SOLUTION: The Face in the Crowd Revisited | My Assignment Tutor

See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/12032911The Face in the Crowd Revisited: A Threat Advantage With Schematic StimuliArticle in Journal of Personality and Social Psychology · April 2001DOI: 10.1037/0022-3514.80.3.381 · Source: PubMedCITATIONS1,038READS6,6413 authors, including:Some of the authors of this publication are also working on these related projects:Newspaper reading View projectProximity of emotional states in the dimensional space View projectDaniel LundqvistKarolinska Institutet84 PUBLICATIONS 3,772 CITATIONSSEE PROFILEFrancisco EstevesMid Sweden University141 PUBLICATIONS 5,421 CITATIONSSEE PROFILEAll content following this page was uploaded by Daniel Lundqvist on 20 May 2014.The user has requested enhancement of the downloaded file.Journal of Personality and Social Psychology2001, Vol. 80, No. 3, 381-396Copyright 2001 by the American Psychological Association, Inc.O022-3514/0l/$5.00 DO1: 10.1037//0022-3514.80.3.381The Face in the Crowd Revisited:A Threat Advantage With Schematic StimuliArne Ohman, Daniel Lundqvist, and Francisco EstevesKarolinska InstitutetSchematic threatening, friendly, and neutral faces were used to test the hypothesis that humanspreferentially orient their attention toward threat. Using a visual search paradigm, participants searchedfor discrepant faces in matrices of otherwise identical faces. Across 5 experiments, results consistentlyshowed faster and more accurate detection of threatening than friendly targets. The threat advantage wasobvious regardless of whether the conditions favored parallel or serial search (i.e., involved neutral oremotional distractors), and it was valid for inverted faces. Threatening angry faces were more quickly andaccurately detected than were other negative faces (sad or “scheming”), which suggests that the threatadvantage can be attributed to threat rather than to the negative valence or the uniqueness of the targetdisplay.Unlike most other musculature, the facial muscles are designedto move skin tissue rather than bones (Fridlund, 1994). In combination with functional, comparative, and developmental considerations (e.g., Ohman & Dimberg, 1984), this anatomical fact suggests that the face has evolved as a specialized module to servenonverbal social interchange (an idea pioneered by Darwin, 1872/1998). Inspired by this evolutionary premise and by a previousarticle by Hansen and Hansen (1988), the research reported in thisarticle examines whether people preferentially direct their attention toward a threatening face in a crowd of faces.Facial threat is typically conveyed by a set of gestures suggesting an emotional expression of anger: pronounced frowning brows,intensely staring eyes, and a shut mouth with lowered corners(Ekman & Friesen, 1975). This is similar to the facial displayshown by confidently dominant primates when they assert theirposition in social hierarchies (e.g., Hinde, 1975). Furthermore,these features figure prominently among ceremonial masks that areunderstood as evil or threatening in diverse cultural contexts(Aronoff, Barclay, & Stevenson, 1988). Because our focus is onthe signaling rather than the expressive role of facial displays, weprefer to discuss threatening rather than angry faces, even thoughin many contexts the terms can be used interchangeably. For thesame reason, we refer to friendly rather than to happy faces.In agreement with the evolutionary scenario, pictorially depictedfacial threat is an efficient cue for human fear conditioning (e.g.,Ohman & Dimberg, 1978). Furthermore, these effects are notdependent on conscious identification of stimuli, because they canArne Ohman, Daniel Lundqvist, and Francisco Esteves, Department ofClinical Neuroscience, Karolinska Institutet, Stockholm, Sweden.The research reported in this article was supported by a grant from theBank of Sweden Tercentennial Foundation. We gratefully acknowledge thetechnical assistance of Jan-Eric Litton and the assistance of Jorge Patraquim and Fredrik Palm in collecting parts of the data.Correspondence concerning this article should be addressed to ArneOhman, Section of Psychology, Department of Clinical Neuroscience,Karolinska Institutet and Hospital, Z6, S-171 76 Stockholm, Sweden.Electronic mail may be sent to arne.ohman@ks.se.be observed in reaction to masked facial stimuli (e.g., Esteves,Dimberg, & Ohman, 1994; see Dimberg & Ohman, 1996, for areview of conditioning to facial stimuli). Recent brain imagingstudies show that nonconscious activation of regional cerebralblood flow to masked angry faces centers on the right amygdala(Morris, Ohman, & Dolan, 1998), to which information is conveyed through subcortical visual pathways (Morris, Ohman, &Dolan, 1999). Neuropsychological studies of patients with bilateralamygdala damage suggest that these patients overestimate thetrustworthiness and approachability of those faces that are rated asmost negative by normal participants (Adolphs, Tranel, &Damasio, 1998). In concert, these findings demonstrate that humans can decode, learn, and emotionally respond to threateningfacial stimuli that they do not consciously perceive and that theseeffects are adaptive and mediated by specialized neural circuitry.Thus, this set of findings may reflect an evolved, specializedbehavior module for responding to emotional facial expressions(Ohman & Mineka, in press; Tooby & Cosmides, 1992).Because facial threat provides a warning that aversive consequences are likely, the evolved module should be biased fororienting attention to salient facial gestures that convey threat. Thishypothesis was tested in a pioneering study by Hansen and Hansen(1988). Using a visual search methodology to separate capacityindependent parallel search from effort-demanding serial searchfor target faces (see Wolfe, 1998), Hansen and Hansen exposedtheir research participants to “crowds” of people composed ofmatrices of individual faces. The participants’ task was to detectwhether all the faces in a crowd showed the same emotionalexpression or whether there was a face with a discrepant expression present in the crowd. In support of the hypothesis, Hansen andHansen’s first experiment showed that participants found threatening faces in friendly crowds significantly faster and with fewererrors than they found friendly faces in threatening crowds. However, other aspects of the data were in less agreement with thehypothesis. For example, participants found threatening faces nofaster than neutral faces in friendly crowds.Because the apparently larger between-subjects variability inthreatening than in friendly or neutral expressions complicated the381382 OHMAN, LUNDQVIST, AND ESTEVESinterpretation of the effect of expression in their first experiment,Hansen and Hansen (1988) changed the composition of the stimuliin their second experiment. Rather than involving different individuals, the crowds in the following experiments consisted ofidentical pictures (i.e., the same individual expressing the sameemotion) with targets of the same individual expressing a differentemotion. With these stimuli, Hansen and Hansen again foundfaster detection of threatening targets in friendly crowds than viceversa. Furthermore, whereas the detection of threatening faces wasunaffected by crowd size, it took significantly longer to detect afriendly target (in a threatening crowd) when the matrix wascomposed of nine than of four faces. This was interpreted as a“pop-out” effect indicating a parallel, preattentive search (see, e.g.,Treisman, 1988) for threatening targets but a serial postattentivesearch for friendly targets. Thus, the studies reported by Hansenand Hansen (1988) appeared to provide support for the hypothesesthat threatening facial gestures effectively command attention andthat the effect was mediated by automatic, preattentive perceptualprocesses.However, subsequent research has cast considerable doubt onthe validity of these conclusions. For example, in support of serialrather than parallel search for threatening targets, significant effects of target location in the matrix have been reported both forthreatening and for friendly targets (Hampton, Purcell, Bersine,Hansen, & Hansen, 1989). Because the latencies Hansen andHansen (1988) found for deciding about target absence wereshorter for friendly than for threatening crowds, finding threatening targets faster may simply depend on discarding friendly distractors more quickly (Hampton et al., 1989). Accordingly, moreefficient processing of friendly than of threatening facial expressions is the common finding in the literature (e.g., Esteves &Ohman, 1993; Harrison, Gorelczenko, & Cook, 1990; Kirouac &Dore, 1984; Wagner, MacDonald, & Manstead, 1986). Finally,and most important, the pop-out effect for threatening targetsappears to be due to a stimulus confound (Purcell, Stewart & Skov,1996). Hansen and Hansen (1988) used high-contrast black-andwhite pictures produced by thresholding (white above the threshold, black below it) the gray scale of Ekman and Friesen’s (1975)facial photographs. As a result, conspicuous dark areas that werenot apparent in the friendly faces accidentally appeared on both oftheir threatening faces, thus introducing a low-level perceptualconfound that provided an alternative explanation of the findings.Purcell et al. (1996) showed that the original faces with the fullgray scale (Ekman & Friesen, 1975) did not produce any threat popout. However, the thresholded threatening faces did, but only forparticipants who reported that they had discovered the confounding dark areas.The more efficient processing of friendly than of threateningfacial expressions (e.g., Esteves & Ohman, 1993; Kirouac & Dore,1984; Wagner et al., 1986) may be due to the fact that friendlyfaces are simply more familiar than threatening faces are. Bondand Siddle (1996) collected several sets of data in which collegestudents reported the frequency with which they encountered individuals exhibiting different prototypical facial expressions intheir environment. It is not surprising (although it is reassuring)that students most frequently encountered happy expressions,whereas they saw angry expressions more seldom. Thus, cognitiverepresentations of friendly expressions should be more or lesspermanently primed, which facilitates many types of cognitiveoperations, including some types of attention. Indeed, extensivepilot work in our laboratory as well as some published work(Byrne & Eysenck, 1995) show that friendly targets are morequickly located in neutral crowds than are threatening targets, atleast when several exemplars are used for each category (i. e.,when the individual faces are relatively unfamiliar).Another problem may be that threatening and neutral faces aremore similar to each other than are friendly and neutral faces.Hansen and Hansen (1988) commented in a footnote on the confusability of angry and neutral faces in their Experiment 1. Indeed,a truly neutral face, lacking any invitation to interact, is easilyinterpreted as slightly hostile.A third problem concerns individual variability in posing different types of emotional facial expressions. Virtually everyonecan provide a reasonably convincing friendly smile, but fewerpersons can produce a convincing threatening, angry expression oncommand. As a result, a threatening crowd is necessarily moreheterogeneous than a friendly or a neutral crowd is. Becausedistractor homogeneity is an important determinant of visualsearch efficiency (Duncan & Humphreys, 1989; Wolfe, 1998), thelarger variability of threatening faces runs the risk of confoundingcomparisons between threatening and friendly crowds when several stimulus individuals are used (Hansen & Hansen, 1988).However, to avoid this problem by using the same individual forall positions in a crowd (Hansen & Hansen, 1988) may introduceother problems. First, idiosyncrasies of the particular individualschosen may introduce confounds (see Purcell et al., 1996), and,second, it provides a loss of ecological validity, because crowds ofclones so far are exceedingly rare in real life.These problems (except that pertaining to ecological validity)could be avoided if schematic facial stimuli such as those shown inFigure 1 were used. Because these stimuli are abstractions fromreal faces, they are all unfamiliar, and thus the differences inpriming real faces of different expressions may be less important.Second, the physical features of the faces can be tightly controlled.In Figure 1, it is obvious that the physical differences between thethreatening and the friendly face, on the one hand, and the neutralface, on the other hand, are identical. Third, the two emotionalfaces contain identical physical components that are organized indifferent ways. For example, the two eyebrows have changedplaces, and the mouths and eyes have been turned upside downwhen comparing friendly and threatening expressions. Fourth,because these stimuli are abstract and schematical, they are prototypical rather than individual instantiations of a particular facialexpression, and thus all crowds are by necessity homogenous.The prototypical nature of schematic faces may be important,because signal evolution is likely to result in stereotypical, exagNeutral Friendly ThreateningFigure I. Schematic faces controlled for physical differences betweenemotional and neutral expressions.THE FACE IN THE CROWD REVISITED 383gerated signaling gestures (Krebs & Davies, 1993). Whether theparticular faces shown in Figure 1 capture the essential features ofeach expression is an empirical question. Lundqvist, Esteves, andOhman (1999) performed a series of rating studies of schematicfaces of this type. They interpreted their result to show thatV-shaped eyebrows allocated a face to a “threat area” within athree-dimensional emotional space defined by orthogonal dimensions of valence, arousal, and dominance. The mouth and the eyesprovided further differentiation within this threat area. The friendlyface, on the other hand, was placed at the opposite end of theemotional space, defined by positive valence, low arousal, and lowdominance. Thus, research participants appeared to provide areasonable emotional differentiation between the two emotionalfaces depicted in Figure 1. The purpose of the five experimentsreported in this article is to use schematic facial stimuli to examinewhether threatening and friendly facial expressions are differentially effective in capturing attention.Experiment 1The first experiment essentially replicates the design of Hansenand Hansen’s (1988) Experiment 1 with schematic faces as stimuli.Thus, we asked research participants to search for discrepant facesin friendly, threatening, and neutral crowds. To allow assessmentof the effectiveness of the search, we set the exposure time of thedisplay at either short (1 s) or long (2 s).MethodParticipants. The research participants were 20 students at the University of Stockholm. They were recruited through advertisements on theStockholm University campus, and they were paid 80 kronor (approximately US $10) for their participation. There were 12 men and 8 women,and the age range was 19-50 years (with a positively skewed distribution).The participants agreed to participate in the experiment on an informedconsent basis.Apparatus. Visual stimuli were presented on the 20 in. (50.80 cm)screen of a Macintosh 8100/100 computer that was activated by a 486Personal Computer, programmed in the Microexperimental Laboratorysoftware (Schneider, 1988) to initiate trials and to measure reaction times(RTs). The participants responded by pressing two different keys on thecomputer keyboard with their left and right index fingers. Except fortiming, the experiment was programmed and presented on the Macintoshcomputer using the Macromedia Director 5 software.Stimuli. The stimuli were matrices composed of nine individual schematic faces arranged in 3 X 3 matrices. The displays with a target areshown in Figure 2. The faces were drawn in black against a whitebackground. The outline of the face and the nose were drawn with 1-pixellines, and the eyebrows, eyes, and mouth were drawn with lines of 2 pixels.The individual faces were 84 X 98 pixels. Half of the matrices werecomposed of faces that all showed the same emotional expression (i.e.,neutral, friendly, or threatening). In the other half of the matrices, one ofthe faces was designated as the target and had a different emotionalexpression from that of the background distractors (see Figure 2). Alldistractor expressions were combined with all target expressions, makingsix different target-distractor combinations (Figure 2). Thus, to assurevaried mapping of targets and distractors (i.e., all stimuli serving in bothroles, which promotes effortful serial search; Schneider & Shiffrin, 1977),we made sure that neutral faces occurred as targets, even though thehypothesis only concerned the difference between friendly and hostiletargets. The target could occur at any of the nine positions in the matrix.Thus, there were 54 different matrices containing a target and threedifferent distractor matrices without targets (neutral, happy, angry).Procedure. The participants were tested individually. They wereseated approximately 1 m from the computer screen in a comfortable chairwhose height could be adjusted so that the participant’s eyes were positioned at the center of the screen. The keyboard was placed to allow the tworesponding fingers to be held on the two response keys with the participant’s arms comfortably rested on the table. The individual face had a sizeon the screen of approximately 3° X 3.5°, and the outline of the stimulusmatrix on the screen gave visual angles of approximately 10° X 11.5°.The general nature of the experiment was first orally outlined to theparticipants, and then more detailed written instructions were presented onthe computer screen. The instructions explained that the task was to detecta discrepant face in a matrix of faces. It was also explained that half of thematrices contained a target and that the participant was expected to pressdifferent keys depending on whether a discrepant target was present in amatrix. Before the task began, the participants were taken through a seriesof self-paced training trials on the computer, which showed and explainedthe stimuli and the nature of the participant’s task, stressing the need todecide quickly whether a target was present in a matrix or not. A positivedecision (target present) was always indicated by the right index finger, anda negative decision (target absent) was indicated with the left index finger.A trial was initiated by the appearance of a fixation point (0.4 cmdiameter) at the center of the screen, located where the center face of amatrix would later appear. The fixation point was on for 2 s and wasimmediately replaced by the matrix. The duration of the matrix was either 1or 2 s. With 54 target matrices, 54 matrices with only distractors, and 2matrix durations, each participant was exposed to 216 randomly orderedtrials. A trial was terminated by the response, and then there was a 4-sintertrial interval before the fixation point reappeared on the screen, initiating a new trial.Design and statistical analysis. The RTs for trials on which participants pressed the wrong button (missing a target or falsely perceiving atarget in a target-absent display) were replaced by the participant’s meanfor the condition. Individual RTs deviating by more than three standarddeviations from the participant’s mean for the general condition (e.g.,target-no target) were replaced by the mean plus or minus three standarddeviations (given the skewed distribution of RTs, this almost exclusivelyhappened for long RTs). The design of the study involved three independent variables: distractor expression, target expression, and matrix duration. Neutral targets were not incorporated in the analysis, because theywere not explicitly part of the hypothesis. Furthermore, they could only beanalyzed against emotional (friendly and threatening) backgrounds. Finally, because they had the unique feature of horizontal lines, they could bevery efficiently picked up by a parallel search (Treisman & Gelade, 1980).Thus, neutral targets among emotional distractors always had the shortestsearch times of all conditions. To allow an overall analysis incorporatingthreatening and friendly targets, we rearranged the distractor to include twolevels, neutral and emotional, in which the latter comprised friendly (withthreatening targets) and threatening (with friendly targets) expressions.Thus, the statistical design was a Target (threatening vs. friendly) XDistractor (neutral vs. emotional) X Exposure Time (1 s vs. 2 s) factorialwith repeated measurements in all factors. Statistical analyses of RTs forcorrect responses and proportion of correct responses were accomplishedby analysis of variance (ANOVA), using Tukey’s honestly significantdifference (HSD) tests as follow-up tests when appropriate. Tests fornormality in the distribution of RT data suggest that a logarithmic transformation was warranted (Ratcliff, 1993). The figures, however, showmean RTs.ResultsMean RTs for all conditions of the Experiment are shown inFigure 3a, and accuracy scores are given in Figure 3b. It appearsin Figure 3a that the participants found the threatening target fasterthan they found the friendly one in all conditions except at the384 OHMAN, LUNDQVIST, AND ESTEVES© ©I 10 ©fl ft© © © ©A SO O J ft© ©© © fl & © © O O 10 ©J 6© ©Figure 2. Examples of 3 X 3 matrices with targets used in Experiment 1.short exposure with emotional distractors. Identification of targetswas more accurate when the targets were threatening than whenthey were friendly and with long rather than with short exposure(Figure 3b).According to the statistics, the participants were, overall, faster,F(l, 19) = 14.90, p