Journal of Experimental Social Psychology 41 (2005) 574–588

www.elsevier.com/locate/jesp

Mood eVects on eyewitness memory: AVective inXuences on susceptibility to misinformation �

Joseph P. Forgas a,¤, Simon M. Laham a, Patrick T. Vargas b

a School of Psychology, University of New South Wales, Sydney 2052, Australiab University of Illinois, Urbana-Champaign, USA

Received 11 February 2004; revised 7 October 2004Available online 3 March 2005

Abstract

Does mood inXuence the accuracy of eyewitness recollections, and people’s susceptibility to misleading information in particular?Based on recent aVect-cognition theories and research on eyewitness memory, three experiments predicted and found that positiveaVect promoted, and negative aVect inhibited the incorporation of misleading information into eyewitness memories. This eVect wasobtained for both positive and negative events (Experiment 1), and for recorded as well as real-life incidents (Experiment 2). Partici-pants had no meta-cognitive awareness of these mood eVects, and aVect-control instructions were ineVective in preventing them(Experiment 3). The cognitive mechanisms responsible for mood eVects on eyewitness memories are discussed, and the implicationsof these Wndings for everyday memories, forensic practice and for current aVect/cognition theorizing are considered. 2005 Elsevier Inc. All rights reserved.

Keywords: AVect and cognition; Eyewitness memory; Social cognition

Accurately remembering personally witnessed eventsis one of the more important cognitive tasks we face ineveryday social life. Most people implicitly assume thattheir eyewitness recollections are generally accurate andreliable, and eyewitness memories are often accordedspecial evidentiary status by judicial and forensic organi-zations. This assumption of eyewitness accuracy, how-ever, may not always be justiWed. Research documentsnumerous instances of inaccuracies in eyewitness testi-mony and constructive memory errors (e.g., Loftus,1979; Malpass, 1996; Penrod & Cutler, 1996; Schooler &Loftus, 1993; Wells & Loftus, 2003).

� This project was supported by a Special Investigator Award fromthe Australian Research Council and by a Fellowship from the Rocke-feller Foundation. The contribution of Stephanie Moylan is gratefullyacknowledged.

¤ Corresponding author.E-mail address: jp.forgas@unsw.edu.au (J.P. Forgas).URL: www.psy.unsw.edu.au/Users/Jforgas/jforgas.htm (J.P. Forgas).

0022-1031/$ - see front matter 2005 Elsevier Inc. All rights reserved.doi:10.1016/j.jesp.2004.11.005

Surprisingly, relatively little attention has been paidto the inXuence of mild mood states, as distinct frommore intense emotions or arousal states on the accuracyof eyewitness memories (Brown, 2003; Wells & Olsen,2003). Moods may inXuence memory at each of the threestages of the eyewitness process: when the event is Wrstwitnessed (encoding stage), later when potentially mis-leading information is encountered (post-event stage),and Wnally, when the information is retrieved and judg-ments are made (retrieval stage). This paper will focus onthe post-event stage, exploring the inXuence of moods onthe incorporation of false information into subsequentrecollections. Drawing on past research on eyewitnessmemory and recent work on mood eVects on informa-tion processing, these three experiments explore the pos-sibility that good moods can accentuate, and bad moodscan inhibit the incorporation of subsequent misleadinginformation into eyewitness reports, consistent with theinformation processing consequences of these aVectivestates. In particular, the more externally oriented and

J.P. Forgas et al. / Journal of Experimental Social Psychology 41 (2005) 574–588 575

systematic information processing style elicited by nega-tive moods may reduce the likelihood of misleadinginformation being incorporated into the eyewitnessrecollections, and thus improve the accuracy of memo-ries. In contrast, the more constructive and assimilativeprocessing style associated with positive moods mayimpair eyewitness accuracy by increasing the likelihoodthat subsequent misleading information will be uncriti-cally incorporated into eyewitness memories (Fiedler,Asbeck, & Nickel, 1991).

Misleading information and eyewitness accuracy

Social perception and judgments require highly con-structive strategies to select, recall and interpret informa-tion about people and events (Fiedler et al., 1991; Loftus,1979; Wells & Loftus, 2003). In a series of highly inXuen-tial experiments, Elizabeth Loftus showed that peopleare easily misled and report erroneous memories as aresult of being exposed to incorrect post-event informa-tion (Loftus, 1979). One common way that eyewitnessescan be exposed to post-event misinformation is by beingasked (mis)leading questions, that is, questions that con-tain information about the observed episode that wasnot in fact part of the original event. For example, a mis-leading question suggesting that there was a ‘give way’sign rather than a ‘stop’ sign in a traYc accident scenecan produce a signiWcant increase in incorrect memoriessuggesting that the ‘give way’ sign was part of the origi-nal scene (Loftus, 1979).

In a similar way, a question like “How fast was thewhite sports car going when it passed the barn whiletraveling along the country road?” produces anincreased tendency by eyewitnesses to report havingseen a barn, even though this was not part of the origi-nal scene (Loftus, 1979). Even such subtle post-eventclues as a change from the indeWnite to the deWnitivearticle (a to the) can produce memory biases. Partici-pants who were asked “Did you see the broken head-light?” were more likely to report seeing it later on thandid those asked “Did you see a broken headlight?”(Loftus, 1979).

This memory-biasing eVect of post-event informationis known as the misinformation eVect (Schooler & Loftus,1993). Some explanations of the eVect suggest that theoriginal memory trace becomes overwritten by the mis-information received later on (Loftus, 1979). Other theo-ries propose however that subsequent misinformationdoes not eliminate the original memory, but simplyinterferes with its accessibility and retrieval (Bekerian &Bowers, 1983). It was also suggested that the eVect maydepend on the strength of the memory trace; strongmemories are more likely to resist misleading sugges-tions than weak memories (Pezdek & Roe, 1995; Reyna& Lloyd, 1997).

Various social factors may also inXuence susceptibil-ity to the misinformation eVect, such as the age and sta-tus of the person making the misleading suggestion (Ceci& Bruck, 1993). When considering these cognitive andsocial inXuences jointly, it seems that the way the personprocesses questions containing misleading informationshould have a critical inXuence on the extent to whichthe false information will be rejected, or incorporatedinto eyewitness memories. More constructive and assim-ilative thinking may increase the tendency to uncriticallyincorporate false details into memory (Fiedler et al.,1991), while externally oriented accommodative process-ing may reduce the misinformation eVect. Interestingly,recent evidence suggests that mild mood states can infact promote exactly these kinds of diVerences in pro-cessing style, promoting assimilative or accommodativethinking (Bless, 2000; Fiedler, 2001; Fiedler & Bless,2001), and may well inXuence people’s susceptibility tothe misinformation eVect.

AVect and eyewitness memories

Although the profound inXuence of aVect on the waypeople think and behave has long been recognized,mood eVects on eyewitness memory have received rela-tively little attention (cf. Eich & Schooler, 2000; Schooler& Eich, 2000). All things being equal, aVective statesseem to facilitate the recall and use of aVect-congruentrather than incongruent information, and informationencountered in a matching rather than a non-matchingmood state (Bower & Forgas, 2001; Clore, Gasper, &Garvin, 2001; Eich & Macauley, 2000). Indeed, suchmood-induced memory biases can even inXuence theway people think about highly familiar and involvingevents, such as incidents in their intimate relationships(Forgas, 1994). By facilitating access to mood-congruentinformation in memory, aVective states can exert apowerful informational inXuence on the kind of evidencepeople access when encoding and recalling the details ofcomplex events (Berkowitz, JaVee, Jo, & Troccoli, 2000;Bower & Forgas, 2001; Eich & Macauley, 2000; Fiedler,2001; Forgas, 1995, 2002). Alternatively, mood states canalso serve as heuristic cues inXuencing evaluative judg-ments in a mood-congruent direction (Clore et al., 2001;Martin & Clore, 2001; Schwarz & Bless, 1991). Becausememory is often better for mood-congruent rather thanmood-incongruent information, we may expect that con-gruence between transient mood and the aVective toneof a witnessed event may oVer some defence against themisinformation eVect and thus contribute to theaccuracy of eyewitness memories, a possibility we inves-tigated in Experiment 1.

In addition to such informational eVects, moods canalso exert a signiWcant processing eVect on how peopledeal with social information. Research on decision-

576 J.P. Forgas et al. / Journal of Experimental Social Psychology 41 (2005) 574–588

making, persuasion, and creativity (Bless, 2000; Bless,Mackie, & Schwarz, 1996; Fiedler, 2001) suggests thatdiVerent mood states produce reliable diVerences ininformation processing style.

Several studies initially suggested that positive moodsmay lead to less eVortful and systematic processing strat-egies (Bodenhausen, 1993; Clark & Isen, 1982). In con-trast, negative moods were thought to facilitate morecareful, vigilant and systematic processing (Schwarz,1990; Schwarz & Bless, 1991). However dysphoria canalso reduce attentional resources (Ellis & Ashbrook,1988). A number of competing explanations for thesemood-induced processing diVerences were proposed.Functional explanations suggested that good mood sig-nals that the situation is favorable and little processingeVort is required, whereas bad moods recruit more sys-tematic and vigilant processing (Schwarz, 1990; Schwarz& Bless, 1991). Other theorists posited a motivationalexplanation, as happy people may try to preserve theirgood mood by avoiding cognitive eVort (mood mainte-nance), and dysphoric individuals increase cognitiveeVort to improve their aversive mood state (moodrepair) (Clark & Isen, 1982). Finally, processing capacitymay also be impaired by positive moods according tosome studies (Stroessner & Mackie, 1992). However,negative moods may also reduce processing capacity(Ellis & Ashbrook, 1988), and some researchers reportthat positive mood may actually enhance rather thaninhibit performance on certain tasks (Bless et al., 1996).

Recently, Fielder and Bless (2001; Fielder, 2001;Bless, 2000) developed a more comprehensive explana-tion of aVective inXuences on information processingstyle. They noted that mood-induced processing diVer-ences are unlikely to be simply due to changes in cogni-tive eVort, as performance on secondary tasks remainsunimpaired by positive mood (Bless, 2000). Instead ofmerely inXuencing processing eVort, these authors sug-gest that diVerent moods actually induce qualitativelydiVerent styles of processing. Negative moods call foraccommodative processing, focused on the actual detailsof the external world. In contrast, positive moods signala benign, predictable environment, and induce assimila-tive and constructive processing where the individualrelies on existing knowledge and heuristic, schematicthinking to perform a task (Bless, 2000; Fiedler, 2001).Recent integrative aVect/cognition theories such as theAVect Infusion Model (AIM; Forgas, 1995, 2002) alsopredict that positive and negative moods should producesuch a diVerence in how novel stimulus information isprocessed (Forgas, 1998a, 2002).

Applying these theoretical accounts to eyewitnessaccuracy, we may expect that people in a negative moodwill be more attentive to situational details and so be lessinXuenced by misleading information than are people ina positive mood. Negative aVect should thus improveeyewitness accuracy by reducing the likelihood that mis-

leading information will be incorporated into eyewitnessrecollections. In contrast, a more constructive and assim-ilative processing style associated with positive aVectmay exacerbate the constructive errors associated withmisinformation eVects. Consistent with this prediction,Fiedler et al. (1991) found that people experiencing apositive mood are more likely to engage in constructiveprocessing and are more inXuenced by prior primingmanipulations when forming judgments about people.Indeed, Fiedler et al. (1991) explicitly argue for the needto examine “the mediating role of mood in eyewitnesstestimony” (p. 376), the main objective of the presentexperiments. Another relevant series of experiments byForgas (1998a) found that people experiencing negativeaVect were less likely to commit the fundamental attribu-tion error, whereas those in a positive mood made moreconstructive internal attributions, apparently ignoringinformation about external constraints.

Aims and predictions

Thus, the aim of this paper is to explore the inXuenceof transient mood states on eyewitness accuracy, and theincorporation of misleading details into eyewitnessmemories in particular. Based on previous theories andresearch, we expected that those in a positive moodshould engage in more constructive, assimilative and lessexternally focused processing, and should be more likelyto incorporate misleading information into their eyewit-ness memories. Negative mood, in contrast, shouldfacilitate a more externally oriented, piecemeal and bot-tom–up processing style, reducing the likelihood thatmisleading information will inXuence eyewitness recol-lections. Experiments 2 and 3 also explored the level ofmeta-cognitive awareness and conWdence people had inthe accuracy of their eyewitness reports. The inXuence ofaVect-control instructions and individual diVerence vari-ables (Hosch, 1994; Schooler & Loftus, 1993; Tomes &Katz, 1997) on these eVects was also assessed.

Experiment 1

The Wrst experiment was designed as an initial test tosee if transient moods can indeed inXuence the mistakenincorporation of false information into eyewitnessreports. We predicted that good mood should increase,and bad mood should reduce people’s susceptibility tomisleading information. Further, based on evidence formood-congruent inXuences on memory (Eich & Macau-ley, 2000; Fiedler, 2001; Forgas, 2002), this experimentalso investigated the possibility that exposure to mis-leading information may have less of an eVect whenmood and the aVective tone of the witnessed event arecongruent rather than incongruent.

J.P. Forgas et al. / Journal of Experimental Social Psychology 41 (2005) 574–588 577

Method

Overview, design, and participantsThe session was introduced as comprising several

unrelated experiments, carried out over a 2-h period.Participants Wrst viewed two A4 size pictures for 1 mineach, showing an image of a complex car crash scene(negative event), and a wedding party scene (positiveevent), and were instructed to ‘look at these pictures as ifyou unexpectedly encountered these events while walk-ing on the street’. After various other activities, about 1 hlater, and allegedly as part of an unrelated study, theyreceived an autobiographical mood induction (they wereasked to re-experience and write about positive, neutraland negative episodes from their own lives). Immediatelyafterwards, participants completed a short questionnaireabout the scenes they saw earlier that either did, or didnot contain misleading information about these events.After a further 45-min interval Wlled with distractiontasks, the accuracy of their eyewitness memory for thescenes was tested. The experiment involves a 3 £ 2 £ 2design, with aVect (happy, control, and sad), the provi-sion of misleading information (present/absent), and theaVective tone of the event witnessed (positive, negative)as the independent variables. Participants were 96 stu-dents (54 female, 42 male) with a median age of 19 whocompleted the experiment as part of their course require-ments.

The target eventsThe target scenes showed either a complex car crash

scene (negative event), featuring a large number of peo-ple, policemen, ambulance personnel and two damagedcars in a complex urban setting, or a wedding celebra-tion (positive event), with a bride, a groom, and a largenumber of guests in a party setting. These images wereselected from some 24 everyday scenes collected fromphoto magazines and the internet, which were pre-ratedby a pilot sample of 16 students for aVective valence,complexity and realism on 7-point scales. The two sceneswere selected as they were most consistently rated ashigh on complexity (M D 5.87 and 6.11; SD D 1.21 and1.30) and realism (M D 6.39 and 5.61; SD D 1.06 and1.26), and were highly and reliably diVerent in terms ofvalence (M D 1.38 and 6.05; SD D .89 and 1.17).

Mood inductionThe mood induction task was described as an unre-

lated study of ‘life events’. Participants were asked to“identify a speciWc social event that has occurred in yourlife that has made you very happy (sad)ƒ imagine thesituation as vividly as you can. Picture the event actuallyhappening to you. Try to experience all the details of thesituation...think through the thoughts that occurred toyouƒ feel the same feelings you feltƒ describe the eventyou remembered as vividly as you can including all the

important details.” In the neutral condition they wereasked to describe their activities while they were gettingready for university this morning. Participants then tookaround 10–12 min to reXect on, and to write down theirpositive or negative experiences. This autobiographicalprocedure was found highly eVective in inducing nega-tive or positive mood states in the past (Forgas, 1995).

The delivery of misleading informationAfter the mood induction, participants completed a

short questionnaire asking four questions each about thecar accident scene, and about the wedding party scene.Each question was prepared in two versions: a directversion containing no misleading information (e.g., ‘Didyou see the overturned car on the roadside?’ ‘Did you seethe police vehicle on the scene?’), and a parallel formcontaining planted, misleading information (set in italicshere: ‘Did you see the overturned car next to the brokenguard rail?’ ‘Did you see the Wreman holding a Wre hose?).Each participant received either the direct form or themisleading form for both scenes in a complete between-subjects design.

Mood validation measureAfter exposure to the misleading questions, a brief

post-experimental questionnaire was administered,designed to validate the eVectiveness of the mood induc-tion. Embedded among several distracter items (e.g., Didyou Wnd the task diYcult? Have you done similar tasksbefore?), participants were also asked to rate their cur-rent mood on seven-point happy–sad and good–badscales.

Measure of eyewitness accuracyAfter performing a series of unrelated distracter

tasks for about 45 min, the measure of eyewitness accu-racy was administered. The questionnaire contained 12true/false questions about each of the two observedscenes. For each scene, four questions evaluated mem-ory for actual, correct details of the scene, four ques-tions tested memory for false, misleading details thatwere only introduced during the previous questioningfollowing the mood manipulation, and four questionsevaluated incorrect details not encountered previously.Based on these responses, three measures were derivedby adding the correct responses in each category: num-ber of correct details recalled, number of misleadingdetails recalled, and number of incorrect detailsrecalled.

DebrieWngThe study concluded with a thorough debrieWng.

There was no evidence of any awareness by partici-pants of the hypotheses and the manipulations. Wetook care to reduce and eliminate any residual moodafter-eVects.

578 J.P. Forgas et al. / Journal of Experimental Social Psychology 41 (2005) 574–588

Results

Mood validationAs participants’ self-rated mood on the seven-point

happy–sad and good–bad scales was highly correlated, asingle combined measure of mood valence was created(Cronbach’s � D .82). An analysis of mood ratings founda highly signiWcant mood eVect F (2,93) D 5.64; p < .01.Participants who thought about a positive life eventwere in a signiWcantly better mood, and those thinkingof a negative event were in a worse mood than did thosewho reXected on a neutral event (M D 2.24, 3.66, 4.98;t (62) D 3.41; p < .01; t (62) D 2.53; p < .01). These resultsconWrm that the aVect induction was highly eVective,and produced signiWcantly diVerent mood states thatendured until well after the delivery of the misleadinginformation (Forgas, 1995, 2002).

Eyewitness accuracyThe three measures of eyewitness memory (recollec-

tion of correct, misleading and incorrect details) for thetwo scenes were Wrst subjected to a 3 £ 2 £ 2MANOVA. Results showed a signiWcant main eVect forexposure to misleading information F (1, 84) D 3.33;p < .05, and a signiWcant interaction between mood andexposure to misleading information, F (2, 84) D 3.95;p < .05 using the Wilks lambda criterion. No other sig-niWcant eVects were found. Further separate analyses ofvariance were conducted on recollections of correct,misleading and incorrect details. We found that neithermood, nor the provision of misleading information, northe valence of the events or their interaction had a sig-niWcant main or interaction eVect on recognition mem-ory for correct and incorrect details. In fact, this is whatwe would expect, as these memories were not directlymanipulated.

As predicted, the provision of misleading informationdid have a signiWcant main eVect on memory for mis-leading details, producing more ‘false alarms’ (recogniz-ing misleading details as part of the original scene),F (1, 84) D 4.86; p < .01. Those exposed to misleadinginformation were signiWcantly more likely subsequentlyto remember such details as part of the original scenesthan were those not exposed to such information, onceagain demonstrating the misinformation eVect (Loftus,1979). Of greatest interest is that there was also a signiW-cant interaction between the presence of misleadinginformation and mood, F (2, 84) D 4.97; p < .01, whichshowed that positive mood increased, and negativemood decreased the tendency to incorporate misleadingdetails into recollections, as illustrated in Fig. 1. In fact,misleading information did not have a signiWcant inXu-ence on ‘false alarms’ by participants experiencing nega-tive moods. In contrast, individuals who experienced apositive or neutral mood while exposed to the mislead-ing information were signiWcantly more likely to incor-

porate these details into their recollections, t (31) D 3.13;p < .01, t (31) D 2.41; p < .05.

Contrary to our predictions, the aVective tone of theobserved event had no main or interaction eVects onmemory. It seems that any memory beneWt produced bymatching the mood of the participants with the aVectivetone of the event was not suYciently large or robust torepresent a signiWcant defense against the incorporationof false details into eyewitness memories.

Signal detection analysisTo provide an integrated analysis of the entire recog-

nition performance, a signal detection analysis based onall responses was carried out. First, the hit rate and falsealarm rate for each participant was calculated and astandard correction was applied when these rates were 0or 1 (Snodgrass & Corwin, 1988). Next, measures of dis-crimination (d�) and bias (C) were calculated using logtransformed data to approximate a normal distributionusing a method suggested by Brophy (1986). Higher val-ues on the discrimination (d�) measure indicate that aparticipant had greater ability to discriminate betweencorrect and incorrect details. Positive values on the bias(C) measure indicate a conservative bias of rejectingdoubtful information as incorrect, while negative valuesindicate a liberal bias of accepting information as cor-rect.

Discrimination (d�) was signiWcantly inXuenced bymood, F (2,90) D 3.99; p < .05. Negative mood at the timeof exposure improved participants’ ability to discrimi-nate between correct and incorrect details compared tothe control group and the positive mood groups,t (63) D 2.12, p < .01; t (63) D 1.79; p < .05 (M D .65, .43,.37), with no diVerence between the neutral and positivegroups. As expected, exposure to false informationimpaired participants ability to discriminate betweencorrect and incorrect details, F (2, 90) D 3.57; p < .05;(M D .54, .35). An analysis of the response bias measure(C) showed no signiWcant mood or exposure eVects,

Fig. 1. The interaction between mood and the presence or absence ofmisleading information on recognition (Experiment 1): positive moodincreased, and negative mood decreased the likelihood that misleadinginformation provided after the event will be incorporated in eye-wit-ness memory.

J.P. Forgas et al. / Journal of Experimental Social Psychology 41 (2005) 574–588 579

F (2, 90) D 1.33; NS; F (2,90) D .86; NS. This is asexpected, as there should be no link between exposure tomisleading information and subsequent changes inresponse bias at the recognition stage. These results areconsistent with the prior analyses, and conWrm thatlooking at recognition performance as a whole, positiveor negative mood while receiving misleading informa-tion did have a signiWcant inXuence on participants’ abil-ity to discriminate between correct and incorrect detailsthey had seen, with negative mood improving, and posi-tive mood impairing memory performance.

These Wndings support our main hypothesis that posi-tive moods promote, and negative moods inhibit thekind of constructive, assimilative information processingstyle that facilitates the incorporation of misleadingdetails into eyewitness accounts (Fiedler et al., 1991).Because one of the greatest threats to the accuracy ofeyewitness memories is precisely the inadvertent, con-structive incorporation of ‘foreign’ details (Schooler &Loftus, 1993), the empirical demonstration that this ten-dency is facilitated by positive mood has a variety ofimportant theoretical and practical implications. Beforediscussing these issues, however, it is important to repli-cate and extend these Wndings in further experiments;this was the objective of the next study.

Experiment 2

The second experiment was designed as a furtherdemonstration that transient moods can indeed have asigniWcant impact on the accuracy of eyewitness reports.In contrast with Experiment 1, a realistic and more com-plex real-life incident rather than photos was used as thetarget event to be remembered. We also used a diVerentmood induction technique (audio–visual mood induc-tion), to establish the generality and robustness of theseeVects. As most mood induction techniques also produceadditional, cognitive and motivational eVects that mightconfound the results, it is important to use a number ofdiVerent aVect-induction strategies in a related series ofinvestigations to establish the convergent validity of themood eVects. It is only by ‘triangulating’ mood eVectsacross a number of induction methods that we can beconWdent that observed eVects are indeed due to aVectivediVerences (Forgas, 2002).

To gain some insight into the degree of meta-cogni-tive awareness people possess about the accuracy oftheir eyewitness memories, participants in this experi-ment were also asked to indicate their conWdence in theirrecollections. Based on most aVect/cognition theoriesthat assume that aVective inXuences on thinking are sub-conscious (e.g., Fiedler, 2001; Forgas, 2002), and studiesshowing that people have little direct access to theirmental processes (Nisbett & Wilson, 1977), we expectedlittle conscious awareness of the misinformation eVect

and therefore no relationship between conWdence judg-ments and accuracy. However, some theories such asSchwarz and Bless (1991) cognitive tuning model, andMartin, Ward, Achee, and Wyer (1993) ‘mood as input’theory imply that processing decisions may often bebased on high-level inferences informed by mood. If so,people may well become aware of their processing strate-gies, and may subsequently report less conWdence whenin a positive mood. ConWdence ratings may thus indicaterespondents’ level of awareness of the processing conse-quences of moods, and also shed some light on the pro-cessing mechanisms that mediate mood eVects on themisinformation eVect.

Method

Overview, design, and participantsStudents in a lecture theatre witnessed what they

believed was an unexpected 5-min aggressive encounterbetween a lecturer, and a female intruder. One week latereyewitnesses to this episode received a mood induction(viewed short 10-min video-Wlms), and then responded toa brief questionnaire about the episode they had wit-nessed that either did, or did not contain misleadinginformation about the event. After a further 45-min inter-val, the accuracy of their eyewitness memory for theevent was tested. The experiment is based on a 3£2design, with mood (happy, control, and sad), and the pro-vision of misleading information (present/absent) as theindependent variables. Participants were 144 students (83female, 61 male) with a median age of 19 who completedthe experiment as part of their course requirements.

Mood inductionVideotapes were used to induce happy, neutral or sad

moods in subjects, in what was described as a separatestudy to validate Wlms for use in a later study. The use ofvideo Wlms to manipulate mood has been extensivelytried and tested both in laboratory and Weld research,and has been found to produce salient and enduringmoods (Forgas, 2002; Forgas & Moylan, 1987). The 10-min Wlms used included scenes from: (a) a popular Brit-ish comedy series (positive mood); (b) a program onarchitecture (control); and (c) a Wlm dealing with deathfrom cancer (negative mood). After the conclusion of theWlms, a short ‘Wlm assessment questionnaire’ was admin-istered, asking subjects among other things to rate theircurrent mood on seven-point happy–sad and good–badscales embedded among several other distracter ques-tions (the mood validation).

The delivery of misleading informationFollowing the mood induction, participants were

asked to answer four questions about the lecture roomincident. Each question was prepared in two versions: adirect version containing no misleading information

580 J.P. Forgas et al. / Journal of Experimental Social Psychology 41 (2005) 574–588

(e.g., ‘Did you see the lecturer removing his microphoneas the woman approached?’ ‘Can you remember theyoung woman Wddling with her scarf as the lecturerspoke to her?’), and a parallel form containing planted,misleading information (set in italics here: ‘Did you seethe lecturer removing his microphone, as the womanwearing a light jacket moved towards him?’ ‘Can youremember the young woman Wddling with her scarf asthe lecturer gave her something from his wallet?’). Half ofthe participants received the direct form, and halfreceived the misleading form.

Measure of eyewitness accuracyNext participants engaged in a variety of unrelated

distracter tasks for about 45 min, involving the comple-tion of several questionnaires and listening to a mini-lec-ture, and performing some calculations. They thencompleted the eyewitness accuracy questionnaire con-taining 12 true/false questions about the classroom inci-dent. Four questions tested the recollection of actual,correct details of the episode (e.g., ‘The lecturer removedhis microphone as the woman approached’). Four ques-tions tested memory for false, misleading details thatwere planted as part of the earlier manipulation (e.g.,‘The woman was wearing a light jacket.’). Four questiontested for incorrect details that were not part of the origi-nal episode (e.g., ‘The lecturer wore a brown stripedshirt’). Participants were also asked to rate their conW-dence in each of their responses. Based on this informa-tion, three sets of measures were calculated for eachparticipant: number of correct details recognised, num-ber of misleading details recognised, and number ofincorrect details recognised, and the conWdence judg-ments associated with each class of judgments.

DebrieWngThe procedure was concluded by a careful debrieWng.

Questioning revealed no evidence of any awareness ofthe manipulations or suspiciousness about the tasks. Wewere careful to ensure that any residual mood after-eVects were removed.

Results

Mood validationAs mood ratings on the happy–sad and good–bad

scales were highly correlated, the two scales were againcombined to create a single measure of mood quality(Cronbach’s �D .80). A univariate ANOVA of moodratings found a highly signiWcant mood eVectF (2, 141) D 5.16; p < .01. Those in the positive conditionrated their mood as signiWcantly better than did thosewho saw a negative Wlm, t (94) D 4.85; p < .01, and bothexperimental groups were signiWcantly diVerent fromcontrols, t (94) D 3.24; p < .01, t (94) D 2.31; p < .01(M D 6.20, 4.65, 3.32). These results conWrm that the

mood induction was highly eVective in generating reli-ably diVerent mood states, as also found in previousexperiments using this method (Forgas, 1995).

Eyewitness accuracyRecognition scores on the three dependent measures

(recollection of correct, misleading, and incorrect details)were subjected to a 3 £ 2 MANOVA using the Wilkslambda criterion, with mood (happy, neutral, and sad)and the absence or presence of misleading informationas the two independent variables. Once again, exposureto misleading information had a signiWcant main eVect,F (2,138) D 3.15; p < .05, and a signiWcant interactionbetween mood and exposure to misleading informationwas also found, F (2, 138) D 3.67; p < .05. To explore theseeVects, further univariate analyses of variance were car-ried out on recollections of correct, misleading andincorrect details separately. Recognition of correct andincorrect details was not inXuenced by either mood, ormisleading information. As the manipulations were notdesigned to inXuence correct and incorrect recognition,these null eVects are not unexpected.

It is memory for misleading details that was of great-est interest here. We found a signiWcant main eVect onfalse alarms due to the presence or absence of misleadinginformation, F (2,138) D 4.81; p < 01, a signiWcant moodmain eVect, F (2,138) D 3.62; p < .05, and also a signiW-cant interaction between the presence of misleadinginformation and mood, F (2, 138) D 5.33; p < .01, on thismeasure. The nature of these eVects is clearly illustratedin Fig. 2.

As Fig. 2 shows, participants who were previouslyexposed to misleading information were signiWcantlymore likely to report ‘false alarms’, in other words, toreport eyewitness memories for details they have not infact seen. Mood also had a main eVect on false alarms,with those in a negative mood showing overall fewerfalse alarms than did those experiencing positive mood.What is most interesting is the interaction between expo-sure to misleading information, and mood state found

Fig. 2. The interaction between mood and the presence or absence ofmisleading information on recognition (Experiment 2): positive moodincreased, and negative mood decreased the inXuence of misleadinginformation on subsequent eye-witness reports (false alarms).

J.P. Forgas et al. / Journal of Experimental Social Psychology 41 (2005) 574–588 581

here. Persons experiencing a positive mood while receiv-ing the misleading information were more likely subse-quently to report it as true than did those not exposed tomisleading details, t (46) D 4.69; p < .01. In contrast, nega-tive aVect seems to have all but eliminated this source oferror in eyewitness recollections: exposure to misleadinginformation did not result in higher false alarm rates inthe sad mood condition, t (46) D 1.12; NS. Participants inthe neutral mood group reported more false alarms afterexposure to misleading information, but this eVect wasnot as strong as in the happy group, t (46) D 2.31; p < .05.

Signal detection analysisAs in Experiment 1, an integrated signal detection

analysis of the entire recognition performance was alsoperformed. Hit and false alarm rates were calculated foreach participant and a standard correction was appliedwhen these rates were 0 or 1 (Snodgrass & Corwin,1988). Using a log transformation method to approxi-mate a normal distribution (Brophy, 1986), measures ofdiscrimination (d�) and bias (C) were calculated. Wefound that discrimination (d�) was again signiWcantlyinXuenced by mood, F (2,138) D 4.20; p < .05. NegativeaVect when receiving misleading information resulted inbetter discrimination between correct and incorrectdetails compared to the control and the positive moodgroups, respectively, t (94) D 2.73, p < .01; t (94) D 1.94,p < .05; (M D .59, .38, .35). There was no diVerence in d�between the control and positive groups. As predicted,exposure to misleading information also impaired par-ticipants’ discrimination ability, F (1,142) D 4.68; p < .05;(M D .56, .32). As in Experiment 1, response bias (C) wasnot inXuenced either by mood, nor by exposure to mis-leading information, F (2, 138) D .88, NS;F (1, 142) D 1.13, NS. The signal detection analysis is con-sistent with predictions, and shows that overall memoryperformance was signiWcantly inXuenced by mood whilereceiving misleading information. Once again, negativemood states produced an improved ability to discrimi-nate between false and correct information.

Analysis of conWdence ratingsConWdence ratings for the recognition judgments

were Wrst subjected to a principal components analysis,that revealed that all conWdence judgments were highlycorrelated, and loaded on a single factor, accounting for61.5% of the variance. The four conWdence judgmentsfor each participant were thus combined into a singlemeasure (Cronbach’s �D .77), and subjected to a 3 £ 2analysis of variance, evaluating the eVects of mood, andthe presence or absence of misleading information.

Results showed only a mood main eVect,F (2, 138) D 3.87, p < .05, indicating that those in a happymood were signiWcantly more conWdent in their recogni-tion accuracy than were people in the neutral,t (94) D 1.74; p < .05 or in the negative mood conditions,

t (94) D 2.81; p < .01, with no diVerence between the neu-tral and the negative mood groups (M D 5.83, 5.01, 4.42).This is an interesting and somewhat counterintuitiveWnding, suggesting that the positive mood group whowere in fact least accurate and most inXuenced by mis-leading information were paradoxically also most conW-dent that their memory was correct (Penrod & Cutler,1996). This pattern seems consistent with a generalmood-congruent eVect on judgments, suggesting that theselective priming and greater availability of positiveinformation in a good mood may have produced moreoptimistic—but also incorrect—conWdence judgments.

The relationship between conWdence ratings andaccuracy was further assessed by calculating an overallcorrelation between individual conWdence judgmentsand associated accuracy scores. Results showed a non-signiWcant correlation, r D ¡.078, conWrming that conW-dence judgments were not related to actual eyewitnessaccuracy. These results suggest that mood eVects on eye-witness accuracy appear to operate at a subconsciouslevel that is not open to meta-cognitive inspection. Thelack of any relationship between conWdence judgmentsand eyewitness accuracy indicates that participants hadno real, direct insight into their own mental processes(Nisbett & Wilson, 1977) and presumably relied on aVec-tively primed information when computing their conW-dence judgments.

Overall, these results support our main prediction,that positive mood states promote a constructive infor-mation processing style that promotes the incorporationof misleading details into eyewitness accounts (Fiedler,2001; Fiedler et al., 1991). Can conscious eVorts to sup-press aVect mitigate these eVects? And what role do indi-vidual diVerence variables play in the misinformationeVect? The next experiment was designed to investigatesome of these questions.

Experiment 3

Experiment 3 sought to replicate the mood eVects oneyewitness memory found previously, and alsoexplored whether participants would be able to activelysuppress the impact of their moods when instructed todo so. This issue is of some interest, because warningand instructing witnesses is still a fairly common strat-egy for limiting undesirable inXuences within judicialand forensic settings. In addition, the role of individualdiVerence variables (such as self-monitoring and socialdesirability) in facilitating aVect suppression, andmood eVects on eye-witness accuracy were also investi-gated, based on previous evidence suggesting that suchindividual characteristics may play an important rolein aVective inXuences on cognition, and in eyewitnessmemory (Hosch, 1994; Rusting, 2001; Schooler & Lof-tus, 1993).

582 J.P. Forgas et al. / Journal of Experimental Social Psychology 41 (2005) 574–588

AVect suppression instructions

Previous studies suggest that increased self-awarenesscan often limit or even reverse mood eVects on judg-ments and cognition (Berkowitz et al., 2000). In contrast,verbal instructions are often ineVectual, or even counter-productive when it comes to the control of cognitive andattentional processes (Wegner & Bargh, 1998). As verbalinstructions are still used in some forensic and judicialsettings, their eYcacy in controlling aVective inXuenceson eyewitness accuracy were also explored here.

Individual diVerences

We also expected that individual diVerence variablescould play an important role in aVective inXuences oneyewitness reports, and on people’s ability to controltheir aVective states when instructed to do so (Hosch,1994; Schooler & Loftus, 1993; Tomes & Katz, 1997).Prior studies suggest that people scoring high on traitssuch as self-monitoring and social desirability tend toengage in more motivated processing, and their judg-ments and behaviors are less open to incidental aVectiveinXuences as a result (Forgas, 2002; Rusting, 2001). Forexample, high self-monitors might be particularly goodat monitoring their moods and may thus be less open tomisinformation eVects when forewarned than are lowself-monitors. Similarly, high concern with social desir-ability may increase the motivation to follow instruc-tions to control mood. One objective of Experiment 3was to explore the inXuence of such individual diVerencevariables on aVect control and on eyewitness memory.

Method

Overview, design, and participantsStudents were shown two 5-min videotapes showing

(a) a robbery in a convenience store, and (b) a weddingscene (the target episodes). They were instructed to“watch these events as if they were unexpected incidentsthey observed while walking on a street.” After an inter-val of some 45 min Wlled with various distracter tasks,they were shown short 10-min video-Wlms (in fact, themood induction), and then received a short question-naire about the episodes they witnessed that either did,or did not contain misleading information about theevent, and that either instructed, or did not instruct par-ticipants to ‘disregard and control their aVective states’(the aVect control manipulation). After a further 45-mininterval Wlled with unrelated distracter tasks, the accu-racy of their eyewitness memory for the two events wastested. The experiment is based on a 2 £ 2 £ 2 design,with mood (happy, sad), the provision of misleadinginformation (present/absent) and the aVect controlinstruction (present/absent) as the independent vari-ables. Participants were 80 students (49 female, 31 male)

with a median age of 20 who completed the experimentas part of their course requirements. All participants alsocompleted the Snyder self-monitoring scale, and theCrowne-Marlowe social desirability scale during a sepa-rate testing session at the beginning of the semester.

Mood induction, misleading information, and aVect suppression manipulations

The same mood induction and mood validation pro-cedure was used as in Experiment 2. However, only posi-tive and negative mood induction tapes were shown.Following the mood induction, participants were askedto complete a brief questionnaire asking four questionsabout each of the three episodes they witnessed. In theaVect suppression condition only, the following instruc-tions were printed at the top of the questionnaire:“While doing this task, please make an eVort to activelysuppress how you might be feeling. This usually requiresa determined eVort to keep your aVective state underconstant control, making sure that you do not giveexpression to your feelings.” Orthogonal to this manipu-lation, half of the participants received misleading ques-tions, and half did not. There were four questions abouteach episode, each prepared in two versions: a direct ver-sion containing no misleading information (e.g., ‘Do youremember the robbers putting handcuVs on the shop-keeper?’), and a parallel form containing planted, mis-leading information (set in italics here: ‘Do youremember the robbers put handcuVs on the shopkeeperbefore they gagged him?). Half of the participantsreceived the direct form, and half received the form con-taining misleading information. A post-experimentalquestionnaire was also administered at this time, askingparticipants to rate their mood on happy–sad and good–bad scales embedded among a number of distracteritems (the mood validation).

Measure of eyewitness accuracyAfter a 45-min distraction period Wlled with unrelated

activities, participants completed the eyewitness accuracyquestionnaire, containing 12 true/false questions abouteach of the witnessed episodes. Again, four questionseach tested the recollection of actual, correct details, mis-leading details planted as part of the earlier manipulation,and incorrect details that were neither seen nor suggested.Participants also rated their conWdence in each of theirresponses. Dependent measures thus included the num-ber of correct, misleading and incorrect details remem-bered. A careful debrieWng concluded the procedure,revealing no awareness of the manipulations.

Results

Mood validationThe happy–sad and good–bad mood validation scales

were again highly correlated, and were combined to

J.P. Forgas et al. / Journal of Experimental Social Psychology 41 (2005) 574–588 583

construct a single measure of mood valence (Cronbach’s� D .80). A t test conWrmed that participants in the posi-tive condition rated their mood as signiWcantly betterthan did those receiving negative feedback, t (78) D 4.46;p < .01 (M D 2.55, 5.89), conWrming that the mood induc-tion was again highly eVective in producing diVerentaVective states.

Eyewitness accuracyRecognition scores on the three dependent measures

(correct, misleading, and incorrect details across the twowitnessed episodes) were Wrst subjected to an overall2 £ 2 £ 2 multivariate analysis of variance, with mood(happy, sad), misleading information (present/absent)and suppression instructions (present/absent) as theindependent variables. Exposure to misleading informa-tion had a signiWcant inXuence using the Wilks lambdacriterion, F (1, 72) D 4.27; p < .05, and there was also a sig-niWcant interaction eVect between mood and exposure toinformation, F (1,72) D 3.51; p < .05. Follow-up univari-ate analyses of variance were also performed on recogni-tion for correct, incorrect and misleading detailsseparately. Results showed no signiWcant eVects for cor-rect and incorrect details, as expected. As predicted, thefalse alarm scores for misleading details were signiW-cantly inXuenced by exposure to misleading questions,F (1, 73) D 13.27, p < .01, and we also found a as signiW-cant mood by misleading questions interaction eVect,F (1, 73) D 4.19, p < .05 (Fig. 3). No other main eVects orinteractions were signiWcant. Thus, it appears that themood suppression instructions had no detectable overalleVect on eyewitness memory.

As Fig. 3 illustrates, exposure to misleading informa-tion signiWcantly increased false alarms, and so reducedeyewitness accuracy, and did so most when people werein a happy rather than a sad mood. For those in a happymood, exposure to misleading information produced sig-niWcantly more false alarms, t (38) D 3.61; p < .01, while

Fig. 3. The eVects of induced mood and the presence or absence of mis-leading information on recognition (Experiment 3): positive moodincreased, and negative mood decreased witnesses’ susceptibility tomisleading information provided after the event, as reXected in falsealarm rates.

those in a negative mood showed a much smaller eVect,t (38) D 2.23; p < .05. Overall, false alarm rates were sig-niWcantly higher for those exposed to misleading infor-mation in the positive than in the negative moodcondition, t (78) D 3.44; p < .01.

Signal detection analysisAn overall signal detection analysis of recognition

performance was also carried out, as described in Exper-iments 1 and 2, based on the corrected hit and falsealarm rates (Snodgrass & Corwin, 1988). Two signaldetection indices, discrimination (d�) and bias (C) werecalculated (Brophy, 1986), measuring ability to discrimi-nate between correct and false details, and conservativevs. liberal bias in rejecting or accepting doubtful infor-mation, respectively. AVect signiWcantly inXuenced dis-crimination (d�), F (1, 72) D 4.85, p < .05, with markedlybetter discrimination performance in negative ratherthan positive mood (M D .51, .27). Neither exposure tofalse information, nor the aVect control instructions hadany inXuence on discrimination measure. The second,bias measure was not inXuenced either by mood, norinformation exposure. However, we found that instruc-tions to control aVect had an unexpected eVect, increas-ing a conservative bias, F (1, 72) D 3.97, p < .05, (M D .13,¡.05). The results of the signal detection analysis againconWrm the beneWcial eVects of negative aVect for subse-quent recognition memory performance. It is interestingthat instructions to control aVect did not actually elimi-nate the mood eVect, but rather, produced a more con-servative response bias leading forewarned participantsto reject doubtful information. This pattern conWrmsthat instructions to control aVect were not eVective inachieving their purpose.

These results are clearly consistent with our primaryhypothesis, that positive mood states promote a con-structive information processing style that increases thelikelihood that incorrect information will be incorpo-rated into eyewitness accounts. Negative aVect in turnappears to reduce susceptibility to misinformation.Because one of the greatest threats to the accuracy ofeyewitness memories is precisely the inadvertent, con-structive incorporation of ‘foreign’ details (Loftus, 1979;Wells & Loftus, 2003), the empirical demonstration thatthis tendency is facilitated by positive mood, and inhib-ited by negative mood may have a variety of importanttheoretical and practical implications.

ConWdence judgmentsConWdence ratings associated with the four recogni-

tion judgments for each of the two episodes were againhighly correlated, and loaded on a single factor(VAF D 59.2%). All conWdence judgments were thuscombined into a single measure (Cronbach’s �D .78),and subjected to a 2 £ 2 £ 2 analysis of variance, evaluat-ing the eVects of mood, and the presence or absence of

584 J.P. Forgas et al. / Journal of Experimental Social Psychology 41 (2005) 574–588

misleading information, and mood suppression instruc-tions on conWdence judgments. We found a signiWcantmood main eVect, F (1,72) D 4.48, p < .05, indicating thatthose in a happy mood were signiWcantly more conWdentin their recognition accuracy than were people in thenegative mood condition (M D 5.34, vs. 4.02).

As also found in the previous experiment, a correla-tional analysis revealed no relationship between self-rated conWdence and eyewitness accuracy, r D .046,conWrming that participants had little apparent insightinto their real level of eyewitness accuracy (Nisbett &Wilson, 1977; Penrod & Cutler, 1996). Rather, positivemood simply increased overall conWdence, consistentwith a general mood-congruent eVect on judgments(Bower & Forgas, 2001; Schwarz & Bless, 1991). Whatrole do individual diVerences play in these eVects? Weshall turn to this question next.

Self-monitoringFirst, high and low self-monitors were identiWed on

the basis of a median split (M D 12.5). We thenincluded self-monitoring as an additional independentvariable in the ANOVA; results showed a signiWcantthree-way interaction between self-monitoring, mood,and the aVect suppression instructions, F (1, 72) D 4.25,p < .05 (see Fig. 4). Among low self-monitors, mood,and the suppression instructions had no consistenteVect on the number of false alarms reported. However,among high self-monitors there was a signiWcant two-way interaction between mood and suppressioninstructions, F (1, 36) D 7.65, p < .01. Suppressioninstructions reduced the number of false alarms in thepositive aVect groups, F (1, 39) D 4.14, p < .05, and mar-ginally increased the number of false alarms in the

negative aVect condition, F (1, 39) D 3.95, p < .06 (Fig.4). It seems that for high self-monitors only, instruc-tions to suppress mood did have an inXuence on eye-witness accuracy, generally reducing the overall moodeVects demonstrated earlier. This pattern indicates thataVect-suppression instructions were only eVectiveamong high self-monitors, and led to happy partici-pants becoming more, and sad participants becomingless accurate in an apparent reduction of the basicmood eVects found here.

Social desirabilityA dichotomous SD variable was created and included

in the analysis by classifying participants as high andlow on the basis of a median split on social desirabilityscores (M D 14.5). This analysis revealed a signiWcantthree-way interaction involving social desirability, moodand suppression, F (1, 68) D 4.32, p < .05. For low SD par-ticipants, there were no interaction eVects between moodand suppression instructions. However, there was a sig-niWcant two-way interaction between mood and sup-pression instructions for high SD participants,F (1,37) D 12.89, p < .01. Among happy, high SD partici-pants, suppression instructions reduced the number offalse alarms reported, F (1,38) D 13.75, p < .01; butamong sad, high SD participants, the eVects of the sup-pression instructions were not signiWcant, F (1, 38) D 2.69,p < .11 (Fig. 5). Once again, we found that the eVects ofthe mood suppression instructions were dependent onindividual diVerence variables. When suppressioninstructions worked, their eVect was to decrease falsealarms (in happy mood), and increase false alarms (insad mood), in essence ameliorating the information pro-cessing consequences of moods.

Fig. 4. The three-way interactive eVects of mood, self-monitoring and aVect suppression on eyewitness accuracy (false alarms for misleading details).For high self-monitors only, instructions to suppress positive aVect reduced, and instructions to suppress negative aVect increased recognition errors.

Fig. 5. The three-way interactive eVects of mood, social desirability and aVect suppression on eyewitness accuracy (false alarms for misleadingdetails). For high social desirability participants, instructions to suppress positive aVect reduced, and instructions to suppress negative aVectincreased recognition errors.

J.P. Forgas et al. / Journal of Experimental Social Psychology 41 (2005) 574–588 585

General discussion and conclusions

Remembering the details of observed events contain-ing complex and often confusing information is one ofthe more demanding cognitive tasks people face ineveryday life. Despite extensive interest in aVective inXu-ences on cognition in recent years (Bless, 2000; Bower &Forgas, 2001; Eich & Macauley, 2000; Fiedler & Bless,2001; Forgas, 2002), the inXuence of aVective states oneyewitness accuracy has received less than adequateattention to date. These three experiments oVer conver-gent evidence that transient moods do have a markedinXuence on people’s susceptibility to misleading infor-mation when remembering witnessed events.

All three experiments showed that positive moodincreased, and negative mood decreased the tendency toincorporate misleading details into eyewitness accounts.We also found that congruence between current moodand the aVective valence of the target event provided nodefence against misleading information (Experiment 1).Memories for real-life events (Experiment 2) were just assusceptible to aVective inXuences as were memories forvideotaped, or photographed scenes, and subjective con-Wdence was unrelated to eyewitness accuracy (Penrod &Cutler, 1996). Paradoxically, happy mood reduced accu-racy yet increased conWdence, suggesting that people hadlittle meta-cognitive awareness of their cognitive pro-cesses (Nisbett & Wilson, 1977), and were unaware ofthe consequences of their mood states for their thinkingand memory (Berkowitz et al., 2000). Instructions to sup-press aVect had no overall eVect, but were selectivelyeVective for participants who scored high on self-moni-toring, and social desirability. These experiments haveseveral important theoretical, as well as some practicalimplications for our understanding of mood eVects onsocial cognition and eyewitness recollections inparticular.

Theoretical implications

Although considerable research now supports theprediction of mood-congruence, and mood-state-depen-dence in memory (Bower & Forgas, 2001; Eich &Macauley, 2000), less is known about the processingconsequences of moods, and how they may impact onconstructive memory processes. Our results are broadlyconsistent with recent aVect-cognition theories that pre-dict that good and bad moods should have an asymmet-ric eVect on processing strategies and outcomes (Bless,2000; Clore et al., 2001; Fiedler & Bless, 2001; Forgas,1995, 2002; Schwarz, 1990).

SpeciWcally, recent experiments suggest that negativemoods often promote a more accommodating, exter-nally oriented and piecemeal information processingstyle that often results in more accurate and less dis-torted judgments and inferences (Fiedler & Bless, 2001;

Fiedler et al., 1991). For example, Forgas (1998a) foundthat happy moods increased and negative moodsdecreased the incidence of judgmental distortions suchas the fundamental attribution error, while negativemoods reduced these eVects. Fiedler et al. (1991) alsoshowed that positive aVect promotes a more constructiveprocessing style, increasing the likelihood that impres-sion formation judgments will be inXuenced by previ-ously primed information. The present experimentsextend this literature by demonstrating a direct linkbetween mood, and the tendency to incorporate mislead-ing information into eyewitness memories.

It is particularly interesting that participants had littleawareness of the eVects of their mood on the accuracy oftheir memory. Indeed, their conWdence was greatest pre-cisely when they were most mistaken. The lack of anyintrospective awareness of aVective inXuences on eyewit-ness accuracy may help to explain why instructions tosuppress mood eVects were also ineVective (Experiment3). However, individual diVerences do seem to play somerole in these eVects. As also suggested in prior experi-ments, people who score high on traits such as self-mon-itoring and social desirability are also more likely toadopt a motivated, directed information processing stylein social situations (Forgas, 1998b). Such individualsmay be better able to act on aVect-suppression instruc-tions, leading to a reduction of both positive and nega-tive mood eVects on memory.

What can these results tell us about the cognitivemechanisms responsible for the misinformation eVect ineyewitness memories (Schooler & Loftus, 1993; Wells &Loftus, 2003)? It is most likely that positive aVect andthe constructive, assimilative processing style it pro-moted simply increased people’s tendency to receive sub-sequent misinformation in an uncritical, acceptingmanner (Bless, 2000; Fiedler & Bless, 2001). Onceencoded, misleading details do not necessarily eliminatethe original memory trace (Loftus, 1979), but simplybecome more accessible and so may interfere with theretrieval of correct details (Bekerian & Bowers, 1983). Itmay well be that these eVects also depend on the strengthof the memory trace, such that strong traces may bemore likely to be resistant to incidental mood eVects(Pezdek & Roe, 1995; Reyna & Lloyd, 1997), an issuethat deserves further investigation.

Several theories speciWcally predict mood-inducedprocessing diVerences consistent with the Wndingsreported here. For example, Schwarz and Bless’s (1991)cognitive tuning model as well as Martin et al.’s (1993)aVect as input model both suggest that positive moodscan inform people of the need to engage in less extensive,vigilant processing. Alternative models emphasize therole of good or bad moods in recruiting more or less sys-tematic processing (Bodenhausen, 1993; Clore et al.,2001; Schwarz, 1990; Stroessner & Mackie, 1992). OurWndings are not simply due to mood-induced diVerences

586 J.P. Forgas et al. / Journal of Experimental Social Psychology 41 (2005) 574–588

in processing eVort and vigilance (Bless, 2000). Rather,aVective states seem to recruit qualitatively diVerent pro-cessing strategies, suggesting a dichotomy between exter-nally oriented, accommodative (in negative mood) vs.constructive, assimilative (in positive mood) processingstyles (Bless, 2000; Fiedler, 2001) that can account forjust the kind of diVerences in eyewitness memories wefound here. Our results also support Fiedler et al.’s(1991) dual-force model that suggests that positive aVectpromotes more constructive processing, for example, inimpression formation judgments. Indeed, based on theirmodel, Fiedler et al. (1991) speciWcally anticipated that“good mood can be predicted to produce more falsealarms in eyewitness reports” (p. 376), exactly what wasfound here.

Practical implications

The ability to correctly remember the events we wit-ness also plays an important role in many spheres ofeveryday life. Eyewitness recollections are especiallyimportant in judicial decisions and in forensic practice.Despite accumulating evidence for the role of aVect inmemory and many other social cognitive tasks (Cloreet al., 2001; Forgas, 1995, 2002; Fiedler & Bless, 2001;Mayer & Hanson, 1995; Sedikides, 1995), not muchattention has been paid to the role of moods, as distinctfrom more intense emotions in eyewitness memories.Forensic research has mainly looked at the eVects ofmore intense arousal states and speciWc emotions, as wellas context eVects on eyewitness accuracy (Davies &Milne, 2003; Malpass, 1996); relatively little work hasbeen done on the eVects of mild, everyday mood states.The Wnding that positive aVect can magnify, and nega-tive aVect can reduce the misinformation eVect suggeststhat even mild aVective states may be of interest in foren-sic practice. Warnings and instructions, sometimes usedto safeguard against distortions in judicial practice, areunlikely to be universally eVective in controlling suchaVective inXuences. AVect-control instructions may onlybe eVective for individuals who are more able (high onself-monitoring) and willing (high on social desirability)to follow such instructions.

Limitations and future prospects

There are also some limitations to these results. Pastevidence suggests that mood eVects on cognition arequite subtle, and often depend on the kind of processingstrategy adopted by people (Forgas, 1995, 2002). It maywell be that in circumstances that call for more moti-vated processing (for example, due to the increased per-sonal relevance of the task; Forgas, 1995, 1998b) theeVects of mood on eyewitness memory may be reduced.These eVects also may well be highly sensitive to prag-matic and situational variables such as the identity,

status and inXuence of the person planting the mislead-ing information, the nature of the event to be remem-bered, and the nature of the memory task. There isconsiderable scope in future studies to explore the role ofvarious pragmatic variables in recruiting diVerent pro-cessing strategies, and thus mediating the ensuing moodeVects on eyewitness memory.

Future experiments may also directly explore the pro-cessing mechanisms responsible for these eVects, forexample by recording the processing latencies involvedwhile happy and sad individuals respond to misleadingquestions. This approach has been successfully used inpast work assessing mood eVects on memory and judg-ments. Results did show that processing latency is oftenan important mediator of mood eVects (Bower & For-gas, 2001; Forgas, 1995; Sedikides, 1995). Further, asnoted before, moods may inXuence memory at each ofthe three stages of the eyewitness process: encoding(when Wrst witnessing an event), subsequent re-process-ing (the post-event misinformation stage), and Wnalretrieval and judgments. These experiments focused onstage two only; future experiments may explore moodeVects on the encoding and retrieval stages as well.

We should also note that eyewitness researchers fre-quently use procedures other than true/false recognitiontests to assess eyewitness accuracy, such as free recalland open-ended responses as suggested by the cognitiveinterview technique (Davies, 1993; Malpass, 1996; Wells& Olsen, 2003). Thus, replicating our results with diVer-ent response paradigms would be desirable. Anotherissue concerns the external validity of our results, a ques-tion of particular importance in studies of eyewitnessmemory. Given the consistency of the results acrossthree experiments, diVerent target events and diVerentmood inductions we can be reasonably conWdent of thereliability of these eVects; nevertheless, it would beimportant to demonstrate corresponding mood eVectson eyewitness memory in naturalistic situations andusing a variety of paradigms (cf. Mayer & Hanson,1995).

Remembering personally witnessed events can be acomplex cognitive task that requires constructive pro-cessing that is open to a variety of distortions (Davies,1993; Loftus, 1979; Malpass, 1966). Our results suggestthat positive mood increases, and negative mooddecreases the likelihood that false information willlater be remembered as true. Much has been discoveredabout the processes that govern constructive memoryprocesses in recent years, yet not enough is knownabout how feelings impact on the accuracy of eyewit-ness memories. Based on recent research on aVect andsocial cognition (Clore et al., 2001; Fiedler, 2001;Forgas, 1995, 2002; Mayer & Hanson, 1995; Salovey &Birnbaum, 1989), our results suggest that both goodand bad mood can have a signiWcant impact oneyewitness memories, due to the kind of information

J.P. Forgas et al. / Journal of Experimental Social Psychology 41 (2005) 574–588 587

processing strategies they generate. Recent aVect-cog-nition theories (Bless, 2000; Fiedler, 2001; Forgas,1995, 2002; Schwarz, 1990) appear particularly relevantto understanding these subtle and process-contingenteVects. Further research on aVect and eyewitness mem-ory should be of considerable theoretical, as well asapplied interest.

References

Bekerian, D. D., & Bowers, J. M. (1983). Eyewitness testimony: Werewe misled?. Journal of Experimental Psychology: Learning, Memoryand Cognition, 9, 139–145.

Berkowitz, L., JaVee, S., Jo, E., & Troccoli, B. T. (2000). On the correc-tion of feeling-induced judgmental biases. In J. P. Forgas (Ed.),Feeling and thinking: The role of aVect in social cognition (pp. 131–152). New York: Cambridge University Press.

Bless, H. (2000). The interplay of aVect and cognition: The mediatingrole of general knowledge structures. In J. P. Forgas (Ed.), Feelingand thinking: The role of aVect in social cognition (pp. 201–222).New York: Cambridge University Press.

Bless, H., Mackie, D., & Schwarz, N. (1996). Mood eVects on encodingand judgmental processes in persuasion. Journal of Personality andSocial Psychology, 63, 585–595.

Bodenhausen, G. V. (1993). Emotions, arousal, and stereotypic judg-ments: A heuristic model of aVect and stereotyping. In D. M. Mac-kie & D. L. Hamilton (Eds.), AVect, cognition, and stereotyping (pp.13–37). San Diego: Academic Press.

Bower, G. H., & Forgas, J. P. (2001). Mood and Social Memory. In J. P.Forgas (Ed.), The handbook of aVect and social cognition (pp. 95–120). Mahwah, NJ: Erlbaum.

Brophy, A. L. (1986). Alternatives to a table of criterion values in sig-nal detection theory. Behavior Research Methods, Instruments, andComputers, 18, 285–286.

Brown, J. M. (2003). Eyewitness memory for arousing events. AppliedCognitive Psychology, 17, 93–106.

Ceci, S. J., & Bruck, M. (1993). Suggestibility of the child witness: Ahistorical review and synthesis. Psychological Bulletin, 113(3), 403–439.

Clark, M. S., & Isen, A. M. (1982). Towards understanding the relation-ship between feeling states and social behavior. In A. H. Hastorf &A. M. Isen (Eds.), Cognitive social psychology (pp. 73–108). NewYork: Elsevier-North Holland.

Clore, G. L., Gasper, K., & Garvin, E. (2001). AVect as information. InJ. P. Forgas (Ed.), The handbook of aVect and social cognition (pp.95–120). Mahwah, NJ: Erlbaum.

Davies, G. E. (1993). Witnessing events. In G. M. Davies, R. H. Logie,(Eds.), Memory in everyday life. Advances in Psychology, 100, (pp.367–401).

Davies, G., & Milne, A. (2003). Eyewitness composite production: Afunction of mental or physical reinstatement of context. CriminalJustice and Behavior, 12, 209–220.

Eich, E., & Macauley, D. (2000). Fundamental factors in mood-depen-dent memory. In J. P. Forgas (Ed.), Feeling and thinking: The role ofaVect in social cognition (pp. 109–130). New York: Cambridge Uni-versity Press.

Eich, E. E., & Schooler, J. W. (2000). Cognition–emotion interactions.In E. E. Eich, J. F. Kihlstrom, G. H. Bower, J. P. Forgas, & P. Nie-denthal (Eds.), Cognition and emotion (pp. 3–29). New York:Oxford University Press.

Ellis, H. C., & Ashbrook, T. W. (1988). Resource allocation model ofthe eVects of depressed mood state on memory. In K. Fiedler & J. P.Forgas (Eds.), AVect, cognition and Social Behaviour (pp. 25–43).Toronto: Hogrefe.

Fiedler, K. (2001). AVective inXuences on social information process-ing. In J. P. Forgas (Ed.), The handbook of aVect and social cognition(pp. 163–185). Mahwah: Erlbaum.

Fiedler, K., Asbeck, J., & Nickel, S. (1991). Mood and constructivememory eVects on social judgment. Cognition and Emotion, 5, 363–378.

Fiedler, K., & Bless, H. (2001). The formation of beliefs in the interfaceof aVective and cognitive processes. In N. Frijda, A. Manstead, & S.Bem (Eds.), The inXuence of emotions on beliefs. New York: Cam-bridge University Press.

Forgas, J. P. (1994). Sad and guilty? AVective inXuences on the expla-nation of conXict episodes. Journal of Personality and Social Psy-chology, 66, 56–68.

Forgas, J. P. (1995). Mood and judgment: The aVect infusion model(AIM). Psychological Bulletin, 116, 39–66.

Forgas, J. P. (1998a). On feeling good and getting your way: MoodeVects on negotiation strategies and outcomes. Journal of Personal-ity and Social Psychology, 74, 565–577.

Forgas, J. P. (1998b). Happy and mistaken? Mood eVects on the funda-mental attribution error. Journal of Personality and Social Psychol-ogy, 75, 318–331.

Forgas, J. P. (2002). Feeling and doing: AVective inXuences on interper-sonal behavior. Psychological Inquiry, 13, 1–28.

Forgas, J. P., & Moylan, S. J. (1987). After the movies: The eVects oftransient mood states on social judgments. Personality and SocialPsychology Bulletin, 13, 478–489.

Hosch, H. (1994). Individual diVerences in personality and eyewitnessidentiWcation. In D. F. Ross & J. D. Read (Eds.), Adult eyewitnesstestimony: Current trends and developments (pp. 328–347). NewYork: Oxford University Press.

Loftus, E. F. (1979). Eyewitness testimony. Cambridge: Harvard Uni-versity Press.

Malpass, R. S. (1996). Enhancing eyewitness memory. In S. L. Sporer &R. S. Malpass (Eds.), Psychological issues in eyewitness identiWcation(pp. 177–204). Hillsdale, NJ: Erlbaum.

Martin, L. L., & Clore, G. (2001). Theories of cognition and aVect. Mah-wah, NJ: Erlbaum.

Martin, L. L., Ward, D. W., Achee, J. W., & Wyer, R. S. (1993). Moodas input: People have to interpret the motivational implications oftheir moods. Journal of Personality and Social Psychology, 64, 317–326.

Mayer, J., & Hanson, E. (1995). Mood-congruent judgment over time.Personality and Social Psychology Bulletin, 21, 237–244.

Nisbett, R. E., & Wilson, T. D. (1977). Telling more than we can know:Verbal reports on mental processes. Psychological Review, 84, 231–259.

Penrod, S., & Cutler, B. (1996). Witness conWdence and witness accu-racy: Assessing their forensic relation. Psychology, Public Policyand Law, 1(4), 817–845.

Pezdek, K., & Roe, C. (1995). The eVects of memory trace strength onsuggestibility. Journal of Experimental Child Psychology, 60, 116–128.

Reyna, V. F., & Lloyd, F. (1997). Theories of false memory in childrenand adults. Learning and Individual DiVerences, 9(2), 95–123.

Rusting, C. (2001). Personality as a mediator of aVective inXuences onsocial cognition. In J. P. Forgas (Ed.), The handbook of aVect andsocial cognition (pp. 371–391). Mahwah, NJ: Erlbaum.

Salovey, P., & Birnbaum, D. (1989). InXuence of mood on health-related cognitions. Journal of Personality and Social Psychology, 57,539–551.

Schooler, J. W., & Eich, E. E. (2000). Memory for emotional events. InE. Tulving & F. I. M. Craik (Eds.), The Oxford handbook of memory(pp. 379–392). London: Oxford University Press.

Schooler, J. W., & Loftus, F. (1993). Multiple mechanisms mediateindividual diVerences in eyewitness accuracy and suggestibility. InJ. M. Puckett & H. W. Reese (Eds.), Mechanisms of everyday cogni-tion (pp. 177–203). New York: Wiley.

588 J.P. Forgas et al. / Journal of Experimental Social Psychology 41 (2005) 574–588

Schwarz, N. (1990). Feelings as information: Informational and moti-vational functions of aVective states. In E. T. Higgins & R. Sorren-tino (Eds.), Handbook of motivation and cognition (Vol. 2, pp. 527–561). New York: Guildford Press.

Schwarz, N., & Bless, H. (1991). Happy and mindless, but sad andsmart? The impact of aVective states on analytic reasoning. In J. P.Forgas (Ed.), Emotion and social judgments (pp. 55–71). Oxford:Pergamon Press.

Sedikides, C. (1995). Central and peripheral self-conceptions are diVeren-tially inXuenced by mood: Test of the diVerential sensitivity hypothe-sis. Journal of Personality and Social Psychology, 69, 759–777.

Snodgrass, J. G., & Corwin, J. (1988). Pragmatics of measuring recogni-tion memory. Journal of Experimental Psychology: General, 117,34–50.

Stroessner, S. J., & Mackie, D. M. (1992). The impact of induced aVecton the perception of variability in social groups. Personality andSocial Psychology Bulletin, 18, 546–554.

Tomes, J. L., & Katz, A. N. (1997). Habitual susceptibility to misinfor-mation and individual diVerences in eyewitness memory. AppliedCognitive Psychology, 11, 233–251.

Wegner, D. M., & Bargh, J. (1998). Automaticity and mental control.In D. T. Gilbert, S. T. Fiske, & G. Linzey (Eds.), The handbook ofsocial psychology (2nd ed., pp. 446–496). New York: McGraw-Hill.

Wells, G. L., & Loftus, E. F. (2003). Eyewitness memory for people andevents. In A. M. Goldstein (Ed.), Handbook of psychology: Forensicpsychology (Vol. 11, pp. 149–160). New York: John Wiley.

Wells, G. L., & Olsen, E. A. (2003). Eyewitness testimony. AnnualReview of Psychology, 54, 277–295.