You are on page 1of 8

ATTITUDES AND SOCIAL COGNITION

Finding the Face in the Crowd: An Anger Superiority Effect


Christine H. Hansen and Ranald D. Hansen
Oakland University

Facial gestures have been given an increasingly critical role in models of emotion. The biological
significance of interindividual transmission of emotional signals is a pivotal assumption for placing
the face in a central position in these models. This assumption invited a logical corollary, examined
in this article: Face-processing should be highly efficient. Three experiments documented an asym-
metry in the processing of emotionally discrepant faces embedded in crowds. The results suggested
that threatening faces pop out of crowds, perhaps as a result of a preattentive, parallel search for
signals of direct threat.

Theories of emotion, both the more traditional facial effer- work of social cognition models constructed around the capac-
ence models (Buck, 1984; Darwin, 1872/1904; Ekman, 1984; ity to process information automatically as well as attentively
Izard, 1977;Leventhal, 1982;Tomkins, 1980) and the less well- {e,g., Bargh, 1984; Logan, 1979; Shiffrin & Schneider, 1977).
known facial afference models (Gratiolet, 1865; Waynbaum, The parameters of automatic or preattentive processing and at-
1906; Zajonc, 1985), embrace the adaptive significance of facial tentive processing have been described extensively. However,
gestures of emotion to explain their existence. Whether trans- they remain controversial and are currently being tested
mission of emotional facial gestures is through cognitive ap- (Cheng, 1985; Kahneman & Treisman, 1983; Schneider &
praisal of another's face (as Leventhal, 1982, for one, suggests) Shiffrin, 1985; Treisman, 1982; Treisman & Paterson, 1984;
or facial mimicry (Hansen & Hansen, 1984), the capacity to Treisman & Souther, 1985). Among other things, automatic or
efficiently pass an emotional signal from one individual to an- preattentive processing is said to be more rapid than attentive
other has obvious survival value. Certainly research has docu- processing and, unlike attentive processing, may be parallel
mented that the human perceptual system is highly practiced, rather than serial and may not be capacity limited. These prop-
if not hard-wired, to detect human faces (Homa, Haver, & erties imply that automatic processing is less (if at all) suscepti-
Schwartz, 1976; Purcell & Stewart, 1981, 1986; Purcell, Stew- ble to distraction and to information load (Bargh & Pietromo-
art, Botwin, & Kreigh, 1983; van Santen & Jonides, 1978; Za- naco, 1982; Beck, 1983; Shiffrin & Schneider, 1977; Treisman
jonc & Markus, 1984). This research, together with numerous & Gelade, 1980; Treisman & Souther, 1985). Preattentive pro-
experiments demonstrating infants' capacity to discriminate cessing can be construed as a filtering of information so that
among emotional facial gestures (Barrera & Maurer, 1981; La- some features or aspects of an array are "passed through" and
Barbera, Izard, Vietze, & Parisi, 1976; Schwartz, Izard, & An- others are "filtered out." By way of this automatic processing, a
sul, 1985; Wilcox & Clayton, 1968), suggests that the detection feature passed through is more likely to come to attention than
and discrimination of emotional faces may be particularly one filtered out: It would seem to the (attentive) individual as if
efficient. Indeed, there is evidence in the literature (Sackett, that feature of the array commanded attention; it would appear
1966; Schwartz et al., 1985) that the mechanism governing the to "pop out" from the array (Treisman, 1982; Treisman & Gel-
processing effaces achieves maximal efficiency at finding facial ade, 1980; Treisman & Paterson, 1984; Treisman & Souther,
signals of potential direct threat; angry faces may be particu- 1985).
larly attention-grabbing for infants early in development (cf.
This theoretical perspective, applied to the hypothesized
Schwartz et al., 1985).
property of face-processing, led to the prediction of a face-in-
This hypothetical anger-superiority property of face-pro-
the-crowd effect: An angry face in a crowd of benign or happy
cessing can be conveniently conceptualized within the frame-
faces should be found more easily than a happy or benign face
in a crowd of angry faces. If facial threat commands attention,
as hypothesized, an angry face in a happy crowd would be found
The authors are indebted to R. B. Zajonc for his insight and wish to quite readily, whereas a happy face in a crowd of attention-grab-
thank D. G. Purcell and four anonymous reviewers for their helpful
bing, distracting, angry faces could easily be overlooked. If an
comments on this article. This research was supported by Grant BNS-
8517416 from the National Science Foundation. individual were given the task of surveying crowds for the pres-
Correspondence concerning this article should be addressed to Chris- ence of a discrepant face, performance when presented with an
tine H. Hansen, Department of Psychology, Oakland University, Roch- angry face embedded in a happy crowd should exceed that when
ester Hills, Michigan 48309-4401. presented with a happy face embedded in an angry crowd. Tc-
Journal of PeraonalHy and Social Psychology, 1988, Vol. 54. No. 6,917-924
Copyright 1988 by the American Psychological Association. Inc. 0022-35I4/88/S00.75

917
918 CHRISTINE H. HANSEN AND RANALD D. HANSEN

ward testing the hypothesized threat-detection orientation of location of the fixation point). Each face subtended a visual angle of
face-processing, we conducted the three experiments reported approximately 1.4° vertically and 1.05° horizontally. The faces were dis-
here to document the existence of this face-in-the-crowd effect. played against a neutral, gray background. The same nine individuals,
five women and four men, appeared in each crowd. On any given trial,
as a function of crowd condition, all of the individuals except the dis-
Experiment 1 crepant face (if one was present) were shown posing happy, angry, or
neutral gestures of emotion. The position of each individual in the 3 X
The subjects were given the task of surveying crowds for the
3 matrix was randomly determined but was the same across all trials for
presence of a discrepant face. They were exposed to photo- all subjects. Thus, individual was confounded with position. This was
graphs of 108 crowds consisting of nine faces. On 54 trials, all not, however, of concern here because neither position nor individual
nine faces displayed the same emotional gesture. On the re- transmitter effects were currently of interest, and the data were col-
maining 54 trials, one of the faces displayed a different emotion lapsed across this factor.
than the other eight. The first factor in the design, then, was Face in the crowd On one half of the trials, a discrepant face was
Trial, designating whether a discrepant face was available (face inserted in the crowd. In happy crowds, the discrepant face was equally
trial) or not available (no-face trial) in the crowd. The second as often angry or neutral. In angry crowds, the face was equally as often
factor in the design was Crowd. On each of 36 trials the crowd happy or neutral. In neutral crowds, the face was equally as often angry
or happy. Each level of this 3 X 2 nested factorial was replicated with
was happy, neutral, or angry.
the face located in each of the nine display matrix positions. In conse-
quence, there were 54 face-in-the-crowd trials. An equal number of no-
Face Trials face trials—18 each of angry, happy, and neutral crowds—comprised
the balance of the experimental trials. Latin square counterbalancing
The remaining two factors in the design (9 X 2) were nested was used to ensure that each face-in-the-crowd condition was preceded
within face trials. The Position factor designated which of the and followed approximately equally as often by the other conditions.
nine faces was the discrepant face. The Face factor described Likewise, it was used to generate the order of position placements in the
the emotional gesture of the discrepant face. Two levels of dis- display matrix. The insertion of no-face trials into the counterbalanced
crepant face were nested within each level of Crowd. On each face trials was randomly determined. One complete order of all 108
of nine angry crowd trials, a happy face or a neutral face was trials and its reverse were used in the experiment. Subjects were ran-
domly assigned to one or the other order. Reaction times and error rates
embedded in the crowd. On each of nine happy crowd trials,
were the data submitted to analysis.
an angry or a neutral face was embedded in the crowd; and on
each of nine neutral crowd trials, a happy or an angry face was
embedded in a neutral crowd. Results
The experiment, then, w a s a 2 X 3 X 2 X 9 (Trial X Crowd X
No-face trials. In crowds in which no discrepant face was
Face X Position) within-subjects nested factorial design. The
present, subjects made more errors when surveying angry
analyses of the data were focused on specific comparisons test-
crowds than when surveying happy or neutral crowds (M =
ing two predictions. First, angry faces in happy crowds should
15.4%, M = 6.8%, and M = 5.6%). Specific comparisons expli-
be found more efficiently (lower reaction times and fewer errors)
cating the significant crowd main effect, F(2, 20) = 27.74, p <
than happy faces in angry crowds. Second, angry faces in neu-
.001, indicated that more false positives occurred in the angry
tral crowds should be found more efficiently than neutral faces
crowd condition than in the happy or neutral crowd conditions,
in angry crowds. A processing asymmetry involving happy
F(l,2l) = 8.06,p<.008,andfi(l,21) = 8.65, p < .007. Because
faces in neutral crowds and neutral faces in happy crowds was
of the difficulty of interpreting reaction time data on error tri-
not anticipated.
als, only data from trials on which a subject made correct re-
sponses were submitted to analyses of the reaction time data.'
Method The time required to survey and correctly report the absence of
Twenty-two undergraduates volunteered to participate as subjects in a discrepant face in the crowd was about the same for angry,
the experiment in partial fulfillment of psychology course require- happy, and neutral crowds (M = 2,384 ms, M = 2,219 ms, and
ments. On each of 108 trials, the subject's task was to discover the pres- M = 2,301 ms). In short, although it took about the same
ence or absence of a discrepant face in a crowd. A discrepant face was amount of time to survey uniformly angry, happy, and neutral
present in one half of the trials (face trials) and absent from the rest crowds, surveillance of angry crowds was more error-prone.
(no-face trials). Crowds were displayed using a four-channel Gerbrands Subjects were more likely to report incorrectly the presence of
tachistiscope. Five hundred milliseconds priorto exposure of the crowd a discrepant face in angry crowds than in happy or neutral
field, a fixation point was displayed in the center of a blank field. The
crowds.
crowd field replaced the fixation field. The crowd field was removed
Face trials. The results for face trials are summarized in Ta-
from view when the subject made a forced-choice (yes or no) verbal
ble 1. It should be evident that the predicted processing asym-
response indicating whether or not a discrepant face had been present.
Reaction time from crowd field onset to verbal response was tracked metries emerged from the data. Because of factor nesting, the
using a Gerbrands voice key. Subjects were given a standard speed-ac-
curacy instructional set: "Answer as rapidly as you can without losing
1
accuracy." Analyses of reaction times including both correct and error trials
Crowds. Each crowd consisted of nine black-and-white, full-face revealed a pattern of effects similar to that reported here, except the
photographs of nine different individuals. The phot 'aphs were ar- face-in-the-crowd effect was substantially weakened in neutral crowd
ranged in a 3 X 3 matrix that subtended a visual angle of approximately conditions. This suggested that angry faces and neutral faces used in the
5.04° in both height and width, centered in the visual field (i.e., the experiment may have been somewhat confusable.
FACE-IN-THE-CROWD EFFECT 919

Table 1 1976) that for infants, neutral faces had about the same atten-
Summary of the Effects of Crowd and Face on Mean Reaction tion-grabbing capacity as angry faces. This alternative is neither
Times and Error Rates for Face Trials devastating to the face-in-the-crowd effect nor particularly dis-
couraging of the hypothesized orientation of face-processing. It
Target face
suggests, however, that the facial trigger underlying the orienta-
Crowd/distract ors Angry Neutral Happy
tion may be broader, perhaps a mechanism in which novel sig-
nals have some of the properties of threat signals.
Mean reaction time (ms) The second alternative is more problematic. There may be
Angry 2,270 2,450 more variance across angry facial gestures than across neutral
Neutral 1,920 2,370 facial gestures and, in turn, more across neutral facial gestures
Happy 1,990 2,010 than happy facial gestures. More varied crowd faces (i.e., an
angry crowd) would better disguise the presence of a face show-
Mean error rate (%)
ing a discrepant emotion. On the other hand, a crowd of less
Angry 71.2 74.8 varied faces (i.e., a happy crowd) would lend itself to the discov-

Neutral 31.3 79.3 ery of an emotionally discrepant face. This explanation, then,
Happy 30.3 —
31.8
— could account for the findings without resorting to the differen-
tial signal value of emotional facial gestures. This alternative
not only accounts for the data just as well as the hypothesized
analysis took the form of a series of planned comparisons test- process—even as amended to incorporate novelty—but has the
ing face-crowd reciprocals (e.g., finding an angry face in a additional appeal of parsimony. It renders the findings of Exper-
happy crowd vs. finding a happy face in an angry crowd). Sub- iment 1 quite equivocal. In the remaining experiments, there-
jects took less time, F(l, 21) = 9.98, p < .008, and made fewer fore, crowds were constructed to ensure that variance across
errors, F(l, 21) = 38.36, p < .001, finding an angry face in a faces in the crowds could not account for the findings.
neutral crowd than finding a neutral face in an angry crowd.
Likewise, subjects found angry faces in happy crowds faster, Experiment 2
f(l,2l) = 8.36, p < .008, and with fewer errors, F(\, 21) =
In Experiment 2, the distractor faces (crowd) as well as the
52.80, p < .001, than happy faces in angry crowds. These pre-
target face (discrepant emotional gesture) were photographs of
dicted asymmetries tended to support the contention that a dis-
the same individual. The crowd was comprised of three identi-
crepant angry face in a nonangry crowd attracts attention and
cal photographs of the individual posing either a happy or an
that angry faces in a crowd distract attention from a discrepant
angry facial gesture. The target face was a photograph of the
nonangry face. But the overall pattern of data left this support
same person posing an angry face in a happy crowd or posing a
somewhat equivocal. Angry faces were found no faster, F(l,
happy face in an angry crowd. A face-in-the-crowd asymmetry
21) < 1, p < .66, and with no fewer errors, F(\, 21) < 1, p <
obtained here obviously could not be dismissed in terms of the
.71, than neutral faces in the same happy crowds. Similarly,
heterogeneity of angry crowds relative to happy crowds. Be-
happy faces in neutral crowds were found in about the same
cause we used a threshold paradigm in Experiment 2, the face-
time, F( 1, 21) = 2.01,p < .\\, with about the same error rate,
in-the-crowd prediction tested was that the threshold to correct
F(l,2l) = 1.18,p<.29, as they were in angry crowds. In short,
location of an angry face in a happy crowd would be lower than
a processing asymmetry emerged here also. Neutral faces in
the threshold to correct location of a happy face in an angry
happy crowds were found faster, F(l, 21) = 9.98, p < .005, and
crowd.
with fewer errors, F( 1,21) = 60.96, p< .001, than happy faces
in neutral crowds.
Method

Discussion Eleven undergraduates volunteered to participate as subjects in par-


tial fulfillment of psychology course requirements. The experiment was
The predicted face-in-the-crowd effect emerged from the designed as a 2 X 2 (Face in the Crowd X Person) within-subjects facto-
data. Angry faces were found more efficiently in neutral crowds rial. As in Experiment 1, subjects were exposed to a series of tachisto-
than were neutral faces in angry crowds. Angry faces were scopically displayed crowds. In one half of the trials, a happy face was
found more efficiently in happy crowds than were happy faces embedded in an angry crowd. In the remaining trials, an angry face
in angry crowds. Alone, these findings tend to support the hy- was embedded in a happy crowd. Each crowd was comprised of four
pothesized threat-detection orientation of face-processing. photographs of the same individual: three identical photographs posing
one emotion (happy or angry) and one posing the other. Black-and-
However, the finding that these predicted processing asymmet-
white photographs of two males (Person) were used. Order of exposure
ries were of about the same magnitude as the unpredicted neu-
to face-crowd condition and to person was randomly determined for
tral-happy asymmetry is disquieting. This unexpected result
each subject.
raises at least two alternative interpretations of the data.2 First, The faces were displayed in the same manner as in Experiment 1. The
angry and neutral faces may be encountered less frequently four faces in each display were placed in the four corners of the matrix
than happy faces. The face-in-the-crowd effect here, then, may
derive not from facial-gestural-threat detection but rather from
2
facial-gestural-novelty detection. Schwartz et al. (1985) have We are indebted to two reviewers for making these hypotheses ex-
offered a similar explanation for the finding (LaBarbera et al., plicit.
920 CHRISTINE H. HANSEN AND RANALD D. HANSEN

used in Experiment 1. Each trial consisted of a centrally located fixation der, 1977; Treisman, 1982; Treisman & Paterson, 1984; Treis-
point exposed for a duration of 500 ms against a blank field, followed man & Souther, 1985). If faces can be preattentively searched
by the crowd field. Crowd-field offset was followed by a 50-ms interstim- in parallel for facial threat features, the latency to discovery of
ulus interval (a blank field) and a 100-ms exposure to a scrambled-letter
an angry face in a happy crowd should not be appreciably in-
pattern mask. The masking stimulus consisted of four scrambled-letter
fluenced by the number of happy-face distractors in the crowd.
patterns presented at the positions used for the crowd faces. The place-
The angry face should pop out of the crowd (cf. Treisman &
ment of the discrepant face on any given trial was randomly determined
for each subject by the computer program. Six replications of each face- Souther, 1985) with about the same latency in small and large
crowd condition for each person condition were used in each of the four happy crowds. However, preattentive search of an angry crowd
matrix locations, resulting in a total of 96 trials. would not render one of the faces distinctive. Finding a happy
The subject's task on each trial was to identify verbally the location face in an angry crowd, then, would require a serial attentive
of the discrepant face: a forced choice of upper right, upper left, lower search through the crowd. The number of angry-face distractors
right, or lower left. The necessary exposure duration of the crowd field would dramatically influence the latency to discovery of a dis-
for correct location (location threshold) was the dependent measure. crepant happy face. The face-in-the-crowd effect would be mag-
The location thresholds for the four Face-Crowd X Person conditions nified as the size of the crowd was increased.
were determined (in 10-ms steps) by the Best PEST algorithm (Lieber-
Experiment 3 was conducted as an additional test for the exis-
man & Pentland, 1982; Purcell & Stewart, 1986), implemented in Pas-
tence of the face-in-the-crowd effect and as an initial step to-
cal for a four-alternative forced choice. Threshold values in each of the
ward uncovering the mechanism of its production. In this ex-
four experimental conditions, then, were based on 24 trials (six replica-
periment, subjects were exposed to crowds of different sizes:
tions at each of the four corner positions).
either four or nine faces. The crowds were either happy or angry.
An angry face appeared in one half of the happy crowds and
Results and Discussion a happy face appeared in one half of the angry crowds. As in
The face-in-the-crowd effect was significant, F(\, 10) = Experiment 2, all of the faces in the crowd were identical photo-
12.73,p < .005. So, too, was the person effect, F(l, 10) = 24.16, graphs of the same individual. Subjects were asked to determine
p < .001. The interaction was not, F(l, 10) = 1.62, p < .23. whether a discrepant face was present in the crowd.
The person effect indicated that thresholds to location of one If, as hypothesized, faces are preattentively searched in parallel
person's discrepant face were lower than those for the other, for features of facial threat, an angry face in a happy crowd would
pop out and not require an extensive, time-consuming, serial at-
probably because of that stimulus person's greater expressive-
ness. The face-in-the-crowd effect supported the prediction that tentive search to discover its presence. However, a happy face in an
discrepant angry faces were easier to locate. Simple effects anal- angry crowd would not be preattentively distinctive; it would not
pop out, and discovering its presence would require a serial search
yses revealed that for both stimulus persons, thresholds to locate
of the faces. This construction of the mechanism underlying the
an angry face in a happy crowd were lower than thresholds to
production of the face-in-the-crowd effect led to the predictions
locate a happy face in an angry crowd, F(l, 10) = 11.94, p <
tested in Experiment 3. First, in crowds of any given size, the la-
.001 (M = 657.3 ms and M = 742.7 ms) for one stimulus person
tency to report the presence of an angry face in a happy crowd
and W, 10) = 6.95, p< .02 (M= 458.2 ms and M = 628.2 ms)
should be less than that to report the presence of a happy face in
for the other.
an angry crowd. Second, the latency to report the presence of an
These data were more persuasive than those of Experiment
angry face in happy crowds should not be substantially influenced
1. With the crowds comprised of identical distractors, the ad-
by the number of happy-face distractors in the crowd. Third, the
vantage of an angry face in a happy crowd over a happy face in
latency to report the presence of a happy face in angry crowds
an angry crowd was clearly evidenced. The data suggest that an
should be greatly influenced by the number of angry-face distrac-
angry target face, which better attracts attention in a crowd of
tors in the crowd: the larger the crowd, the longer the latency. Fi-
happy faces, also better serves to distract attention from a happy
nally, we would expect a fourth effect if subjects come to recognize
face. This should be interpreted cautiously, for although these
that a "pop-out" is contingent on the presence of an angry face in
data argued strongly in favor of the existence of the predicted
a happy crowd. If this contingency is recognized, subjects may
face-in-the-crowd effect, they were mute with regard to the pro-
adopt a simplifying strategy. Notably, when no pop-out occurs,
cesses underlying its production. It was toward defining these
subjects need only establish that the crowd is happy in order to
processes that Experiment 3 was conducted.
execute a correct no response. In this case, the latency to report the
absence of angry faces in happy crowds would not be appreciably
Experiment 3 influenced by the number effaces in the crowds. This type of effect
was noted explicitly by Treisman and Souther (1985) and is quite
The face-in-the-crowd effect was predicted from the hypothe-
apparent in the data reported by Schneider and Shiffrin (1977,
sis that face-processing is oriented toward threat detection.
Experiment 2). Unlike the pop-out of angry faces in happy crowds,
Theoretically, this could be achieved if faces were preattentively
however, this simplifying strategy would evolve over experimental
processed in parallel for features of facial threat. A parallel
trials. In short, the latency to report the absence of an angry face
search, not necessitating attentional shifts, could be completed
in a happy crowd should be refractory to crowd size only on later
more rapidly than a serial attentive search effaces. More telling,
trials.
a preattentive, parallel search for a feature would be relatively
insensitive to distraction whereas an attentive search, entailing Method
serial shifts of focus, would be susceptible to distraction (Bargh, Fourteen subjects volunteered to participate in the experiment, con-
1982, 1984; Bargh & Pietromonaco, 1982; Shiffrin & Schnei- structed a s a 2 x 2 x 2 x 2 (Matrix X Crowd X Trial X Stimulus Person)
FACE-IN-THE-CROWD EFFECT 921

mixed factorial design. The experiment was comprised of 288 trials. Face trials. Simple effects analyses revealed that, for face tri-
The lachistoscopic procedures were the same as those used in Experi- als, the matrix size and the crowd main effects, F(l, 13) = 33.73,
ment 1. On one half of the trials, subjects were exposed to a four-face p<.001,and.F(l,13) = 167.13, p< .001, as well as the Matrix
crowd, arranged in a 2 X 2 matrix as described in Experiment 2. On the
Size X Crowd interaction, F(l, 13) = 30.24, p < .001, were all
remaining one half of the trials, subjects were exposed to a nine-face
significant. Angry faces in happy crowds were discovered faster
crowd arranged in a 3 X 3 matrix as described in Experiment 1. On
one half of the trials, subjects were exposed to an angry crowd; on the than happy faces in angry crowds, both when the target ap-
remaining trials they were exposed to a happy crowd. All of the faces in peared with three distractors, F( 1, 13) = 74.78, p < .001 (M =
all of the crowds seen by any one subject were identical photographs of 874ms and M= 1,173 ms), and when the target appeared with
the same individual. Subjects were randomly assigned to see crowds eight distractors, F(\, 13)= 170.58,p< .001 (M= 910ms and
made up of identical photographs of one of the two stimulus persons, M = 1,426 ms). More important, a happy target face embedded
one man and one woman, used in the experiment. Stimulus Person, in a crowd with eight angry-face distractors required signifi-
then, was a between-subjects factor. cantly more time to discover than a happy face embedded in a
On one half of the trials, one of the faces in the crowd was a discrepant crowd with three angry-face distractors, F( 1, 13) = 44.82, p <
target face. One half of the angry crowds had a happy target face and
.001. However, the time to discover an angry face appearing in
one half of the happy crowds had an angry target face. The angry target
a crowd with eight happy-face distractors was no longer than
face was the same photograph used to create the angry crowd, and the
happy target face was the same photograph used to create the happy the time to discover an angry face appearing in a crowd with
crowd. The subject's task was to report the presence or absence (yes or three happy-face distractors, F(l, 13) = 2.30,p < .15. This pat-
no) of a discrepant face in the crowd on each trial. The target angry face tern, of course, matched the predictions and tended to support
was rotated through all four positions of the small crowd matrix nine the hypothesis that angry faces pop out from happy crowds as a
times and through all nine positions of the large crowd matrix four result of preattentive processing, but happy faces in angry
times. Likewise, the happy target face was rotated nine times through crowds can be discovered only after an attentive serial search of
all four positions of the small crowd matrix and four times through the the faces in the crowd.
nine positions of the large crowd matrix. As a result, an angry target No-face trials. The crowd matrix size and the crowd main
face was embedded in a crowd of happy distractor faces in 72 trials:
effect, F(l, 13) = 24.32, p < .001, and F(l, 13) = 52.08, p <
in 36 trials it appeared with three happy distractors and in 36 trials it
.001, as well as the Matrix x Crowd interaction, F(l, 13) =
appeared with eight happy distractors. Likewise, a happy target face was
116.15, p < .001, also were significant for no-face trials. It took
embedded in a crowd of angry distractor faces on 72 trials—36 trials in
each crowd size. The remaining 144 trials in the experiment were an longer to report the absence of a discrepant face in large angry
equivalent number of angry crowds and happy crowds without target crowds than in small angry crowds, F[\, 13) = 58.41, p < .001
faces, half of each being crowds of four and half being crowds of nine. (M = 1,774 ms and M = 1,325 ms). In contrast, it took no
Trial order was randomly determined. One order of trials and its reverse longer to report the absence of a discrepant face in crowds of
were used in the experiment. Subjects were randomly assigned to order. nine than it did in crowds of four happy faces, F(\, 13)= 1.38,
Trials were run in three blocks of 72 trials with a 2-min interblock rest p < .26 (M = 1,351 ms and M = 1,292 ms). As noted earlier,
period. Trials on which subjects made errors were rerun at the end of this was the pattern anticipated if subjects became aware of the
the block in which they occurred. Subjects' responses and response late-
pop-out contingency for an angry face in happy crowds. How-
ncies were recorded as described in Experiment I.
ever, recall also that subjects' reliance on the absence of a pop-
out in happy crowds to simplify the detection of no angry target
Results
face was predicted to emerge only on later trials. To test this
The analyses fully supported the predictions. The 2 X 2 X prediction, mean reaction time from the first 10 trials in each
2 x 2 analyses of variance (ANOVAS) conducted on error rates condition (early) and the last 10 trials in each condition (late)
and reaction times indicated that Stimulus Person produced no were submitted t o a 2 x 2 X 2 X 2 (Time X Trial X Matrix X
effects, so the data were collapsed across this factor. Error rates Crowd) ANOVA. The focus of this analysis was on the four-way
were acceptably low. The crowd matrix size main effect was sig- interaction which, as expected, was significant, F(\, 13) = 5.77,
nificant, F(\,l3) = 6.6l,p< .03, and indicated that more errors p < .04. Simple effects analyses revealed the anticipated pattern
were made on large crowd trials than on small crowd trials of effects. On early trials, it took longer to report the absence of
(M = 2.9% and M = 1.7%). Bear in mind, however, that trials an angry face in a happy crowd of nine than in one of four, F( 1,
on which subjects made errors were rerun, so reaction time data 13) = 24.61, p<. 001 (M= 1,693ms and M= 1,388ms). On
from error trials were excluded from the reaction time analyses. late trials, it took about the same amount of time to report the
Analysis of the reaction time data indicated that the main absence of an angry face in a crowd of nine as in a crowd of
effects due to trial, matrix, and crowd were all significant, F( 1, four,fl(l, 13)= l , p < . 8 9 ( M = 1,225 m s a n d M = 1,217ms).
13)= 141.50, p<.001; f X l , 13) = 31.72, p < .001;/^1, 13) = In short, as expected from the hypothesis, the use of the absence
152.51, p < .001, respectively. Further, the Trial X Matrix, the of a pop-out in a happy crowd as a diagnostic for the absence of
Trial x Crowd, and the Matrix X Crowd interactions were all an angry target face emerged only on later trials. Of course, it
significant, F(l, 13) = 7.11,p < .02;F(l, 13) = 24.01, p< .001; should be noted that the pop-out of angry faces in happy crowds
andf(l, 13)= 105.17, p<. 00 ^respectively. All of these effects, was evident in the absence of a crowd effect on both early trials,
however, tended to be explained by the significant Trial X Ma- P(l, 13) = 2.29, p<. 16, and late trials,^!, 13)= l , p < . 7 7 .
trix Size X Crowd interaction, F(l, 13) = 13.72, p < .003. Sim- The contrast across experimental conditions is quite evident
ple effects analyses were used to explicate this interaction. The in Figure 1. When the size of the angry crowd was increased,
mean reaction times underlying this effect are illustrated in Fig- the time to report the absence of a happy target face increased
ure 1. by about 90 ms for each additional distractor. Increasing the
922 CHRISTINE H. HANSEN AND RANALD D. HANSEN

Number of Distractors (on Face-Present Trials)/


1800 Crowd Size (on Face-Absent Trials)
Three Distractors/
1700 Four Faces
Eight Dish-actors'
1600 Nine Faces

1500

1400

1300

1200

1100

1000

900

800

Happy Crowd Angry Crowd Happy Crowd Angry Crowd


Angry Face Happy Face Angry Face Happy Face
Present Present Absent Absent
Face Trials No Face Trials

Figure 1. The effects of crowd size on time to report the presence of an angry face in a happy crowd, the
presence of a happy face in an angry crowd, and the absence of target faces in happy and angry crowds.

size of the angry crowd lengthened the time to report the pres- crowds were found more efficiently than neutral faces in angry
ence of the happy target face by about 52 ras for each additional crowds. In Experiment 2, thresholds to correct location of an
distractor. This suggests that subjects tended to engage in an angry face in a happy crowd were lower than thresholds to cor-
exhaustive serial search of angry crowds, searching the entire rect location of a happy face in an angry crowd. In Experiment
crowd before reporting no target face but terminating the 3, it took less time to determine the presence or absence of an
search when a happy target face was found (Sternberg, 1966; angry face in happy crowds than of a happy face in angry
Treisman & Souther, 1985). In contrast, increasing the size of crowds, and the asymmetry of this face-in-the-crowd effect in-
the happy crowd lengthened the time to report the presence of creased as the size of the crowd was increased.
an angry target face by about only 8 ms and the time to report These findings, then, also support the hypothesis from which
the absence of an angry target by about 12 ms for each happy the face-in-the-crowd effect was predicted: that faces can be pre-
distractor added. Each happy distractor increased reaction attentively processed for features of facial threat. The nature
times more on angry-face-present trials than on angry-face-ab- and consequences of the hypothesized preattentive processing
sent trials in a ratio of 1.5:1. should be noted. Detection and location of a feature can be ac-
This implies that if subjects became aware of the pop-out con- complished preattentively, whereas discrimination requires at-
tingency for happy crowds, they did not trust it sufficiently to tentional processing (Bergen & Julesz, 1983; Cheesman & Mer-
report the absence of a target without at least a cursory search ikle, 1986; Sagi & Julesz, 1985a, 1985b). An angry face could
of the crowd. This seems reasonable because in this experiment, be detected and located in a happy crowd, but its content—
all of the experimental conditions were interspersed. The ab- that it was, in fact, an angry face—could not be discriminated
sence of a pop-out was variably mapped (cf. Shim-in & Schnei-
preattentively. The consequence of preattentive face processing
der, 1977) to responses of present (i.e., a happy face in an angry
from which the face-in-the-crowd effect derived, then, would be
crowd) and absent (i.e., no angry target face in a crowd of happy
a shift of attention to a preattentively located point in the
faces). In short, given no pop-out, before reporting the absence
crowd. Given time to complete an attentional shift to that point,
of a target subjects probably engaged in a minimal search of the
the target could be seen as an angry face. And, indeed, the data
crowd to ensure that it was comprised of happy faces.
obtained, particularly in Experiment 3, are consistent with the
contention that angry faces pop out of a happy crowd (Treis-
Discussion man, 1982; Treisman & Gelade, 1980; Treisman & Paterson,
The data from these three experiments established the exis- 1984; Treisman & Souther, 1985).
tence of the predicted face-in-the-crowd effect. In Experiment It is important to note that our findings did not address the
1, angry faces in happy crowds were found more efficiently than nature of the feature that is preattentively processed. The face,
happy faces in angry crowds. Likewise, angry faces in neutral after all, is a complex stimulus; amazingly complex, it would
FACE-IN-THE-CROWD EFFECT 923

seem, to be considered a perceptual unit, a feature having an used to address this critical issue. Facial expressions are mim-
identity with a "specialized population of detectors" (Treisman icked (Dimberg, 1982,1986; McHugo, Lanzetta, Sullivan, Mas-
& Paterson, 1984, p. 13) or a basic perceptual element rather ters, & Englis, 1985; Ohman, 1986) and, hence, subjects' mim-
than a molecular conjunction of features. Clearly the face-in- icry might be used as a psychophysiological measure of emotion
the-crowd effect could derive from preattentive detection of a (cf. Cacioppo & Petty, 1979; Cacioppo, Petty, Losch, & Kim,
local facial feature distortion unique to anger. This prediction 1986) that is potentially independent of attentional processing.
could be derived from the texton theory (Julesz, 1981; Sagi & It would be interesting to discover, for example, whether the
Julesz, 1985a, 1985b) that predicts that differences in some lo- threshold to mimicry of an angry face precedes or occurs after
cal features (textons) can be preattentively detected, but detect- thresholds to its detection and to identification.
ing the conjunction of local features requires attentive process- Further, the conjunction of our work with that of Agostinelli,
ing. Recent psychophysiological research (Cacioppo, Martzke, Sherman, Fazio, and Hearst (1986) implies that it might be
Petty, & Tassinary, 1987) indicated that the region of the brow fruitful to explore asymmetries in the efficiency of detecting,
is a probable candidate. Within this view, the face would be identifying, and mimicking a face in a happy crowd that
construed as a conjunction of local features and the face-in-the- changes from angry to happy and a face in an angry crowd that
crowd effect based on preattentive detection of the angle of the changes from happy to angry. The predictions from these two
brow (local feature) produced by the activity of the corrugator lines of research converge. An angry face in a happy crowd is
supercilii muscle in anger but not in happiness. In this case, the face in the crowd most likely to be given attentional process-
of course, preattentive face processing is reduced to stimulus ing; it is the face most likely to result in a highly accessible me-
feature detection, and although it is interesting to speculate morial representation (Agostinelli et al., 1986). Detection and,
about the parallel evolution or acquisition of nonface-specific particularly, identification (Agostinelli et al., 1986) of an angry
feature detectors and emotional facial gestures in which these face changing to a happy face should be rather efficient—for the
features happen to inhere, it is probably not particularly fruitful enemy who blends into the friendly crowd can be treacherous,
to do so. Nevertheless, it is too early to dismiss the hypothesis indeed.
that emotional facial displays are perceptual elements. If the
human perceptual system has the capacity to be mapped to
References
complex stimuli, genetically available (cf. Hubel & Wiesel,
1968) or socially acquired (e.g., Shiffrin & Schneider, 1977), Agostinelli, G., Sherman, S. J., Fazio, R. H., & Hearst, E. S. (1986).
maps of faces would be high on the list of probable candidates Detecting and identifying change: Additions versus deletions. Journal
(Darwin, 1872/1904; Zajonc, 1985). of Experimental Psychology: Human Perception and Performance,
It also is important to be precise about how preattentive pro- 12, 445-454.
cessing could contribute to the observed face-in-the-crowd Bargh, J. A. (1982). Attention and automaticity in the processing of self-
relevant information. Journal of Personality and Social Psychology,
effects. Response latencies were sufficiently long to suggest that
43, 425-436.
both preattentive and conscious attentive processes play roles
Bargh, J. A. (1984). Automatic and conscious processing of social infor-
in responses. If subjects had an attentional search strategy—if
mation. In R. S. Wyer, Jr., & T. K. Srull (Eds.), Handbook of social
they wished to be certain of their identification—they would
cognition (Vol. 3, pp. 1-43). Hillsdale, NJ: Erlbaum.
have had to serially survey a minimum of three faces in any Bargh, J. A., & Pietromonaco, P. (1982). Automatic information pro-
crowd. The preattentive advantage accrued to angry faces in cessing and social perception: The influence of trait information pre-
happy crowds, then, represents the likelihood that the angry tar- sented outside of conscious awareness on impression formation.
get face in a happy crowd was the first or at least an early focus Journal of Personality and Social Psychology, 43, 437-449.
of attention. Further research will have to be conducted using Barrera, M. E., & Maurer, D. (1981). The perception of facial expres-
convergent paradigms (e.g., Treisman, 1982; Treisman & sions by the three-month-old. Child Development, 52,203-206.
Beck, J. (1983). Texture segregation. In J. Beck (Ed.), Organization and
Souther, 1985) to isolate the preattentive components from the
representation in perception (pp. 285-317). Hillsdale, NJ: Erlbaum.
attentive components of the face-in-the-crowd effect and con-
Bergen, J. R., & Julesz, B. (1983). Parallel versus serial processing in
clusively demonstrate that facial features acquire signal value
rapid pattern discrimination. Nature, 303,696-698.
at the level of preattentive processing.
Buck, R. (1984). The communication of emotion. New York: Guilford
However, documenting the preattentive processing of faces Press.
would be only the first step. Clearly, one larger issue raised by Cacioppo, J. T, & Petty, R. E. (1979). Attitude and cognitive response:
our data is whether the face-in-the-crowd effect is the result of An electrophysiological approach. Journal of Personality and Social
processes more appropriately labeled cognitive or emotive. Psychology, J7. 2181-2199.
Most researchers would agree that preattentive processing has Cacioppo, J. T., Martzke, J. S., Petty, R. E., & Tassinary, L. G. (1987).
the capacity to direct or call attention to some aspect of a stimu- Specific forms of facial EMG response index emotions during an in-
terview: From Darwin to the continuous flow Hypothesis of affect-
lus array (Cheesman & Merikle, 1985, 1986; Sagi & Julesz,
laden information processing. Unpublished manuscript, University of
1985a, 1985b; Shiffrin & Schneider, 1977; Treisman & Gelade,
Iowa, Iowa City.
1980; Treisman & Paterson, 1984; Treisman & Souther, 1985).
Cacioppo, J. T., Petty, R. E., Losch, M. E., & Kim, H. S. (1986). Electro-
Other researchers seem to claim more for preattentive process- myographic activity over facial muscle regions can differentiate the
ing: that emotion may occur preattentively (Kunst-Wilson & valence and intensity of affective reactions. Journal of Personality and
Zajonc, 1980; Seamon, Brody, & Kauff, 1983; Seamon, Marsh, Social Psychology, 50.260-268.
& Brody, 1984; Zajonc, 1980). A thorough investigation of the Cheesman, J., & Merikle, P. M. (1985). Word recognition and con-
processing underlying the face-in-the-crowd effect might be sciousness. In D. Besner, T. G. Waller, & G. E. MacKinnon (Eds.),
924 CHRISTINE H. HANSEN AND RANALD D. HANSEN

Reading research: Advances in theory and practice (Vol. 5, pp. 311- Sackett, G. P. (1966). Monkeys reared in isolation with pictures as visual
352). New York: Academic Press. input: Evidence for an innate releasing mechanism. Science, 154,
Cheesman, J., & Merikle, P. M. (1986). Distinguishing conscious from 1468-1473.
unconscious perceptual processes. Canadian Journal of Psychology, Sagi, D., & Julesz, B. (1985a). Detection versus discrimination of visual
40, 343-367. orientation. Perception, 14, 619-629.
Cheng, P. W. (1985). Restructuring versus automaticity: Alternative ac- Sagi, D., & Julesz, B. (1985b). "Where" and "what" of vision. Science,
counts of skill acquisition. Psychological Review, 92, 414-423. 228, 1217-1219.
Darwin, C. (1904). The expression of emotions in man and animals. Schneider, W., & Shiffrin, R. M. (1977). Controlled and automatic hu-
London: Murray. (Original work published 1872) man information processing: I. Detection, search, and attention. Psy-
Dimberg, U. (1982). Facial reactions to facial expressions. Psychophysi- chological Review, 84, 1-66.
ology, J9, 643-647. Schneider, W., & Shiffrin, R. M. (1985). Automatic and controlled pro-
Dimberg, U. (1986). Facial expressions as excitatory and inhibitory cessing revisited. Psychological Review, 91, 269-276.
stimuli for conditioned autonomic responses. Biological Psychology, Schwartz, G. M., Izard, C. E., & Ansul, S. E. (1985). The five-month-
22, 37-57. old's ability to discriminate facial expressions of emotion. Infant Be-
Ekman, P. (1984). Expression and the nature of emotion. In K. R. havior and Development, 8,65-77.
Scherer & P. Ekman (Eds.), Approaches to emotion. Hillsdale, NJ: Seamon, J. G., Brody, N., & Kauff, D. M. (1983). Affective discrimina-
Erlbaum. tion of stimuli that are not recognized: Effects of shadowing, masking,
Gratio\et,P.(l%65).Delaphysionomieetdesmouvememsd'expression and cerebral laterally. Journal of Experimental Psychology: Learn-
[About physiognomy and movements of expression]. Paris: Hetzel. ing, Memory, and Cognition, 9, 544-555.
Hansen, R. D., & Hansen, C. H. (1984). Toward a facial primacy theory Seamon, J. G., Marsh, R. L., & Brody, N. (1984). Critical importance
of emotion. Unpublished manuscript, Oakland University, Rochester of exposure duration for affective discrimination of stimuli that are
Hills, MI. not recognized. Journal of Experimental Psychology: Learning,
Homa, D., Haver, B., & Schwartz, T. (1976). Perceptibility of schematic Memory. andCognition, 10, 465-469.
face stimuli: Evidence for a perceptual gestalt. Memory and Cogni- Shiffrin, R. M., & Schneider, W. (1977). Controlled and automatic hu-
tion, 4, 176-185. man information processing: II. Detection, search, and attention.
Hubel, D. H., & Wiesel, T. N. (1968). Receptive fields and functional Psychological Review, 84, 127-190.
architecture of monkey striate cortex. Journal of Physiology, 195, Sternberg, S. (1966). High-speed scanning in human memory. Science,
215-243. 153. 652-654.
Izard, C. E. (1977). Human emotions. New York: Plenum Press. Tomkins, S. S. (1980). Affect as amplification: Some modifications in
Julesz, B. (1981). Textons, the elements of texture perception and their theory. In R. Plutchik & H. Kellerman (Eds.), Emotion: Theory, re-
interactions. Nature, 290, 91-97. search and experience (pp. 141 -164). New York: Academic Press.
Kahneman, D., & Treisman, A. (1983). Changing views of attention Treisman, A. (1982). Perceptual grouping and attention in visual search
and automaticity. In M. Kubovy & J. R. Pomerantz (Eds.), Varieties for features and for objects. Journal of Experimental Psychology: Hu-
of attention. New York: Academic Press. man Perception and Performance, 8,194-214.
Kunst-Wilson, W. R., & Zajonc, R. B. (1980). Affective discrimination Treisman, A., & Gelade, G. (1980). A feature-integration theory of at-
of stimuli that cannot be recognized. Science, 207, 557-558. tention. Cognitive Psychology, 12, 97-136.
LaBarbera, J. D., Izard, C. E., Vietze, P., & Parisi, S. A. (1976). Four- Treisman, A., & Paterson, R. (1984). Emergent features, attention, and
and six-month-old infants' visual responses to joy, anger, and neutral object perception. Journal of Experimental Psychology: Human Per-
expressions. Child Development, 47, 535-538. ception and Performance, 10, 12-31.
Leventhal, H. J. (1982). The integration of emotion and cognition: A Treisman, A., & Souther, J. (1985). Search asymmetry: A diagnostic for
view from the perceptual-motor theory of emotion. In M. S. Clark & preattentive processing of separable features. Journal of Experimen-
S. T. Fiske (Eds.), Affect and cognition (pp. 121-156). Hillsdale, NJ: tal Psychology: General. 114, 285-310.
Erlbaum. Van Santen, J. P. H., & Jonides, J. (1978). A replication of the face-
Lieberman, H. R., & Pentland, A. P. (1982). Microcomputer-based esti- superiority effect. Bulletin of the Psychonomic Society, 12, 378-380.
mates of psychophysical thresholds: The Best PEST. Behavior Re- Waynbaum, I. (1906). La physionomie humaine: Son mecanisme et son
search Methods & Instrumentation, 14,21-25. role social [The human physiognomy: Its mechanism and its social
Logan, G. D. (1979). On the use of a concurrent memory load to mea- role]. Paris: Alcan.
sure attention and automaticity. Journal of Experimental Psychology: Wilcox, B. M., & Clayton, F. L. (1968). Infant visual fixation on motion
Human Perception and Performance, 5, 189-207. pictures of the human face. Journal of Experimental Child Psychol-
McHugo, G. J., Lanzetta, J. T, Sullivan, D. G., Masters, R. D., & Englis, ogy. 6, 22-32.
B. G. (1985). Emotional reactions to a political leader's expressive Zajonc, R. B. (1980). Feeling and thinking: Preferences need no infer-
displays. Journal of Personality and Social Psychology, 49, 1513- ences. American Psychologist, 35, 151-175.
1529. Zajonc, R. B. (1985). Emotion and facial efference: A theory reclaimed.
Ohman, A. (1986). Face the beast and fear the face: Animal and social Science, 228. 15-21.
fears as prototypes for evolutionary analyses of emotion. Psychophys- Zajonc, R. B., & Markus, H. (1984). Affect and cognition: The hard
iology,23, 123-145. interface. In C. E. Izard, J. Kagan, & R. B. Zajonc (Eds.), Emotions,
Purcell, D. G., & Stewart, A. L. (1981). A face superiority effect. Bulle- cognition, and behavior, (pp. 73-102). Cambridge, England: Cam-
tin of the Psychonomic Society, 24, 118-120. bridge University Press.
Purcell, D. G., & Stewart, A. L. (1986). The face-detection effect. Bulle-
tin of the Psychonomic Society, 24, 118-120.
Purcell, D. G., Stewart, A. L., Botwin, M., & Kreigh, R. (1983). A face Received November 14, 1986
superiority effect: Hemiretina effects. Bulletin of the Psychonomic So- Revision received November 2, 1987
ciety, 21, 351. Accepted November 16, 1987

You might also like