate approach- gersapositive-negativeaffectiveresponseandpromotes
Winkielman et al. / AFFECTIVE PRIMING 123
approach-avoidance motivation. That is, modern theo- strong notion of unconscious cognition (Kihlstrom,
riescombineaffectandmotivationinthesameexplana- Mulvaney,Tobias,&Tobis, 2000).Anunconscious affect,
toryframework(Toates,1986).Motivationalstates,such in the stronger sense, would be a reaction caused by
as thirst, are still important, but they work by directly valenced stimuli and with valenced behavioral conse-
influencingaffectiveandmotivationalresponsestothe quences,whichnonethelessisnotsubjectivelyfelt,even
relevantfeaturesofthestimulus.Thisinfluencecanbe upon introspection.
observed in a phenomenon known as alliesthesia— Several psychologists and neuroscientists have pro-
change in incentive value as a function of a relevant posedtheexistenceofunconsciousaffectiveprocesses,
motivational state (Cabanac, 1971). For example, peo- basedonstudiesofanimals,patients,drugaddicts,and
ple perceive taste of water more favorably and want it people under hypnotic analgesia (Berridge, 1999;
morewhentheyarethirsty,butnotwhentheyarehungry Kihlstrometal.,2000;Lang,1993;LeDoux,1996).How-
(Rolls, Rolls, & Rowe, 1983). Neuroscientific research ever,unconsciousaffectivereactionshavenotbeendem-
highlightsthecloseconnectionbetweenaffectandmoti- onstrated in normal adult human participants. One of
vation postulated by incentive theories. For example, thefewindicationsofunconsciousaffectcomesfroma
animal studies suggest that the mesolimbic dopamine studythatfoundmodulationofparticipants’preference
system,thenucleusaccumbens,andtheamygdalasup- ratings by subliminal happy or angry expressions
port both affective and motivational responses (Winkielman, Zajonc, & Schwarz, 1997). Despite
(Berridge, 1996). Human neuroimaging studies also changesintheirpreferences,inexitinterviews,partici-
show activation of these systems to incentive stimuli, pantsdeniedexperiencingchangesinconsciousmood,
such as drugs or money (Knutson et al., 2001) and, eventhoughtheywerealertedtosuchapossibility.Fur-
important for the current studies, beverages (Berns thermore,whengivenanalternativeattributionfortheir
etal.,2001)andemotionalfacialexpressions(Critchley mood changes, participants still showed the affective
et al., 2000; Whalen et al., 1998). priming effect. Nevertheless, failure to report an emo-
Thetightlinkbetweentheaffectiveandmotivational tion or change its attribution might be due to inatten-
systemssuggeststhatstimuli,suchasfacialexpressions, tion, lack of motivation, or memory distortion.
thatelicitbasicaffectiveresponsesshouldmodulatethe A more convincing demonstration of unconscious
incentivevalueofasubsequenthedonictarget,suchasa affect requires obtaining on-line ratings of conscious
beverage,eventhoughtheseeventsareostensiblyunre- feelings immediately after subliminal exposure to the
lated.Thus,wepredictedthatexposuretohappyexpres- valenced stimulus. Thus, in current studies, we asked
sionsshouldtemporarilyincreasetheincentivevalueof someparticipantstoratetheirmomentarymoodimme-
thebeverageandfacilitateconsumption,whereasexpo- diatelyafteraseriesofsubliminalexpressions.Ifpartici-
sure to angry expressions should temporarily decrease pantsdidnotreportanymoodchangeimmediatelyafter
the incentive value of the beverage and suppress con- the subliminal exposure, it would be difficult to argue
sumption.Furthermore,wepredictedthatthemodula- thattheirfailuretodosowasduetoattentionormemory
tionofincentivevalueshoulddependonthepreexisting deficits, especially if they subsequently demonstrated
motivational state of the individual, just as the normal behavioralandjudgmentalconsequencesoftheiraffec-
incentive value of a drink is modulated by preexisting tive reaction. To ensure that participants had a fair
thirst (Rolls et al., 1983). If such a modification is ob- chancetodetectanyinfluenceontheirfeelings,wegave
served, it would suggest that preferences are jointly them two scales tapping various aspects of emotional
determined by basic affective processes, influenced by experience. As described in more detail later, Study 1
subliminal expressions, and basic motivational pro- usedatwo-itemscaleaskingparticipantstoreporttheir
cesses, influenced by a physiological need. current mood on a broad positive-negative dimension
and their current arousal. Study 2 used a Positive and
Affect and Awareness
NegativeAffectSchedule (PANAS)scalecontaining20
Can affective reactions to subliminal expressions adjectives to assess mood using differentiated terms
occurwithoutawarenessofthosereactions?Mostscien- (e.g.,excited,scared,irritated)orusingageneralpositive-
tistsagreethatatriggeringaffectivestimuluscanbeun- negativeindex.ThePANASscalehasgoodreliability,is
conscious but assume that the resulting affective reac- sensitivetochangesovertime,andisconsideredoneof
tionisitselfconscious(e.g.,Clore,1994;Öhman,Flykt, the best measures of current mood (Watson, Clark, &
& Lundqvist, 2000; Zajonc, 2000). In fact, the scien- Tellegen, 1988). Finally, we interviewed participants at
tific term unconscious affect is typically used to refer to theendofthestudyabouttheiraffectivereactionsand
the unconsciousness of affect induction, not to the re- testedwhethertheycoulddetectthesubliminalprimes.
sulting affective state. However, a stronger notion of Across these measures we predicted no effect on con-
unconscious affect may be considered, similar to the scious feelings, despite evidence of affective influence
124 PERSONALITY AND SOCIAL PSYCHOLOGY BULLETIN
ondrinkingmeasures.Thispredictionwasbasedonthe ateabeverage.Tominimizedemandcharacteristics,par-
assumption that subliminal facial expressions activate ticipants received instructions via computer and re-
low-level affect mechanisms that can function inde- corded their ratings on an anonymous questionnaire,
pendently from mechanisms underlying subjective ex- with the experimenter out of view. After the study, all
perience,asobservedinourearlierwork(Winkielman participants were debriefed and thanked.
et al., 1997).
Initial motivational state. Before participants were
introduced to the computer task, they completed a
OVERVIEW
“background”questionnaireratingtheircurrentlevelof
In summary, the current studies were designed to thirst(0=notatall,11=verythirsty)andhunger(0=notat
addressthefollowingquestions.First,canbasicaffective all, 11 =very hungry).
reactionstriggeredbysubliminalexpressionsinfluence
Sequence of experimental events. Figure 1 shows the
the pouring and consumption of a beverage? Second,
sequence of experimental events. Participants were
does motivation (thirst)change theimpact ofaffective
primedwithsubliminalexpressionsembeddedinagen-
primes? Third, do subliminal facial expressions alter
der classification task and then performed a beverage
conscious feelings? We investigated these questions in
task and a mood rating task, as explained shortly. To
two studies using a modified version of the subliminal
establishabaselineofparticipants’drinkingandrating
affectiveprimingparadigm.Inbothstudies,participants
responses, the experiment started with a set of eight
first rated their motivational state—thirst and hunger.
baseline trials with neutral primes, followed by a set of
Next,participantswereexposedtoaseriesofsubliminal
eighttesttrialsinwhichprimevalencewasmanipulated
happyorangryfacesinataskrequiringgenderclassifica-
between participants. Participants then received
tionofsupraliminalneutralfaces.Afterpriming,partici-
anothersetofbaselinetrialsfollowedbyanothersetof
pants received a hedonic stimulus—a fruit-flavored
testtrialswithexpressionsofreversedvalencefromthe
drink. We chose this specific drink because previous
initialtest.Nosignificanteffectsemergedintheselater
studiesshowed motivationalmodulation ofitshedonic
testtrialsinStudy1orStudy2,sotheyarenotfurtherdis-
value,suggestingthatemotionalexpressionscouldexert
cussed. The absence of priming on the later test trials
a similar effect (Laeng et al., 1993). Study 1 examined
couldreflecthabituationofaffectiveresponsestofacial
affective influence on behavior toward the beverage.
expressions(Wrightetal.,2001)orresistancetoinduc-
Participantsfreelypouredthemselvesthebeverageand
tionofoppositevalenceoncestimulusvalencehasbeen
consumedasmuchastheywanted.Study2testedaffec-
established(Murphy&Zajonc,1993;Winkielmanetal.,
tiveinfluenceonperceptionofthebeverage’svalue.Par-
1997).
ticipantstastedasmallpredeterminedsampleofthebev-
erageandthenratedit.Inbothstudies,presentationof
Priming trials: Gender classification task. Subliminal
thebeveragetaskwascounterbalancedwithself-reports
expressionswereembeddedina“gender-classification”
ofcurrentfeelings,oneithera2-itemscale(Study1)ora
task,inwhichparticipantsreportedthegenderofeasily
20-item scale (Study 2). We predicted the following:
visible neutral faces. Each trial started with a forward
First,theimpactofaffectiveprimesshouldbemostpro-
mask(50-mscross),followedbyasubliminalprime(16-
nounced on variables tapping the beverage’s hedonic
ms expression), which was immediately replaced by a
value,suchasconsumption(Study1)andratingsofbev-
backward mask (400-ms neutral male or female face).
eragedesirability(Study2).Second,theimpactofaffec-
Each baseline and test priming sequence contained
tiveprimesshoulddependonmotivationalstate(thirst).
eight trials with subliminal expressions of the same
Third, the impact of affective primes should occur
valence(i.e.,allhappy,allneutral,orallangry),withdif-
without a change in conscious mood.
ferentindividualfacesoneachtrial.Wechoseeighttrials
pereachprimingsequencebecausethisnumberofsub-
STUDY 1 liminalexpressionsappearssufficientforatransientacti-
vationofthelow-level affectivecircuitry,whereas more
Method
trialsleadtohabituation(Whalenetal.,1998).Wechose
Participants.Thirty-nineundergraduates(14men,25 happy and angry faces for three reasons. First, these
women, mean age = 22 years) gave informed consent expressionsworkedinpreviousaffectiveprimingstudies
and participated in individual sessions for extra credit. (Murphy&Zajonc,1993;Winkielmanetal.,1997).Sec-
Uponarriving,participantslearnedthatthestudyinves- ond,itiseasiertoextractnegativevalencefrombriefly
tigated how “biological rhythms influence reaction presentedexpressionsofangerthanfromfear,disgust,
times,sensations,andmood”andrequiredthemtoper- or sadness (Murphy, 2001). Third, we did not want to
formarecognitiontask,toratetheirmood,andtoevalu- elicit a specific consumption-related reaction, such as
Winkielman et al. / AFFECTIVE PRIMING 125
Figure 1 Sequence of events in Study 1 testing the influence of subliminal expressions on consumption behavior and feelings.
NOTE:Participantswereprimedwithsubliminalfaces(baselinetrials=8neutralexpressions;testtrials=8happy,8neutral,or8angryexpressions)
embeddedinataskrequiringgenderclassificationofvisibleneutralfaces.Primingtrialswereimmediatelyfollowedbyeitherratingsofmoodand
arousal or by presentation of the drink (in counterbalanced order).
disgust, but rather a general positive-negative affective right now, at this very moment?” (–5 = unpleasant, 5 =
reaction. pleasant)and“Howmucharousaldoyoufeelrightnow,
Thefacestimuliwere24neutralfaces,whichservedas atthisverymoment?”(–5=low,5=high).Inthebeverage
primesandmasks,and8happyand8angryfaces,which task,participantsreceivedanopaquepitchercontaining
servedonlyasprimes.Ofthese40faces,20werefemale 600 ml of a beverage prepared with water, sugar, and
and20male,and20wereJapaneseand20Caucasian.All lemon-lime-flavored Kool-Aid powder. To ensure that
came from the Japanese and Caucasian Facial Expres- thedrinkappeared novelaftereachsetofthepriming
sions of Emotion (JACFEE) set by Matsumoto and trials,thepitcherpresentedaftereachsetofprimingtri-
Ekman (1988). Images were 8 × 8 cm, shown in gray alscontainedabeveragemadeoffourdifferentpropor-
scale, centrally on a 15-inch monitor, 50 cm from the tionsofsugarandpowder,counterbalancedacrosspar-
participant, using PsyScope software on a McIntosh ticipants.Participantspouredasmuchofthebeverageas
computer. theywantedintoa250-mlcupandconsumedasmuchas
theywanted.Unknowntotheparticipants,theamounts
Beverage and feelings task. After the gender classifica- poured and consumed were recorded using an elec-
tion task, participants rated their current feelings and tronic scale. After pouring and drinking, participants
performedabeveragetask,incounterbalancedorder.In gavefourratings:“Howdeliciousisthisdrink?”(0=not
thefeelingstask,weaskedparticipants,“Howdoyoufeel delicious,10=extremelydelicious);“Howmuchwouldyou
126 PERSONALITY AND SOCIAL PSYCHOLOGY BULLETIN
liketodrinkrightnow?”(0=none,5=2pints);“Howwell entered the baseline response, thirst, and priming. In
did it quench your thirst?” (0 = not at all, 10 = extremely thesecondblock,weenteredthecrossproductofthirst
well); and “How much would you pay for a can of this andpriming.Thefirstregressionblockrevealedonlya
drink?” (1 =10 cents, 10 =1 dollar). main effect for prior pouring levels: Participants who
poured more during the baseline phase also poured
Prime perceptibility task. A forced-choice recognition
moreduringthevalencedphase.Thenonstandardized
task evaluated participants’ awareness of 16 emotion
regression coefficient indicated that for every 1-ml
expressions presented during the test sequences
increaseinpouringduringthebaselinephase,pouring
(Winkielmanetal.,1997).Oneachof16trials,apartici-
inthevalencedphaseincreasedby.32ml,t(35)=2.06,
pantwasfirstsubliminallyflashedanemotionalexpres-
p< .05.
sion masked by a neutral face. Immediately afterward,
Thesecondregressionblockrevealedaninteraction
theparticipantsawtwofacespresentedfor2secondson
betweenthirstandpouring,B=6.52,t(34)=3.02,p<.01.
theleftandrightpartofthescreen (oneofwhichhad
The interaction is shown in the top panel of Figure 2,
beenthesubliminalexpression)andwasaskedtodecide
whichplotstheamountpouredasafunctionofpriming
whichfaceswerepresentedintheflash.Performanceon
andthirst,controllingforbaselinepouring.Atthehigh
thistestwas51.6%,whichwasnotsignificantlydifferent
thirst level (1 SD above the sample mean for thirst),
fromchance,t(38)=1,p=.30.Regressiontestsrevealed
pouringincreasedasafunctionofpriming,goingup21
no relation between performance in the forced-choice
mlbetween angry(37ml),neutral(58ml),andhappy
testandtheinfluenceofsubliminalprimingonthecriti-
(79ml)primes,B=21.31,t(34)=3.53,p<.01.Thatis,
cal dependent variables, including pouring, drinking,
thirstyparticipantspoured114%moreofthebeverage
and mood (ts < 1). Thirst level did not correlate with
afterhappyprimesthanafterangryprimes.Atthemean
forced-choiceperformance(r=.05).Theseresultssug-
levelofthirst,pouringexhibitedamarginallysignificant
gestthatconsciousrecognitionwasnotresponsible for
increase as a function of priming, going up 7.52 ml
the observed effects of emotional expressions.
between angry (47ml), neutral (55ml), and happy
Results (62ml) primes, B = 7.52, t(34) = 1.80, p < .08. That is,
moderatelythirstyparticipantspoured32%moreofthe
Descriptive statistics and analytic method. We analyzed
beverageafterhappythanangryprimes.Atthelowlevel
data using linear regressions with valenced phase re-
ofthirst(1SDbelowthemeanofthirst),thedifference
sponses(pouring,drinking,mood,arousal)ascriterion
between priming conditions was not significant, B =
variablesandbaselinephaseresponses,primingcondi-
–6.27,t(35)<1.Insummary,thirstyparticipantspoured
tion,andthirstlevelaspredictorvariables.Ourcoding
morethantwicetheamountofthebeverageafterhappy
for linear regression analyses assumed equal intervals
primesthanangryprimes,butprimingdidnotinfluence
between angry and neutral and between neutral and
pouring of nonthirsty participants.
happyconditions(–1=angry,0=neutral,1=happy,with
higher scores indicating more positive valence). An
Amountconsumed.Inthebaselinephase,participants
alternative strategy was to dummy code separately for
drank27.06ml(SD=22.76)or48%ofwhattheypoured.
angryandhappyprimingconditions,whichallowsinde-
Inthevalencedphase,participantsdrank26.79ml(SD=
pendentcomparisonsoftherelativeimpactofangryver-
26.44) or 50% of what they poured. To determine
sus happy priming on the criterion variable. However,
whether priming and thirst influenced drinking in the
thiscomparisonwasnotcentraltothecurrentresearch,
valenced phase, we regressed valenced phase drinking
andwedidnotcollectalargeenoughsamplesizetopur-
ontobaselinedrinking,thirst,priminginthefirstregres-
sue selective comparisons between angry and happy
sionblock,andthecrossproductofthirstandprimingin
primes. Exploratory analyses did reveal, however, that
the second block. The first block revealed only a main
angryandhappyprimesinteractedwiththirstinoppo-
effect for prior drinking: Participants who drank more
site directions. This result is consistent with our theo-
before priming also drank more after priming. The
retical framework and our coding strategy assuming
nonstandardized regression coefficient indicated that
equal intervals.
forevery1-mlincreaseindrinkingbeforepriming,post-
Amount poured. Participants poured 56.21 ml (SD = primedrinkingincreasedby.58ml,t(35)=3.47,p<.001.
24.63)ofthebeverageduringtheneutralbaselinephase As with pouring, the second block revealed an interac-
and53.92ml(SD=23.19)inthevalencedtestphase.To tionofthirstandprimingondrinking,B=6.02,t(34)=
investigate the predictors of pouring, we regressed 2.59, p < .01. This interaction is shown in the bottom
valenced-phase pouring onto baseline-phase pouring, panelofFigure2.Athighlevelofthirst,primingcaused
thirst,priming,andthecrossproductofthirstandprim- anincreaseindrinking,goingup17mlfromangry(12
ing. In the first block of a hierarchical regression, we ml)toneutral(29ml)andtohappy(46ml)primes,B=
Winkielman et al. / AFFECTIVE PRIMING 127
Figure 2 Amountofbeverageinmilliliterspoured(toppanel)and Figure 3 Ratingsofmood(toppanel)andarousal(bottompanel)asa
consumed(bottompanel)asafunctionofsubliminalprim- functionofsubliminalprimingandinitialthirst.Moodand
ing and initial thirst. arousalwereeachmeasuredona–5to+5scale,rangingfrom
NOTE: Values are estimated by linear regression (priming coded: positive-negative to low-high, respectively. Values are esti-
angry=–1,neutral=0,happy=+1)atdifferentthirstlevels(low=1SD matedbylinearregressionandadjustedforbaselinelevels.
belowthemean,mean=samplemean,high=1SDabovethemean)
and adjusted for baseline levels.
participants’moodtrendednonsignificantly(p=.35)in
17.11,t(34)=2.73,p<.01.Inpercentages,thirstypartici- theprime-incongruentdirection(2.34afterangry,2.02
pants drank 280% more of the beverage after happy after neutral, and 1.70 after happy primes), and their
primes than after angry primes. At the mean level of patternforarousalwasnearlyflat(0.36afterangry,0.38
thirst,primedidnotsignificantlyincreasedrinking,B= after neutral, 0.40 after happy primes).
4.38,t(34)<1.00(23mlafterangry,27mlafterneutral, Median-splitonthirst.Thejustpresentedanalysesused
32 ml after happy primes, representing an 18% differ- linearregressionstoestimatetheprimingeffectatthree
ence).Finally,atthelowlevelofthirst,primingalsodid levels of thirst.The results showed thatathigh level of
notinfluencedrinking,B=–8.36,t(35)=1.22,p<.24,34 thirst,affectiveprimes reliably influenced participants’
ml after angry, 26 after neutral, and 17 ml after happy consumptionwithoutinfluencingtheirfeelings.Similar
primes). In summary, thirsty participants drank more resultsholdwhenthedataareanalyzedbydividingpar-
beverage after happy primes than angry primes, but ticipants into two independent groups using a median
nonthirsty participants did not. splitonthirst.Toillustratethis,Figure4plotstheprime
effect on consumption behavior and subjective experi-
Feelingratings.Equivalentanalysesalsoassessedmood
ence for participants who scored at or above the thirst
and arousal ratings. We regressed valenced-phase feel-
median. As shown, priming influenced consumption
ing ratings onto baseline-phase feeling ratings, thirst,
behaviorofthirstyparticipants(leftpanel),butnottheir
priming,andthecrossproductofthirstandprime.The
subjective experience (right panel).
results,showninFigure3,revealedonlyamaineffectof
prior levels. Participants who reported more positive Correlations between consumption and feeling measures.
moodinthebaselinephasealsoreportedmorepositive Finally,weanalyzedthecorrelationsbetweenfeelingrat-
moodinthevalencedphase,B=1.09,t(35)=10.35,p< ings and consumption behaviors during the valenced
.01.Likewise,participantswhoreportedmorearousalin phase.Mooddidnotcorrelatewithpouring,r(39)=.03,
the baseline phase also reported more arousal in the ordrinking,r(39)=.05.Similarly,arousaldidnotcorre-
valencedphase,B=.939,t(35)=21.5,p<.01.Nomain late with pouring, r(39) = .01, or drinking, r(39) = .01.
effects or interaction of priming and thirst occurred, Moreimportant,theselowcorrelationswerenotdueto
regardlessofthelevelofthirst,evenwhenfeelingswere thelackofvariabilityorsensitivityofoursubjectivereport
rated immediately after the presentation of valenced measures. This is indicated by a significant correlation
priming(beforedrinking).Infact,thepatternforthirsty betweenarousalandmood,r(39)=.52,p<.01,replicating
128 PERSONALITY AND SOCIAL PSYCHOLOGY BULLETIN
Figure 4 Means and standard errors for measures of consumptionbehavior(pouringand drinking)and subjectiveexperience(mood and
arousal) among thirsty participants.
NOTE: Values are for participants at or above median thirst and are adjusted for baseline levels.
the previously documented associationbetweenmoder-
atearousalandpositivemood(Thayer,1996). tive state, even when their mood was assessed imme-
diately after the subliminal primes. Nonthirsty partici-
Beverage ratings during the baseline phase. To assess
pants’ pouring and drinking did not change after sub-
whether participants’ motivational state influenced
liminal priming. Apotentially puzzling findingwasthe
theirperceptionoftheinitialvalueofthedrink,wecor-
absenceofprimingeffectsonbeverageratings,obtained
relatedtheirratingsofthebeverageduringthebaseline
after participants finished drinking as much as they
phase with their initial level of thirst. Thirst correlated
wished.Afterall,previousstudiesreportedeffectsofsub-
positively with ratings of the beverage’s deliciousness,
liminalexpressionsonratingsofvisualpatterns,suchas
r(39) = .39, p < .05, and its thirst-quenching ability,
cartoons or ideographs (Murphy & Zajonc, 1993;
r(39) = .36, p < .05. This effect represents an enhance-
Niedenthal,1990;Winkielmanetal.,1997).Thediscrep-
mentofstimulusvaluebyappetitestate(alliesthesia),as
ancy could be due to procedural differences between
documented previously (Cabanac, 1971).
previous studies and our study. First, previous studies
Beverage ratings during the valenced phase. As the final exploredimmediateimpressionsofvisualpatternspre-
task,afterpouringandconsumingthebeverageandrat- sented for only 1 or 2 seconds. In contrast, we allowed
ingsubjectiveexperience, participantsratedthebever- participants to pour and consume as much and for as
age.Analysesrevealednomaineffectofprimingandno longastheywantedbeforegivingratings,resultingina
interactioneffectofprimingandthirstonbeveragerat- greaterdelaybetweensubliminalprimingandtargetrat-
ings.Weaddressthisabsenceofbeverageratingeffects ing. If subliminal priming influences only the initial
in the next study. impressions,thenitseffectcouldhavedissipatedbythe
timeourparticipantsratedthedrink(foradiscussionof
Discussion
thisissue,seeZajonc,2000).Second,thirstyparticipants’
Study1resultsconfirmedourpredictionsthatsublim- consumption of different amounts (after happy versus
inalfacialexpressionswouldalterconsumptionandthat angryfaces)mighthaveconfoundedanyprimingeffect
theirinfluencewoulddependonrelevantmotivational ondrinkratings,especiallyifadditionalconsumptionof
state.Inparticular,thirstyparticipantspouredmoreand a moderately tasty beverage depressed subsequent rat-
drankmoreofabeverageafterexposuretohappyfaces ings.Third,we foundthatthestrongestpriming effect
but poured less and drank less afterexposure to angry required high thirst. This raises the possibility that let-
faces. Remarkably, despite these changes in behavior, ting participants quench their thirst before rating the
thirstyparticipantsreportednochangeintheirsubjec- beveragecaneliminatetheprimingeffectonratings.To
Winkielman et al. / AFFECTIVE PRIMING 129
addresstheseissues,Study2modifiedtheprocedurefor sensoryaspectsofthebeverage,thequestionaboutthirst
assessment of beverage ratings. quenchingwasreplacedbythequestion,“Howsweetis
this drink?” (0 =not sweet at allto 10 =extremely sweet).
STUDY 2 Moodratings.Participantsratedtheiraffectiveexperi-
ThegoalofStudy2wastoreassesstheeffectofsublim- enceonthePANASquestionnairebyindicatingona5-
inalprimingonbeverageratings.IncontrasttoStudy1, point scale how they felt “right now, at this very
participants did not pour the drink themselves but moment,” using thefollowing 20 items: interested, dis-
insteadreceivedafixedandsmallquantityofthebever- tressed, excited, upset, strong, guilty, scared, hostile,
agetosamplewithonlyonesip.Inaddition,becausepar- enthusiastic, proud, irritable, alert, ashamed, inspired,
ticipantsdidnothavetopourthebeveragethemselves, nervous,determined,attentive,jittery,active,andafraid
they tasted it immediately after priming. We predicted (anchors ranging from 1 = slightly or not at all to 5 =
that under these conditions, subliminal affective prim- extremely).
ingshouldmoreclearlyinfluenceperception ofbever-
Prime perceptibility task. The subliminal presentation
agevalue.Wealsousedamoredifferentiatedmeasureto
was verified using the same methods as in Study 1 and
examine the impact of subliminal expressions on con-
again did not significantly exceed chance (54%). Nei-
sciousmood.Thetwo-itemglobalscaleusedinStudy1
therwasthescoreintheforced-choicetestsignificantly
capturesonlylimitedattributesofaffectiveexperience.
relatedtotheinfluenceoffacialexpressionsoncritical
So,perhapssomeaspectofmoodchanged,butthescale
dependentvariables(ratingsofdrinkvalueandratings
did not allow participants to report it. Therefore, in
of mood) or on level of thirst (ts < 1).
Study2,weusedaPANASquestionnairetomeasure20
nuancesofexperience(Watsonetal.,1988).Thisscale Results
providedamorestringenttestofourhypothesisthatsub-
Willingnesstopay.Participantsindicatedtheirwilling-
liminal priming induces an affective reaction that is
nesstopayforahypotheticalcanofthebeverage ona
introspectively inaccessible. The design of Study 2 was
scalerangingfrom10centsto1dollar(U.S.).Afterthe
otherwise similar to that of Study 1. Participants first
valencedprimingphase,participantswerewillingtopay
ratedtheircurrentlevelofthirstandhunger.Next,they
31 cents. Because priming influenced only immediate
wereexposedtoeightsubliminalhappyorangryfacial
reactions in Study 1, we tested whether the effects dif-
expressionsembeddedinthegenderclassificationtask.
feredbetweenparticipantswhoratedthedrinkimmedi-
Onthefirstsetofprimingtrials,allsubliminalfaceswere
atelyafterprimingandparticipantswhofirstcompleted
neutraltoestablishabaselineofratingresponses.Inthe
the intervening mood questionnaire. A multivariate
secondsetoftrials,wemanipulatedthevalenceofsub-
analysis of variance (MANOVA) assessed the effects of
liminal expressions between participants. To simplify
independent variables prime (happy vs. angry) and
thedesign,theprimesincludedhappyandangryexpres-
order(beverageratingfirstvs.moodratingfirst),aswell
sions,butnotneutralexpressions.Aftereachsetofeight
as the covariate of baseline willingness to pay, on the
primingtrials,participantsperformedabeveragerating
dependent variable of postpriming willingness to pay.
task or a mood rating task in counterbalanced order.
Theanalysisrevealedaprime-by-orderinteraction,F(1,
Method 28)=4.41,p<.05.Testofsimpleeffectsrevealedthatpar-
ticipantswhoratedthebeveragefirstwerewillingtopay
Participants. Twenty-nine undergraduates gave in-
37 cents after happy primes and 19 cents after angry
formedconsentandparticipatedforextracreditinapsy-
primes, F(1, 28) = 4.72, p < .04. For participants who
chology course (6 male students, 23 female students,
rated mood first,priming had no influence on willing-
mean age = 20 years). After completing the study, all
ness to pay (33 vs. 34 cents, p = .44). In short, affective
participants were debriefed and thanked.
primingnearlydoubledthedrink’smonetaryvalue,but
Beverage sampling and ratings. Consumption was re- this effect was short-lived and only occurred when
strictedbyofferingparticipantsonlya40-mlbeveragein participants responded immediately after priming.
an 80-ml cup. Participants took a single sip and then A MANOVA testing for the interaction of priming,
answered the following questions. Three questions order, and thirst (as measured by our continuous 11-
assessedperceptionofthehedonicandincentivevalue pointscaleonbeverageratings)revealedonlyamargin-
of the beverage: (a) “How delicious is the drink?” (b) allysignificantprime-by-thirsteffect,F(2,28)=3.43,p=
“How much of this drink would you like to drink right .07.However,becausesuchinteractionwassignificantin
now?” and (c) “How much would you pay for a can of Study1,andbecausethestatisticalpowerofStudy2was
this drink?” (scale anchors as in Study 1). To assess low compared to Study 1, we performed linear regres-
whetheraffectivepriminginfluencedperceptionofthe sionstestingforeffectsofprimingatdifferentlevelsof
130 PERSONALITY AND SOCIAL PSYCHOLOGY BULLETIN
primingatdifferentthirstlevels.Linearregressionanal-
yses revealed a significant effect of priming for
participantsathighthirstlevel,whowantedtodrink1.02
unitsmore afterhappyprimes thanafterangryprimes
(2.0,orabouthalf-cupversus0.98,orabout1-2sips),B=
1.02,t(28)=2.16,p<.05.Atthemeanthirstlevel,there
wasanonsignificantincrease of.48unitsinwantingto
drinkmoreafterhappythanangryprimes(1.48vs.1.0),
B=.48,p=.16.Atthelowthirstlevel,primehadnoeffect
onwantingtodrinkmore(.98vs.1.03),B=.05,p=.92.
No such effects were obtained with hunger (ts < 1). In
summary,thirstyparticipantswantedtodrinkmoreafter
beingprimedwithhappyversusangryfacialexpressions,
but nonthirsty participants did not.
Otherratings.Subliminalaffectiveprimingdidnotin-
fluence participants’ ratings of liking or sweetness. We
return to this issue in the discussion.
Moodratings.Table1presentsPANASratingsbyprim-
ingconditionanditemtype.Weexaminedthepriming
influenceonmoodinseveralways.First,weanalyzeda
Figure 5 Ratingsofthedrinkasafunctionofsubliminalprimingand subsetofpositiveitems(α=.83).Second,weanalyzeda
initial thirst. subsetofnegativeitems(α=.84).Third,weanalyzeda
NOTE:Toppanel:willingnesstopay(incents)amongparticipants
global mood index, created by subtracting negative
whoratedthebeveragefirst.Bottompanel:wantingformorebever-
age,regardlessoforder(scalevalues:none,1-2sips,half-cup,1cup,1 items from positive items (α = .72, mean after happy =
pint,and2pints).Valuesareestimatedbylinearregressionatdifferent 1.18,SE=.08,meanafterangry1.14,SE=.08).Priming
thirst levels and adjusted for baseline levels.
influencednoneofthesemoodscores,regardlessoftask
thirst.Thoseregressionstreatedthevalenced-phaserat- order and thirstlevel. Finally, we analyzed the priming
ingsascriterionvariablesandbaseline-phaseratingsas effect on each of the 20 mood scale items. At the high
covariates.Primingwascodedsothathigherscoresindi- andmeanlevelofthirst,participantsshowednoprime
cated more positive valence (0 = angry, 1 = happy). effectsonanymooditem.Atthelowlevelofthirst,partic-
ThetoppanelinFigure5showsthepatternofresults. ipantsshowedonemarginaleffect(p=.08),ratingthem-
Atthehighlevelofthirst(1SDabovethemean),partici- selves as slightly more irritated after angry (1.55) com-
pantswerewillingtopay28centsmoreafterhappythan paredwithhappyprimes(0.97).But,again,atthehigh
angryprimes(38vs.10cents),B=.28,t(13)=2.25,p= thirstlevel,therewasnohintofthiseffect(p=.67),with
.05.Atthemeanlevelofthirst,participantswerewilling participants rating their irritation as lower after angry
topay17centsmoreafterhappythanangryprimes(36 (1.59)thanhappyprimes(1.77).Furthermore,asingle
vs. 19 cents), which was marginally significant, B = .17, marginaleffectatlowthirstshouldbeinterpretedwith
t(13)=2.03,p<.08.Atthelowlevelofthirst(1SDbelow caution given there were 20 mood ratings. As a whole,
themean),participantswerewillingtopaynomoreafter theabsenceofmoodeffectsisconsistentwithStudy1,in
happythanangryprimes(36vs.28cents),B=.08,t(13)= whichprimingalsodidnotinfluenceratingsofsubjec-
.67,p<.53.Hungerdidnotinfluencewillingnesstopay, tive experience.
suggestingmotivationalspecificity(ts <1).Insummary, Mediansplitonthirst.Inthejustpresentedanalyses,we
thirsty participants who rated the drink immediately usedlinearregressionstoestimatetheprimingeffectat
after priming were willing to pay triple the price after threelevelsofthirst.Wefoundthatatahighthirstlevel,
happy primes than after angry primes, but nonthirsty priming reliably influenced selected beverage ratings
participants were not. withoutinfluencingmood.Similarresultsholdwhenthe
dataareanalyzedbydividingparticipantsintotwoinde-
Wantingformorebeverage.Participantsratedhowmuch pendentgroupsusingamediansplitonthirst.Toillus-
beveragetheywantedtodrinkona0to5scale(withval- trate,Figure6plotstheprimingeffectonbeveragerat-
ueslabeledasnone,1-2sips,half-cup,1cup,1pint,and2 ingsandmoodforparticipantswhoscoredatorabove
pints).Participantsratedtheirpostpriminglevelofwant- the thirst median. As shown, for those thirsty partici-
ingmoreofthebeverageat1.27,orslightlymorethan1 pants, priming influenced beverage evaluation (left
to2sips.ThebottompanelofFigure5plotstheeffectsof panel), but not subjective experience (right panel).