Watch Out! That Could Be Dangerous: Valence-Arousal Interactions in Evaluative Processing

13 Pages·2004·0.12 MB·English

1PR0Eo.Rb1i1Sn7Os7oN/nA0 e1Lt4I a6Tl1Y. 6 /A7 2VN0AD4L2 ES6ON66CC4IE7A-ALR POSYUCSHALO LINOTGEYR BAUCLTLIOETNISN
Watch Out! That Could Be Dangerous:
Valence-Arousal Interactions in
Evaluative Processing
Michael D. Robinson
North Dakota State University
Justin Storbeck
University of Virginia
Brian P. Meier
Ben S. Kirkeby
North Dakota State University
Seven studies involving 146 undergraduates examined the andarousal are largely independent intheireffectson
effects of stimulus valence and arousal on direct and indirect experience(Russell&Barrett,1999),physiology(Lang,
measuresofevaluativeprocessing.Stimuliwereemotionalslides Greenwald,Bradley,&Hamm,1993),andbrainactivity
(Studies1to6)orwords(Study7)thatsystematicallyvariedin (Heller&Nitschke,1998),andthereforeshouldbecon-
valenceandarousal.Evaluativecategorizationwasmeasured sidered as orthogonal dimensions of appraisal and
byreactiontimestoevaluatethestimuli(Studies2,3,and7), emotion.
Whatiscommontomultipletheories,however,isthe
latenciesrelatedtoemotionalfeelings(Study3),andincidental
assumed independence of valence and arousal. And
effects on motor performance (Studies 4 and 5). A consistent
indeed, there are dependent measures, such as self-
interaction was observed such that evaluation latencies were
reported emotional experience, that favor the inde-
fasterifanegativestimuluswashighinarousalorifapositive
pendence of the dimensions. However, this independ-
stimuluswaslowinarousal.Studies1,6,and7establishthat
enceneednotbetruewhenexamininginitialstagesof
thefindingsarenotduetostimulusidentificationprocesses.The
stimulus appraisal. This is particularly true given some
findingsthereforesuggestthatpeoplemakeevaluativeinferences
evidencethatduringinitialstagesofencoding,arousing
on the basis of stimulus arousal.
objects are presumed to be potentially dangerous,
whereas nonarousing objects are presumed to be safe.
In this connection, Schneirla (1959) attempted to
Keywords: affect; appraisal; cognition; valence; arousal; emotion characterizeemotionalvalenceandarousalastheyapply
V tolower animals,suchastheantandtheparamecium.
alence and arousal are the most important dimen-
He suggested that approach and avoidance behaviors
sionsofconnotativemeaning(e.g.,Osgood,1969),feel-
seemtobegovernedbyacurvilinearrelationshiprelat-
ing states (e.g., Russell & Barrett, 1999), and emotion-
ingemotional arousaltoemotional valence.Atmoder-
relatedbehavior(e.g.,Lang,1995).However,thereare
ate levels of intensity,approach is facilitated.Athigher
some divergent theoretical proposals related to the
question of how valence and arousal might interact in
determining these outcomes. Some scholars, such as Authors’Note:GrantsupportisacknowledgedfromtheNationalSci-
Watson(Watson,Wiese, Vaidya,& Tellegen, 1999) and ence Foundation (9817649) and the National Institute of Mental
Lang(1995),proposethatarousalisnottrulyindepend- Health (068241). Correspondenceshould be addressed to Michael
Robinson, Psychology Department, North Dakota State University,
entofvalence.Rather,arousalcaneitherbeofthenega-
Fargo, ND 58105; e-mail: [email protected].
tivevariety,reflectingtheactivationofanavoidancesys-
PSPB,Vol. 30 No. 11, November 2004 1472-1484
tem,orofthepositivevariety,reflectingtheactivationof
DOI:10.1177/0146167204266647
anapproachsystem.Othertheoristsarguethatvalence © 2004 by the Society for Personality and Social Psychology, Inc.
1472
Robinson et al. / VALENCE-AROUSAL INTERACTIONS 1473
levels of intensity, by contrast, avoidance is facilitated. wordsofÖhman(1997),preattentiveaspectsoforient-
For example, a paramecium will approach a light of a ing “are determined by simple stimulus dimensions
moderateintensitybutwillavoidalightofhighintensity suchassuddenonsetorhighintensity....Abrupt,high-
(Schneirla,1959).Amonghumaninfants,asimilarpat- intensity stimulation has been associated with danger
ternmayapply(Izard,1993).Researchandtheoryinfact throughout evolution, and therefore sensory systems
suggests thatthe ability to regulate intense stimulation havetunedimmediatelyandautomaticallytorespondto
developsearlyinchildhood,specificallywiththematura- such stimulus parameters” (p. 167).
tion of the prefrontal cortex (Harman & Fox, 1997). An implication of this view is that arousing stimuli,
Because infantsareunable toregulate intensestimula- regardlessofwhethertheyarepositiveornegative,may
tion, an intense stimulus tends to overwhelm the triggerpreattentiveaspectsoforientingandinturnbe
attentionalandcopingabilitiesoftheinfant,leadingto associated with preattentive evaluations indicative of
apparentdistressandavoidance(Harman&Fox,1997; potential danger. For example, a roller coaster ride,
Rothbart, Posner, & Boylan, 1990). whichisenjoyableformanywhorideit,maynevertheless
Valence-arousal interactions of the type mentioned beautomaticallyevaluatedasnegativebasedonitsinten-
above are probably not unique to lower animals and sity and capacity for quick motion.
humaninfants.Tounderstandwhythisistrue,itisuseful
From Independence to Integration
to appreciate the fact that there may be two distinct
preattentivemechanismsrelatedtoemotionalappraisal
On one hand, positive stimuli induce an approach
(Robinson, 1998).
orientation, whereas negative stimuli induce an avoid-
ance orientation (Neumann, Förster, & Strack, 2003).
Preattentive Mechanisms in Emotional Appraisal
Ontheotherhand,intenseandnovelstimulitriggeran
A good deal of evidence suggests that people auto- avoidanceorientation,whereasmildandfamiliarstimuli
maticallyevaluatestimuli,atleastundercertaintaskcon- triggeranapproachorientation(Zajonc,2001).Accord-
ditions (Klauer & Musch, 2003). Such evidence has ingtoRobinson(1998),bothofthesecomputationsare
comefromlatency-(e.g.,Bargh,Chaiken,Govender,& performedindependentlyatthepreattentivelevel.How-
Pratto, 1992) and judgment-based (e.g., Murphy & ever,Robinsonalsoproposedthatthesetwomechanisms
Zajonc, 1993) versions of the affective priming para- contributetoone’sconsciousunderstandingofthestim-
digm.Italsohascomefromotherparadigmssuchasthe ulus.Thatis,outputfromthetwomechanismsmustbe
emotionalStrooptask(e.g.,Pratto,1994),theaffective integrated prior to conscious evaluation.
Simon task (e.g., De Houwer & Eelen, 1998), and the Because intense stimuli are potentially dangerous
implicit association test (e.g., Greenwald, McGhee, & ones,thisintegrativeactivitymaybeeasierwhenastimu-
Schwartz, 1998). The valence of an object can be lusisbothnegativeandhighinarousal.Forexample,a
decoded quickly, a point that is apparent in findings stimulussuchasascowling,barkingdog,whichisboth
showingthatsubliminalpresentationsofvalencedstim- negativeandarousing,isclearlyapotentialthreattothe
uli are effective in altering evaluations (Dijksterhuis & self.Suchastimulusshouldbeevaluatedquicklyatthe
Aarts, 2003; Murphy & Zajonc, 1993). consciouslevel.Bycontrast,astimulussuchasaninjured
However,thereisalsoevidenceforasecondmecha- kittenmightbemoredifficulttoevaluate.Ononehand,
nism that orients attention toward novel and intense thesightofahelplessanimalengenderssomepityand
stimuli. Reliable output from this mechanism include approachmotivation;ontheotherhand,theinjurypro-
skinconductanceresponses,adecreaseinheartrate(for vokesnegativeaffectintheviewer.Thecombinationmay
visualstimuli),andenhancedattention(Lang,Bradley, resultinaconflictedstateinwhichbothapproachand
& Cuthbert, 1997). Stimulus intensity can be coded avoidance motives are present (Lang, 1995).
preattentively,asestablishedbyalargebodyoffindings Bycontrast,evaluativedecisionsmaybeslowedwhen
showingthatsubliminalstimuliarecapableofeliciting a stimulus is both positive and arousing. The intense
theorientingresponse(Öhman,1997).Althoughatthe aspectofthestimulusmaypromoteacertaindefensive
conscious level orienting occurs for both positive and orientationuntilthestimulusisbetterunderstood.And
negativestimuli(Langetal.,1997),thereisareasonto yet, the positive affective nature of the stimulus pre-
think that, at the preattentive level, the orienting cludesmakinganegativeevaluation.Theresultshould
response may be associated with the fear system beconsiderableuncertaintyastowhetherthestimulusis
(Öhman, 1997; Robinson, 1998). Of interest, however, beneficial or harmful during early stages of stimulus
its inputs are not specific to threatening stimuli (e.g., encoding. For example, there are few acts that bring
snakes,angerfaces).Rather,itmayrespondtoabroader morepleasurethansex.Andyet,wedonotexpecttosee
range of inputs that are indicative of potential danger, attractivenakedstrangersinourlivingroom,andweare
such as intensity and capacity of quick motion. In the notimmediatelypreparedtocopewithsuchopportuni-
1474 PERSONALITY AND SOCIAL PSYCHOLOGY BULLETIN
ties.Bycontrast,apositivelowarousalstimulussuchasa Method
pieceofchocolatecakebringspleasurebutnoimmedi-
Participants. Participants were 20 undergraduates
ate hint of danger or uncertainty. In this case, a
from the University of Illinois who received course
preattentiveevaluationofthestimulusaspositivewould
credit.
havenocompetitionfromasecondpreattentivemecha-
nism devoted to intensity or arousal. If these specula- Apparatus. A DOS-based computer controlled two
tions are correct, valence and arousal may interact in Kodak Ektapro projectors. The projectors were
determining conscious evaluation speed. These ideas designedtointerfacewithacomputerandbecapableof
motivated the present studies. randomized presentation sequences. Projectors were
placedwithinahandmadecasethatrestedontopofa5-
Overview of Studies
ftfilingcabinetthatprojectedimagesfrombehindand
Stimuliformostofthestudiesconsistedofslidesfrom abovetheparticipantduringthecourseofthestudy.Pre-
theInternationalAffectivePictureSystem(IAPS;Lang, cise timing was accomplished by using high-speed
Bradley, & Cuthbert, 1999). To study direct aspects of Uniblitzshuttersaffixedtotheprojectorlenses(approx-
evaluative processing, we simply asked people to, as imately 1 ms random error).
quicklyandaccuratelyaspossible,categorizestimulias Participants sat in a standard chair equipped with a
positiveornegativeinthreeseparatestudies(Studies2, smalldesk.Thechairwasplaced3ftinfrontofalarge(5
×
3, and 7). Because we hypothesized that emotional ft 5 ft) projection screen. Emotional images were
×
arousal is negatively coded, we expected valence and approximately3ft 3ftandwereprojectedontothecen-
arousaltointeractsuchthatnegative/higharousaland terofthescreen.Theenvironmentwasimmersiveinthat
positive/low arousal stimuli would be evaluated more images were typically life sized or larger.
quickly.
Stimuli. Emotional slides were chosen from Lang
Intwoadditionalstudies(Studies4and5),weasked
et al.’s (1999) IAPS. We were particularly interested in
peopletoevaluatetheslidescovertlyandthenperforma ×
choosingslidestorepresentallcellsofa2(valence) 2
second task. When two tasks are closely sequenced in
(arousal) design. After some preliminary testing (e.g.,
time,performanceonthesecondtaskisdependenton
Robinson&Clore,2001),wesettledonasetof56slides.
resolutionofthefirsttask,asknownintheattentionand
Twenty-eightofthemwerenegativeinvalence,whereas
performanceliterature(Pashler,Johnston,&Ruthruff,
× 28werepositive.Withineachvalence,therewere14low
2000). We expected to observe a similar Valence
arousal and 14 high arousal slides.1
Arousal interaction in these studies. Of importance,
It is useful to statistically establish that valence and
because stimulus and response factors were independ-
arousalweremanipulatedindependentlyinthepresent
ently varied in Studies 4 and 5, interactions must be
design.Toestablishthispoint,wecreatedadatasetwith
interpretedintermsofencodingoperationsratherthan
slideasunitofanalysis(N=56).Wethenentereddummy
response factors.
codes for our dichotomized valence (positive vs. nega-
BecauseStudies1through6allusedemotionalslides
tive)andarousal(lowvs.high)factors.Wethenadded
asstimuli,itseemedpossiblethatvalence/arousalinter-
valence and arousal means from Lang et al.’s (1999)
actionscouldbespecifictopictorialimages.Tobroaden
slide norms. Finally, we computed a valence extremity
thescopeofthecurrentframework,wesoughttorepli-
score for each slide, which was defined in terms of the
cate key results in Study 7, this time in the context of
extenttowhichLangetal.’svalencemeanfortheslide
stimulus words rather than images.
differed from 5.0, the midpoint of their scale.
A series of one-way ANOVAs established that our
STUDY 1 valence designation of slides was a strong predictor of
Prior to conducting Studies 2 through 7, we con- Langetal.’s(1999)valencenorms,F(1,54)=715.47,p=
ducted Study 1, which was concerned with stimulus .00,butdidnotpredictarousalnorms,F(1,54)=1.36,p=
selectionissues.Wesoughttoselectemotionalslidesthat .25,orvalenceextremityscores,F(1,54)=2.27,p=.14.A
were independent in valence and arousal. In addition, second series of one-way ANOVAs established that our
×
Study1soughttoshowthatthereisnoValence Arousal arousal designation of slides was a strong predictor of
interactionrelatedtothevisualcomplexityoftheslides. Langetal.’s(1999)arousalnorms,F(1,54)=186.15,p=
Totheextentthatthereisnot,subsequentfindingscan .00,butdidnotpredictvalencenorms,F<1,orvalence
beinterpretedintermsofevaluativeprocessingspecifi- extremityscores,F(1,54)=1.97,p=.17.Finally,inathird
×
cally rather than object identification processes more series of 2 (dichotomized valence) 2 (dichotomized
generally.TheidentificationnormscollectedinStudy1 arousal) ANOVAs, we confirmed that dichoto-
also were examined in Study 6 reported later in the mizedvalenceanddichotomizedarousaldidnotinter-
article. actinpredictingLangetal.’s(1999)valencenorms,F(1,
Robinson et al. / VALENCE-AROUSAL INTERACTIONS 1475
52)=2.03,p=.16,arousalnorms,F(1,52)=1.42,p=.24, 820
or valence extremity scores, F(1, 52) = 1.08, p = .30.
Valence and arousal were therefore manipulated inde-
800
pendently in the stimulus materials.
s
d
Procedures. Participants were tested within individual on
c 780
testing sessions. Following a 10-min dark adaptation se
period,theyengagedinthemainpartofthestudy.Par- Milli
n
ticipantsweretoldthatwewereinterestedinhowlongit e i 760
m
would take them to identify the main theme of slides. n Ti LHoigwh AArroouussaall
They were instructed to respond to “gist” information. o
ati 740
Specifically,westated,“Ifyouseeadog,youshouldhit u
al
thespacebarbeforeyouknowexactlywhetherthedogis Ev
a golden retriever or a German shepherd.” These an 720
e
M
instructions were designed to preclude extensive and
identification-irrelevant examination of the slides.
700
There was a 3-s delay between trials and identification
responses were made by hitting the spacebar on a
680
computer keyboard. Negative Positive
Results Figure 1 Evaluation latencies by valence and arousal, Study 2.
Prior to analyzing identification times, we log-
transformed them to normalize the distribution. We except that Study 2 used a response box rather than a
thenreplacedoutliersthatwere2.5standarddeviations computer keyboard.
beloworabovethegrandlatencymeanwiththesecutoff
× Procedures. After dark adaptation, participants were
scores.Latencymeanswereanalyzedina2(valence) 2
askedtoevaluatetheslidesasquicklyandaccuratelyas
(arousal) repeated-measures ANOVA. There was no
possible,indicatingwhethereachslideispositive(1key;
effectforvalenceonidentificationtimes,F(1,19)=1.14,
leftindexfinger)ornegative(4key;rightindexfinger).
p=.30.Therewas,however,amaineffectforarousal,F(1,
Slideswereshowntwice,eachtimeinadifferentrandom
19)=8.19,p=.01.Thethemeofhigharousalslideswas
order.2
identified faster(M=1,046ms)thanthethemeoflow
arousalslides(M=1,152ms).Thisresultcouldbesome- Results
whatconsistentwiththefindingthathigharousalslides
Inaccurate trials were dropped and times were log-
trigger more active processing within the visual cortex
transformed.Wethenreplacedtimes2.5standarddevia-
(Lang et al., 1998). Of more importance, however, the
tionsbelowandabovethegrandlatencymeanwiththese
interactionbetweenvalenceandarousalwasnotsignifi-
cutoffscores. Evaluation latencies were analyzed in a 2
cant,F<1.Becausetheinteractionwasnotsignificant,
×
(valence) 2(arousal)repeated-measuresANOVA.Nei-
interactive effects of valence and arousal cannot be
therthemaineffectforvalence,F<1,northemaineffect
viewed as consequences of visual complexity.
forarousal,F<1,wassignificant.Ashypothesized,how-
×
ever,therewasaValence Arousalinteraction,F(1,20)=
STUDY 2
19.75, p = .00. The means, reported in Figure 1, reveal
InStudy1,weshowedthatthepresentslideswerenot thatnegativeimageswereevaluatedmorequicklywhen
×
associatedwithaValence Arousalinteractiononiden- highinarousal,whereaspositiveimageswereevaluated
tificationlatencies.InStudy2,wesoughttoshowthatthe more quickly when low in arousal.3
×
sameslidesproduceaValence Arousalinteractionper-
Discussion
taining to evaluation latencies.
×
Although Study 1 indicated no hint of a Valence
Method
Arousalinteractioninfluencingidentificationprocesses,
×
Study 2 shows that there is a rather robust Valence
Participants. Participants were 21 undergraduates
Arousal interaction influencing evaluation processes.
from the University of Illinois who received course
The latter result provides support for the valence-
credit.
arousal conflict theory pursued in this article.
Stimuliandapparatus.Thestimuliwerethesameasin Atthispoint,itisworthconsideringandrejectingan
Study1.Thehardwareforthestudywasalsoverysimilar alternative account of Study 2 results. Specifically, it
1476 PERSONALITY AND SOCIAL PSYCHOLOGY BULLETIN
might be proposed thatpeople seek tomaximize their 1140
exposuretopositiveeventsandminimizetheirexposure
tonegativeevents.Asaresult,theymaybereluctantto s
d 1120
n
terminate a positive image by making an evaluation, o
c
e
resulting in slower evaluation times. However, this s
potentialaccountofourresultsisflawedforatleastthree Milli 1100
n
reasons.One,suchanexplanationforthefindingsfails e i
m
toaccountforthelackofaninteractioninStudy1.Two, Ti 1080
itfailstoexplainpreviousdatashowingthatwhenpartic- ment LHoigwh AArroouussaall
ipantsaregivenfreedomtolookataslideforasshortor g
d
u 1060
aslongasdesirable,itisthearousallevelofthestimulus, e J
v
rather than its valence, that influences viewing time ati
u
(Langetal.,1997).Three,thisaccountcouldpredicta al 1040
× Ev
valencemaineffect,butnottheValence Arousalinter- n
a
action reported in Study 2. In Study 3, we sought to Me 1020
replicate and extend Study 2.
1000
STUDY 3 Negative Positive
Method Figure 2 Evaluative judgment latencies by valence and arousal,
Study 3.
Participants. Participants were 20 undergraduates
from the University of Illinois who received course
credit.
the1key.Asinpriorstudies,wedeletedinaccuratetrials,
Stimuliandapparatus.Emotionalstimuliandappara- log-transformedlatencies,andreplacedonesthatwere
tuswerethesameasthosereportedinStudy2.Torepre- 2.5 standard deviations above or below the mean.
sent the taskpertaining to a particular block, a second Latencymeanswereexaminedina2(valence:positive
slideprojectordisplayedtheslide“evaluate”or“experi- vs.negative) × 2 (arousal: low vs.high)× 2 (judgment:
ence” during the relevant trials. evaluation vs. feeling) ANOVA.
Noneofthemaineffectsweresignificant,Fs<1.Also,
Procedures.Participantswereinformedthattheywould
× ×
thethree-wayValence Arousal Judgmentinteraction
(a) evaluate slides, judging whether they were good or
wasnotsignificant,F(1,19)=1.50,p=.24,indicatingthat
bad, and (b) indicate whether the slide causes more
similarfindingspertainedtotheevaluationtaskasper-
pleasantorunpleasantfeelings.Tomakethetasksseem
×
tainedtothefeelingjudgmenttask.Two-wayValence
somewhatdistinct,weinformedparticipantsthatevalua-
×
Judgment, F(1,19)=2.26,p=.15,andArousal Judg-
tionisprimarilybasedonstimulusqualitiesratherthan
ment, F < 1, interactions were similarly not significant.
experience, whereas feelings are primarily based on
×
However,asexpected,therewasasignificantValence
experienceratherthanonstimulusqualities.Eachslide
Arousalinteraction,F(1,19)=13.86,p=.00,whichalso
was paired with each task exactly once. Trials were
blockedsuchthatparticipantsperformedeachtaskfor was replicated within the evaluation task, F(1, 19) =
10 trials in a row. 4.43, p = .05, and the feeling judgment task, F(1, 19) =
17.00, p = .00, considered separately. Figure 2, which
When participants were ready to make a judgment,
reports means collapsed across task, indicates that the
theypressedthe1keyontheresponsebox.Whenthe1
×
Valence Arousal interaction was parallel to that
key was pressed, the emotional slide was removed and
reportedinStudy 2.4
theparticipantspressedthe2key(“bad”or“unpleasant
feelings”)orthe3key(“good”or“pleasantfeelings”)to
Discussion
make their ratings. Under these procedures, time to
press the 1 key becomes the important index of judg- ApotentialconcernwithStudies2and3isthatpeople
ment time (see Robinson & Clore, 2002). This said, mayhesitatetoevaluatepositivestimuliaspositiveifthey
effectswerecomparablewhenlatencytopressthe1key produce states related to lust, excitement, or joy.
was added to latency to make the rating. Althoughsuchstatesarenotsociallyundesirableinany
fundamental way, they may be seen as instrumentally
Results
undesirableinthattheyareperceivedasinterferingwith
Asexpected,nofactorsinfluencedtimetopressthe2 performanceefficiencywithineverydaylifetasks(Erber
or3key,ps>.15.Wethereforefocusedontimetopress &Erber,2000).Totheextentthatthisistrue,theindivid-
Robinson et al. / VALENCE-AROUSAL INTERACTIONS 1477
ualmaychoosetorecognizeandreportsuchstatesonly 484
whentheyareundeniable.Althoughwedonotfeelthat
suchresponse-relatedexplanationsforourfindingsare
s
legitimate, we also recognize that every public evalua- nd
o 480
c
tion carries some potential for scrutiny from others. e
s
Ifso,itseemsdesirabletoshowthatthesameValence× Milli
Arousalinteractioncharacterizesperformanceontasks n
inwhichthereisnopublicevaluationatall.Tothisend, me i 476
Ti
wedesigned Studies4and5toexamine theincidental on Low Arousal
consequences ofevaluation.Participantswere askedto ati High Arousal
n
evaluate the slides, ostensibly in preparation for later crimi 472
questions. Following each slide, they also engaged in a s
Di
second,neutraltask.Ifevaluationsaredelayedbecause ot
D
the participant is encountering difficulty in deciding n
a 468
e
whetheraslideispositiveornegative,suchevaluativedif- M
ficulties should be reflected in secondary task perfor-
mance(Pashleretal.,2000).Onthebasisofthisreason-
464
ing, we hypothesized that latencies on a neutral
Negative Positive
×
secondarytaskwouldexhibitasimilarValence Arousal
interaction as obtained in prior studies. Figure 3 Dotdiscriminationlatenciesbyvalenceandarousal,Study4.
STUDY 4
presentation (see Robinson & Compton, 2004, for
Method related studies).Fifthandfinally,thetrialended when
the participant pressed the 1 key on the button box if
Participants. Participants were 42 undergraduates
onedotwaspresentedorhitthe2keyonthebuttonbox
from North Dakota State University.
if two dots were presented. Following this response,
Stimuli and apparatus. The emotional stimuli and there was a 5 s interval before the next trial began.
apparatus were the same as used in Study 2. A second
Results
slide projector was equipped to display dot stimuli
requiring a response.
Inaccurate responses were deleted and times were
Procedures.Participantsweretoldthattheyhadtwopri- log-transformed. Subsequently, we replaced latencies
marytasks.First,theyshouldattendtoandevaluateemo- 2.5standarddeviationsaboveandbelowthegrandmean
tionalslidesthatwerepresentedatthestartofeachtrial withthesecutoffvalues.Meanswerecomputedforeach
×
becausetheymaybeaskedabouttheslideslaterinthe cellofthe2(valenceofprime:positivevs.negative) 2
×
study. Second, they were instructed to determine, as (arousalofprime:lowvs.high) 2(visualfield:leftvs.
quickly and accurately as possible, whether one or two right)design.Meanswerethenanalyzedinarepeated-
dots were presented following each emotional slide. measures ANOVA. All effects were not significant, ps >
×
This dot discrimination task provided the latency data .10, except for one. Specifically, the Valence Arousal
for the study. interaction was significant, F(1, 41) = 10.01, p = .00. As
Figure3shows,theinteractionreplicatesthosereported
The56slideswere showntwiceeach,eachtimeina
inthepriorstudies,butdoessowithinthecontextofan
different random sequence. Target slides consisted of
incidental evaluation paradigm.
oneortwodotspresentedtoeithertheleftorrightvisual
field.Dotslideswerepresentedlaterallyatadistanceof
Discussion
approximately2.5degreesvisualangleleftorrightoffix-
ation. Dot slides were randomly paired with emotional ThefindingsfromStudy4addtoourconfidencein
slides. the valence-arousal conflict theory. When an environ-
Each trial had the following sequence. First, a ran- mental stimulus is arousing, it is apparently judged,
domly selected emotional slide was presented for 1 s. preattentively,tobepotentiallythreatening.Thisfacili-
Second, the slide disappeared followed by a 500 ms tates negative evaluations but inhibits positive evalua-
blankinterval.Third,alaserfixationwaspresentedfor tions. Although such a conflict is eventually resolved,
200ms.Fourth,adotslidewaspresented800msfollow- sucharesolutiontakestime,thereforeslowingconcur-
ing the offset of the laser fixation. Dot slides were pre- rent performance. Study 5 sought to replicate Study 4,
sentedfor100ms,aprocedurethatensureslateralized again using an incidental evaluation paradigm. How-
1478 PERSONALITY AND SOCIAL PSYCHOLOGY BULLETIN
ever, we replaced the dot discrimination task with a 390
motor task.
STUDY 5
s
nd 380
Method o
c
e
s
Participants. Participants were 15 undergraduates Milli
from the University of Illinois. e in
m 370 Low Arousal
Stimuliandapparatus.Stimuliwerethesame56emo- Ti High Arousal
nt
tionalslidesusedinpriorstudies.Hardwareforthestudy e
m
wasalsothesame,includingtheEktaproprojectorand ve
o
M
thecustom-madebuttonbox.However,theparticipant n
a 360
sat in a chair without a desk portion. In addition, the e
M
chairwasplacedclosertotheslidescreensothatpartici-
pants could easily reach it. Finally, the button box was
affixedtotheslidescreensuchthatitslocationwasinthe
middle, both horizontally and vertically, of projected 350
Negative Positive
images.
Figure 4 Motor latencies by valence and arousal, Study 5.
Procedures.Participantswereinformedthattherewere
twotasks.One,theyshouldevaluatetheemotionalslides
astheywerepresented.Andtwo,theyshouldperforman
fasterinthebackwarddirection(M=155ms)thaninthe
indicatedmovementasquicklyaspossibleuponasignal
forwarddirection(M=578ms).Inaddition,therewasa
from the laser pen. During half of the blocks, partici-
×
Valence Arousal interaction, F(1, 14) = 5.24, p = .04.
pants began each trial with their right hands on their
TherelevantmeansareshowninFigure4.Noneofthe
right knees. With a signal from a laser pen, they were
other effects were significant, allps > .50.5
instructed to reach up as quickly as possible and press
anyofthefourkeysonthebuttonbox.Duringtheother Discussion
halfoftheblocks,participantsbeganeachtrialwiththeir
index fingers depressing the 1 key of the button box. TheresultsofStudies4and5,similartotheresultsof
With a signal from a laser pen, they were instructed to Studies2and3,cannotbereconciledwiththeviewthat
remove their fingers from the button as quickly as valence and arousal are completely independent in
possible. evaluativeprocessing,andtheycannotsupporttheview
that arousal merely reflects activation of approach or
Emotionalslideswereshowntwiceeach,eachtimein
avoidance systems. Instead, our effects are compatible
a different random order. Movement direction—press
withproposalsthatintensestimuli,whichelicittheori-
orrelease—wasrandomlypairedwithslides.Trialswere
enting response (McCurdy, 1950), are categorized as
blocked such that participants performed the same
aversiveduringearlystagesofencoding(Zajonc,1998).
movementforseventrialsinarow.Trialsconsistedofthe
Beforeturningtotheissueofwhetherourresultsare
following sequence. First, a picture appeared. Second,
uniquetopictorialstimuli,wesoughttoconductameta-
therewasa2-sdelayuntilalaserpenfireda150-mssig-
analysisofthefindingsreportedinStudies1through5.
nal. This signal was the impetus for performing the
Therelevantanalysesservetwopurposes.One,withthis
movement. Third, the participant performed the
new dataset,we were able to control for identification
relevant action.
latencies(collectedinStudy1).Andtwo,wecouldtreat
Results valenceandarousalinadimensional,ratherthancate-
gorical,manner.Forthesakeofconvenience,wereferto
Movement times were log-transformed to normalize
this meta-analysis as Study 6.
the distribution. For each movement separately, we
replaced scores that were 2.5 standard deviations from
the grand latency mean for that movement direction. STUDY 6
×
Latency means were examined in a 2 (valence) 2
× Method
(arousal) 2(movementdirection)repeated-measures
ANOVA. There was a main effect for movement direc- Wecreatedanewdataset,withslide(N=56)asthe
tion,F(1,14)=324.69,p=.00,suchthatmovementswere unitofanalysis.Increatingthisdataset,wefirstaddedz
Robinson et al. / VALENCE-AROUSAL INTERACTIONS 1479
scored norms related to valence and arousal and their TABLE 1: Multiple Regression Predictors of Evaluation Latencies
and Incidental Motor Performance, Study 6
interaction,basedontheLangetal.(1999)means.We
thenaddedlog-transformedidentificationmeansfrom Predictor
Study1.Suchidentificationmeanswerecomputedsepa-
Dependent Measure Identification Valence Arousal Interaction
rately for each slide, averaged across all of the partici-
pants in that study. Finally, we added log-transformed
Evaluation time, Study 2 .44* .09 .00 .45*
evaluationlatenciesforeachslide,whichwerecollected Evaluation time, Study 3 .55* .21 .09 .25*
inStudies2and3,aswellaslog-transformedincidental Feeling time, Study 3 .41* –.02 .02 .31*
Dot discrimination, Study 4 .23 –.08 .17 .29*
performancemeasures,whichwerecollectedinStudies
Motor behavior, Study 5 .26 –.02 .03 .27*
4 and 5.
NOTE:Withslide(N=56)asunitofanalysis,weperformedfivemulti-
Results pleregressions,oneforeachofthedependentmeasuresreportedin
thistable.Predictorsweresimultaneouslyentered.Depictedintheta-
Inaninitialmultipleregression,weenteredzscores ble are standardized Betas resulting from these regressions.
*p< .05.
forvalence,arousal,andtheirinteractioninaregression
predictingidentificationlatencies.Theregressionindi-
catedthatthemorearousingtheslide,thefasteritsmain
actionwouldgeneralizetotheevaluationofemotional
themewasidentified,t=–3.20,p=.00.Therewasalsoa
words. We sought to investigate this issue in Study 7.
marginal hint in this regression that slides higher in
Method
valence were identified more quickly than were slides
lowerinvalence,t=–1.94,p=.06.However,theinterac- Participants. Participants were 28 undergraduates
tionbetweenvalenceandarousalwasnotsignificant,t< from North Dakota State University.
1. This regression confirms that the interactions
Stimuli. Word stimuli were chosen on the basis of
reported in this article are not due to the visual
normscompiledbyBradleyandLang(1999).Theinves-
complexity of the slides.
tigatorsaskedraterstousepictorialscalestodetermine
Wenextsoughttoexaminetheinteractiveeffectsof
thevalence(1=unpleasant,9=pleasant)andarousal(1
valence and arousal in predicting the dependent mea-
=low,9=high)ofeachofthewords.BradleyandLang
sures collected in Studies 2 through 5. In these regres-
also obtained word frequency norms from Kucera and
sions,wesimultaneouslyenteredthefollowingterms:z
Francis (1967).
scoredidentification,zscoredvalence,zscoredarousal,
InselectingstimulifromtheBradleyandLang(1999)
andatermtorepresent thevalencebyarousalinterac-
norms, we sought to satisfy multiple constraints. First,
tion.Theresults,intermsofBetacoefficientsandsignifi-
lowandhigharousalwordsshouldbeequalinvalence.
cance, are reported in Table 1.
Second, the extremity of negative and positive words
Inthestudies(2and3)involvingexplicitevaluations,
should be equal. Third, negative and positive words
identificationlatencieswereapositivepredictorofper-
shouldbeequalinarousal.Andfourth,allfourcatego-
formance. In other words, knowing what something is riesofwordsshouldbeequalinwordfrequency.Acare-
(i.e.,identification) is an important predictor of being fuliterativeprocesssatisfiedalloftheseconstraintsand
abletodeterminewhetheritisgoodorbad(i.e.,evalua- produced 35 words per category.6
tion). Identification latencies also positively predicted To examine the effectiveness of our selection
incidentalmotorperformance(Studies4and5),butnot procedures, we created dummy codes to represent
×
significantly so. The Valence Arousal interaction was valence(–1=negative,+1=positive)andarousal(–1=
significant in all five regressions. In other words, the low, +1 = high) selection criteria. We then used these
morearousingtheslide,thefaster(implicitorexplicit) dummycodestoexaminetheoriginalratingnormspro-
evaluation latencies were for negative slides, but the videdbyBradleyandLang(1999).Theseanalyseswere
slower they were for positive slides. conducted with word (N = 140) as the unit of analysis
within repeated-measures ANOVAs. With valence rat-
ingsasthedependentmeasure,valencehadasignificant
STUDY 7
effect (Ms = 3.27 and 6.73 for negative and positive
Althoughtheuseofaconstantsetofstimulihasbene- words,respectively),F(1,138)=3187.57,p=.00,whereas
×
fits, as the meta-analysis shows, this stimulus selection arousalandtheValence Arousalinteractiondidnot,
procedurealsomayhavecosts.Specifically,becausewe Fs<1.Withextremityasthedependentmeasure,none
onlyusedemotionalslidesasstimuliinthepriorstudies, of the factors were significant predictors, Fs < 1. With
×
itisunknownwhetherasimilarValence Arousalinter- arousalratingsasthedependentmeasure,arousalhada
1480 PERSONALITY AND SOCIAL PSYCHOLOGY BULLETIN
significant effect (Ms = 4.23 and 5.81 for low and high 920
arousalwords,respectively),F(1,138)=349.04,p=.00,
×
whereas valence andtheValence Arousal interaction ds 900
n
didnot,Fs<1.Finally,withwordfrequencynormsasthe o
c
e
dependentmeasure,noneofthefactorsweresignificant s
predictors,Fs<1.Insum,thewordselectionprocesswas Milli 880
n
successful. cies i 860 LHoigwh AArroouussaall
Procedures.Thecollection ofthedatawasconducted en
at
inthecontextofsmallgroupsessions.Wordstimuliwere L
nt 840
displayed on computer monitors. Participants were e
m
asked to both identify and evaluate the 140 stimulus dg
u 820
words.Weusedtwocounterbalancingprocedures.First, J
viarandomassignmentprocedures,12ofthe28partici-
800
pantswereassignedtoaconditioninwhichtheymade Negative Positive
identifications before making evaluations; the remain-
Figure 5 Judgment latencies by valence and arousal, Study 7.
ing participants completed the tasks in the opposite
order.Second,intheevaluationtaskitself,wecounterbal-
ancedwhetherthe1keyatthetopofthekeyboard(n =
× ×
Valence TasknortheArousal Taskinteractionswere
13) or the 9 key (n = 15) represented a positive
significant,ps > .05.
evaluation.
×
Of most importance was the significant Valence
Intheidentification task,participantswere asked to
Arousalinteraction,F(1,26)=43.23,p=.00.Theinterac-
pressthespacebarassoonastheycouldidentifytheword
tionwascross-overinnature.AsshowninFigure5,par-
in question. In the evaluation task, participants were
ticipantswerefastertoprocess(i.e.,identifyorevaluate)
askedtoevaluatethewordbypressingeitherthe1or9
negativewordswhentheywerehighinarousal;bycon-
keyatthetopofthekeyboard.Inneitherofthetasksdid
trast,theywerefastertoprocesspositivewordswhenthey
we provide accuracy feedback. Words were randomly
were low in arousal. Finally, there was some indication
selected foreachparticipantforeachtask.Trialproce- ×
that the Valence Arousal interaction was more pro-
duresinvolvedthepresentationofasinglewordonthe
nouncedfortheevaluationtaskthanfortheidentifica-
computer screen. The timing began with the appear- × ×
tion task, but the Valence Arousal Task interaction
ance of the word and ended with a relevant response.
wasnotsignificant,F(1,26)=2.83,p=.10.Wesoughtto
Afteraresponse,therewasa500-msdelayuntilthenext
supplement these results with additional results involv-
word was presented.
ing word as the unit of analysis.7
Results
Word-levelanalyses.Giventhattaskorderdidnotaffect
Participant-levelanalyses.Priortoconductingparticipant- thepatternofresultsinanysubstantialway,meanswere
levelanalyses,wedeletedtheinaccurateevaluations.We collapsedacrossthisvariable.Wecreatedanewdataset
thenlog-transformedlatenciesandreplacedtimesthat withword(N=140)astheunitofanalysis.Foreachword
were2.5SDsbeloworabovethegrandlatencymean,sep- separately,weenteredindependentvariablesrelatedtoz
aratelyforeachtask.Log-latencymeanswerethenaver- scoredvalenceandarousal.Bothwerecontinuousmea-
×
agedwithincellsofa2(valence:negativevs.positive) 2 sures and both were based on the Bradley and Lang
×
(arousal:lowvs.high) 2(task:identificationvs.evalua- (1999) norms. We also entered frequency norms from
tion) within-subject design. We also retained the KuceraandFrancis(1967).Finally,foreachwordsepa-
between-subjects factor of task order in the relevant rately, we obtained identification and evaluation laten-
ANOVA.Significanteffectswillbereportedintermsof cies averaged across participants.
milliseconds. We initially performed a multiple regression
Therewerenoeffectsfortaskorder,ps>.20,savean designedtoreplicatetheanalysesreportedabove,here
× ×
Arousal Task TaskOrderinteraction,F(1,26)=6.87, withwordastheunitofanalysis.Inthisregression,log-
p=.01,thatwasneitherpredictednorinformative.How- transformedevaluationlatencieswerepredictedonthe
ever,thereweremaineffectsforvalence(Ms=904and basis of z scores related to valence, arousal, and their
861 ms for negative and positive words, respectively), interaction.Themaineffectforvalencewasnotsignifi-
F(1,26)=8.99,p=.01,arousal(Ms=866and900msfor cant, t = –1.11, p = .27. However, we replicated a main
lowandhigharousalwords,respectively),andtask(Ms= effectforthearousalleveloftheword,t=2.60,p=.01,as
×
789and977msforidentificationandevaluationtasks, wellastheValence Arousalinteraction,t=3.76,p=.00.
respectively), F(1, 26) = 10.35, p = .00. Neither the Estimatedmeans(+or–1SD),obtainedfromaregres-
Robinson et al. / VALENCE-AROUSAL INTERACTIONS 1481
sion equation, revealed that the interaction was quite examining evaluation latencies. As expected, we
×
parallel to that reported above. observed a significantValence Arousal interaction in
Inasecondmultipleregression,wesoughttotestthe Studies 2 and 3, such that negative/high arousal and
robustness of the interaction with other factors con- positive/lowarousalslideswereevaluatedmorequickly
trolled. In this equation, the dependent measure con- than their valence-matched counterparts. Study 7
sistedoflog-transformedevaluationlatencies.Predictor showedthataparallelpatternisapparentwhenpeople
variables consisted of identification latencies and word evaluate word stimuli.
frequency values in addition to z scores related to Althoughthepossibilityseemssomewhatunlikelyto
valence, arousal, and their interaction. As in Study 6, us,particularlyaslatencieswereinvolvedandresponses
identification latencies were a significant predictor of weremadeinthedark,itcouldbethecasethatpartici-
evaluationlatencies,t=3.72,p=.00.However,frequency pantsfeltsomewhathesitantinreportingpositiveevalua-
andvalencemaineffectswerenotsignificant,ts<1.Rep- tionsofslideswithinthepositive/higharousalcategory.
licating the analyses reported above, there was a main Similar concerns,however,cannotbe offeredconcern-
effectforarousal,t=2.59,p=.01,aswellasthehypothe- ingthecovertevaluationfindingsreportedinStudies4
×
sized Valence Arousal interaction,t= 3.05,p= .00. and5.Inthesestudies,participantsmadenopubliceval-
uationsoftheslides.Rather,slideevaluationdifficulties
Discussion
wereinferredfromperformanceonaneutralsecondary
×
Pictorialandwordstimulidifferinanumberofways, task. Because we observed the same Valence Arousal
suchasimageabilityandcapacitytoinducestrongphysi- interactionsinthesestudies,weconcludethatevaluative
ologicalactivity(Langetal.,1997).Inaddition,accessto processing was genuinely delayed in the case of
word meaning obviously relies to a greater degree on negative/low arousal and positive/high arousal slides.
symboliclinguisticactivity.Onemightthereforeexpect It is important to note that effects pertaining to the
×
some differences in the processing of pictures and Valence Arousalinteractionwereapparentbothwith
words.Indeed,whereashigharousalpictureswereiden- participantastheunitofanalysis(Studies2-5and7)and
tified more quickly (Study 1), high arousal words were with stimulus as the unit of analysis (Studies 6 and 7).
identified more slowly (Study 7). A possible reason for Thelatterfindingsaddtothestrengthsofthearticlein
thediscrepancyisthathigharousalstimuliboth(a)acti- thatcognitivepsychologistsareoftenmoreimpressedby
vatetherighthemisphere(Robinson&Compton,2004) findingsthatreplicateacrossbothparticipant-leveland
and(b)therighthemisphereisbetteratpictureprocess- item-level analyses. More important, however, the rele-
×
ing, whereas the left hemisphere is better at word pro- vant analyses were able to confirm that the Valence
cessing(Hellige,1990).However,asystematiccompari- Arousalinteractionwasreplicatedwithinthecontextof
son of word and picture processing was clearly not the continuousvariationsinvalenceandarousal,evenwith
focus of the present studies and therefore this stimulus identification processes controlled.
interpretation should be viewed with caution.
Stimulus Arousal and Negative Connotation
Moreimportantthantheidentification-relateddiffer-
encesweretheresultsthatreplicatedacrossbothpicture
Thepresentresults,althoughnovel,dovetailtosome
×
andwordstimulussets.Specifically,therewasaValence
extentwithpriorresults.First,Langetal.(1997)report
Arousal interaction such that high arousal facilitated
that it is extremely difficult to find strongly negative
evaluationsofnegativestimuliandinterferedwithevalu-
slides that are low in emotional arousal. This is also
ationsofpositivestimuli.Giventheparallelinteractions
trueofemotionalwordsandsounds(Langetal.,1997).
involvingtheevaluationofpicturesandwords,thecur-
One inference from these prior norming data is that
rentfindingscannotbeduetothenatureoftheparticu-
valence and arousal covary in the real world such that
larstimuliinvolved.Rather,abroaderprincipleappears
negativestimulitendtobearousing.Ifthisistrue,itis
to be operative.
notcompletelyerroneoustoinferthat,pendingfurther
analysis, an arousing stimulus is likely to be a negative
GENERAL DISCUSSION one. Two, Lang (1995) has reported some preliminary
dataindicatingthatnotallnegativeslideselicitthesame
Major Findings
level of affect-modulated startle responses. Specifically,
Insevenstudies,wefoundconsistentsupportforthe stimulithatarenegativeandlowinarousal,suchasthose
ideathatstimulusvalenceandarousalinteractivelyinflu- related to pity, tend to elicit smaller amplitude startle
enceevaluativeprocessing.Study1firstestablishedthat responses thancould beexpected onthebasisoftheir
×
therewasnoValence Arousalinteractioninvolvingthe valence.Bycontrast,stimulithatarenegativeandhighin
visual complexity of slides. This was an important null arousal,suchasthoserelatedtofear,tendtoelicitlarger
result because it paved the way for subsequent studies amplitudestartleresponsesthancouldbeexpectedon