Determine the cost factors (distance, weight, and destiny) that may influence the cost of goods.

Research Article

Prejudice From Thin Air The Effect of Emotion on Automatic Intergroup Attitudes David DeSteno,1 Nilanjana Dasgupta,2 Monica Y. Bartlett,1 and Aida Cajdric1

1 Northeastern University and

2 University of Massachusetts–Amherst

ABSTRACT—Two experiments provide initial evidence that spe-

cific emotional states are capable of creating automatic prej-

udice toward outgroups. Specifically, we propose that anger

should influence automatic evaluations of outgroups because of

its functional relevance to intergroup conflict and competition,

whereas other negative emotions less relevant to intergroup

relations (e.g., sadness) should not. In both experiments, after

minimal ingroups and outgroups were created, participants

were induced to experience anger, sadness, or a neutral state.

Automatic attitudes toward the in- and outgroups were then

assessed using an evaluative priming measure (Experiment 1)

and the Implicit Association Test (Experiment 2). As predicted,

results showed that anger created automatic prejudice toward

the outgroup, whereas sadness and neutrality resulted in no

automatic intergroup bias. The implications of these findings for

emotion-induced biases in implicit intergroup cognition in par-

ticular, and in social cognition in general, are considered.

Since the heyday of frustration-aggression and scapegoating theories

of prejudice (e.g., Dollard, Doob, Miller, Mowrer, & Sears, 1939),

social psychologists have recognized that intergroup relations, and the

stereotypes and prejudices that inevitably accompany them, are in-

fluenced by perceivers’ emotional states. As in the case of attitudes

more generally, emotions have been found to influence when, and to

what extent, people express positive or negative attitudes toward, and

beliefs about, members of in- and outgroups (Bodenhausen, Muss-

weiler, Gabriel, & Moreno, 2001; Fiske, 1998; cf. Petty, DeSteno, &

Rucker, 2001). For example, anger and happiness are known to en-

hance heuristic processing of social information that, in turn, ex-

acerbates stereotypic judgments of outgroups (Bodenhausen, Shep-

pard, & Kramer, 1994; Tiedens & Linton, 2001). Sadness, however,

has been shown to promote systematic processing of information that,

in turn, decreases stereotypic judgments (Lambert, Khan, Lickel, &

Fricke, 1997). These and similar findings have led to wide

acceptance of the view that specific emotions can influence people’s

beliefs about social groups.

It is important to note, however, that thus far, the growing corpus of

research on emotion and intergroup cognition has focused exclusively

on the effects of emotion on self-reported, or explicit, judgments of

social groups (for a review, see Bodenhausen et al., 2001). Such

judgments involve conscious deliberation and are, therefore, clearly

under perceivers’ voluntary control. Indeed, if people suspect that

incidental emotion may unduly influence an unrelated judgment, they

often correct for the perceived bias (Lambert et al., 1997; cf. DeSteno,

Petty, Wegener, & Rucker, 2000). Moreover, happy individuals, who

typically engage in heuristic processing, are able to process system-

atically when instructed to do so (Queller, Mackie, & Stroessner,

1996) or when counterstereotypic information motivates them to do so

(Bless, Schwarz, & Wieland, 1996). Such control, however, is not

available for all types of judgments, especially automatic ones (Banaji

& Dasgupta, 1998; Greenwald & Banaji, 1995). In the domain of

intergroup cognition, automatic attitudes stand as an unconscious

analogue to self-reported or conscious attitudes; that is, they rep-

resent evaluations of social groups whose initiation and modification

typically operate without volitional control (Fazio & Towles-Schwen,

1999; Greenwald & Banaji, 1995). Understanding the conditions that

lead to the formation and exacerbation of automatic prejudice is im-

portant not only because of its pervasiveness, but also because of

accumulating evidence that automatic prejudice does not remain

confined to mental life—it diffuses into people’s behavior toward

outgroup members (Dovidio, Kawakami, & Gaertner, 2002; Fazio,

Jackson, Dunton, & Williams, 1995; McConnell & Leibold, 2001).

We believe that people’s emotional states at the time of intergroup

judgment ought to influence their automatic evaluations of social

groups by moderating or even creating intergroup biases outside of

awareness. This hypothesis stems from a functional view of emotions

as phenomena designed to increase adaptive responding to en-

vironmentally significant stimuli (Damasio, 1994; Keltner & Gross,

1999; LeDoux, 1996). 1 From an adaptiveness standpoint, it seems

The first two authors contributed equally to this work. Address correspondence to David DeSteno, Department of Psychology, Northeastern University, Boston, MA 02115, e-mail: d.desteno@ neu.edu, or to Nilanjana Dasgupta, Department of Psychology, Tobin Hall, 135 Hicks Way, University of Massachusetts, Amherst, MA 01003, e-mail: dasgupta@psych.umass.edu.

1 The influence of emotion on cognition and behavior is theorized to produce

adaptive responses that prepare organisms to meet environmental challenges. However, the influence of emotions may also diffuse into new situations; that is, a preexisting or incidental emotion may influence interactions with a sub- sequent target (cf. Bodenhausen et al., 2001; Petty et al., 2001). Any biases that stem from the influence of incidental emotions on judgments of subsequent targets need not represent an adaptive response.

PSYCHOLOGICAL SCIENCE

Volume 15—Number 5 319Copyright r 2004 American Psychological Society at UNIV TEXAS AT TYLER on August 22, 2016pss.sagepub.comDownloaded from

 

 

reasonable to expect that specific emotions should facilitate people’s

ability to evaluate social groups quickly and automatically, as well as

slowly and carefully. We predict that to the extent that outgroups often

signify sources of conflict, competition, or blockage of goals (Brewer &

Brown, 1998; Neuberg & Cottrell, 2002), and to the extent that

emotions help individuals meet environmental challenges by acti-

vating goal-driven action tendencies (Frijda, 1986; LeDoux, 1996),

emotions that prepare organisms to meet challenges related to conflict

or competition (e.g., anger) should bias automatic intergroup evalua-

tions in accord with these functional goals.

EMOTION AND AUTOMATIC INTERGROUP ATTITUDES

Although no evidence directly bears on this hypothesis, findings from

three lines of research lend credence to the idea that emotion ought to

shape automatic attitudes toward social groups. First, cognitive neu-

roscience research has begun to identify subcortical structures in-

volved in automatic evaluative appraisals of social groups (Phelps

et al., 2000) and has found these structures to be reciprocally linked

to both cortical and subcortical regions of the brain involved in the

experience of emotion (Ochsner, Bunge, Gross, & Gabrieli, 2002).

Such reciprocal pathways suggest not only that automatic appraisals of

particular stimuli can trigger emotion, but also that extant emotional

states can influence subsequent appraisals. Given these linkages, it is

possible that an emotional state renders individuals more vigilant

against certain threats in the environment and that such vigilance

modulates subsequent automatic evaluations of relevant social stim-

uli. Because automatic evaluations facilitate rapid responses when

strategic analysis is unavailable, it seems reasonable to expect that

these responses may be an important medium through which emotions

allow organisms to meet environmental challenges; for example, cer-

tain emotion-driven automatic responses may act as the first line of

defense against threatening stimuli.

Second, the functional view of emotion readily extends into the

realm of intergroup relations. Recent work has begun to find that

appraisals of social groups evoke specific emotional states, goals, and

action tendencies that facilitate the successful negotiation of group

interactions (Mackie & Smith, 2002). Given this link between social

groups and specific emotions, it is conceivable that the experience of

such emotions, even when their source is incidental to intergroup

relations, may influence people’s perceptions of in- and outgroups in

accord with their functional significance.

Finally, for emotion-based moderation of automatic intergroup at-

titudes to occur, such attitudes must show some degree of flexibility.

Recent research has supported this view, providing evidence that

automatic beliefs and attitudes toward groups are not as immutable as

previously theorized, but rather are quite sensitive to external cues

such as social context (Dasgupta & Greenwald, 2001; Wittenbrink,

Judd, & Park, 2001). Consequently, emotion, given its context-rel-

evant signaling value, ought to act as an internal cue capable of

moderating automatic intergroup attitudes.

EXPERIMENT 1

We used a minimal-group procedure to provide an initial test of the

hypothesis that specific emotions can bias automatic attitudes toward

social groups. Minimal groups provided a clean assessment of the

primary hypothesis because participants had no preexisting attitudes

or emotional reactions toward them. Thus, any automatic preference

for one group over another could be interpreted as a new attitude. 2 To

the extent that outgroups signify sources of conflict and competition,

they may evoke feelings of anger and contempt (cf. Brewer & Brown,

1998; Neuberg & Cottrell, 2002). We propose that just as anger can

originate from current interactions with groups, so may incidental

feelings of anger from an unrelated situation affect automatic ap-

praisals of social groups in a subsequent situation because the emo-

tion signals a hostile environment and prepares individuals to act

accordingly. Specifically, we propose that incidental feelings of anger

are likely to increase automatic bias against an outgroup because

anger increases negativity toward the outgroup, decreases positivity,

or both. According to a functionalist perspective, the emergence of

outgroup bias should be specific to feelings of anger as opposed to

other negative emotions that are typically less relevant to intergroup

relations (e.g., sadness). To examine this hypothesis, we assigned

participants to minimal groups, induced one of three emotional states

(i.e., anger, sadness, neutrality), and then assessed participants’ au-

tomatic attitudes toward these groups with an evaluative priming task.

Method

Participants

A community sample of 87 New York City residents (50 females, 37

males) participated in exchange for $10.

Manipulations and Measures

Creation of Minimal Groups. To create minimal groups, we had par-

ticipants complete a bogus personality test in which they estimated

the frequency of various events (e.g., ‘‘How many people ride the New

York subway every day?’’). After they completed the test, the computer

ostensibly analyzed their responses and informed them that they were

either an ‘‘overestimator’’ or an ‘‘underestimator.’’ In reality, each

participant had been randomly assigned to one of these two groups. To

ensure that participants remembered their group membership

throughout the experiment, we instructed them to wear wristbands

designating their group: red wristbands for underestimators and blue

ones for overestimators. Participants were then shown pictures of 6

ingroup members and 6 outgroup members. 3 The backdrops of these

pictures were color-coded red for underestimators and blue for over-

estimators. By matching participants’ wristbands to the color of the

photographs, we sought to make group membership readily rec-

ognizable by a visually salient characteristic.

Assessment of Automatic Intergroup Attitudes. An evaluative priming

task was used to measure automatic intergroup attitudes (elements of

this priming task were borrowed from Fazio et al., 1995, and Payne,

2001). In the first block of 12 trials, participants categorized the

pictures of in- and outgroup members that they had seen previously

during the minimal-group assignment procedure as belonging to the

2 There is debate about whether evaluative biases captured by response la-

tency measures ought to be interpreted as personal attitudes or as cultural associations learned by exposure to particular stimulus pairings in the en- vironment (Karpinski & Hilton, 2001). The evaluative biases captured in the present experiments cannot be attributed to cultural associations given that the target stimuli were experimentally created minimal groups.

3 Individual pictures assigned to the overestimator and underestimator

groups were counterbalanced.

320 Volume 15—Number 5

Prejudice From Thin Air

at UNIV TEXAS AT TYLER on August 22, 2016pss.sagepub.comDownloaded from

 

 

ingroup (‘‘us’’) or outgroup (‘‘them’’). These images were presented one

at a time in a random order and later served as target stimuli. In the

second block of 12 trials, participants learned to classify valenced

words as good or bad; these later served as primes. The third block of

24 trials allowed participants to practice the standard evaluative

priming procedure. In each priming trial, several stimuli were pre-

sented in rapid succession in the following order: (a) an orienting

stimulus (n) for 500 ms, (b) a word prime for 200 ms, (c) a target

picture for 200 ms, and (d) a gray mask that stayed on screen until

participants pressed the appropriate key on a computer keyboard to

indicate whether the target picture belonged to their ingroup (‘‘us’’) or

outgroup (‘‘them’’). A 500-ms pause separated individual trials. Par-

ticipants were instructed to attend to all stimuli presented on screen,

but to categorize only the pictures. For each trial, the prime and target

were selected randomly from a pool of 12 primes and 12 targets. Once

practice was over, participants experienced the emotion induction

(described in the next paragraph). They then completed two blocks of

45 data-collection trials each, received a second round of the emotion

induction, and finally completed two more blocks of 45 data-collection

trials (total of 180 critical trials).

Emotion Induction. The emotion-induction task was introduced as a

study of people’s memories. Participants were asked to write in detail

about an autobiographical event from the past that had made them

very angry, very sad, or emotionally neutral (control condition). The

duration of the initial writing task was 4 min. Participants were told

that they would have an opportunity later to continue writing about

their memory. In the second round of the induction procedure, par-

ticipants were told to continue writing from where they had left off for

another 2 min. 4

Emotion Manipulation Check. Emotional states were assessed using

5-point adjective rating scales known to tap sadness and anger

(DeSteno et al., 2000). The anger subscale consisted of angry, an-

noyed, frustrated, and irritated (a 5 .90). The sadness subscale con- sisted of sad, gloomy, and down (a 5 .91).

Procedure

Participants arrived at the lab for what they thought was an experi-

ment on people’s personalities. They first completed the minimal-

group assignment task, which they believed to be a measure designed

to determine their personality type. Immediately following this ma-

nipulation, participants completed the evaluative priming task, which

served as a measure of their automatic attitudes toward the ingroup

and outgroup. This task was introduced as a measure of ‘‘hand-eye

coordination’’ that was allegedly necessary to serve as a baseline

because of individual differences in people’s speed of responding to

visually presented stimuli. As noted earlier, the emotion induction was

embedded before the first and again before the third data-collection

blocks of this priming task. After the priming task, participants

completed an emotion-manipulation check and were debriefed. All

data were collected and instructions and stimuli presented via com-

puter using MediaLab (Jarvis, 2002) and Inquisit (Draine, 2000).

Results and Discussion

Manipulation Check

The emotion manipulations were successful in producing the expected

3 (emotion-induction condition) � 2 (emotion rating) interaction, F(2, 85) 5 15.04, p < .001. That is, participants in the angry condition

reported more anger (M53.32) than sadness (M52.57), t(30)53.01,

p < .01, d 5 0.54; participants in the sad condition reported more

sadness (M 5 3.28) than anger (M 5 2.36), t(24) 5 3.35, p < .01,

d 5 0.67. Neutral participants reported low levels of both emotions

(Msadness 5 1.40, Manger 5 1.53).

Automatic Attitudes Toward Social Groups

A 3 (emotion) � 2 (prime) � 2 (target) mixed analysis of variance revealed that the experience of specific emotional states differentially

influenced automatic attitudes toward the target groups, as indicated

by the three-way interaction, F(2, 85)52.50, p5 .08 (see Fig. 1). 5 A

Prime � Target interaction emerged among angry participants, in- dicating that, as predicted, the outgroup became a strongly valenced

attitude object, F(1, 30)54.95, p5 .03, d50.57. 6 More specifically,

angry participants were slower to associate positive attributes than

negative attributes with the outgroup, t(30)52.35, p5 .03, d50.42.

There was no difference in the speed with which they associated

positive versus negative attributes with the ingroup (t < 1), indicating

a neutral evaluative stance toward this group. Moreover, as expected,

no intergroup bias (i.e., Prime � Target interaction) emerged for neutral or sad participants (Fs < 1.3). These data suggest that anger

exerted a functional influence on automatic attitudes and, in so doing,

created automatic prejudice where none had previously existed.

However, before placing confidence in this finding, we wanted to at-

tempt a cross-method replication, especially given that the omnibus

test of the three-way interaction did not reach the conventional level

of statistical significance.

EXPERIMENT 2

In this experiment, we used a different measure to assess the effect of

emotion on automatic attitudes—the Implicit Association Test (IAT).

We used the IAT for two reasons. First, it provided the opportunity to

conduct a cross-method validation of the findings of Experiment 1.

Evaluative priming and the IAT share several commonalities: (a) Both

tasks assume that if an attitude object evokes a particular evaluation,

it will facilitate responses to stimuli that are evaluatively congruent

versus neutral or incongruent, and (b) both tasks interpret response

facilitation as a measure of the strength of association between the

object and attribute (Bargh, Chaiken, Govender, & Pratto, 1992;

4 To avoid confounding the anger induction with the priming of information

related to intergroup conflict, we screened participants’ responses for memories that were intergroup in nature. None of the memories involved intergroup themes; all were interpersonal.

5 Analyses involving response latencies were conducted using log-trans-

formed values to normalize the distributions. For easier interpretation, how- ever, we present descriptive statistical information using the millisecond metric.

6 We were agnostic about whether increased intergroup bias would be driven

by less positivity or greater negativity toward the outgroup. Previous work on automatic prejudice in particular, and automatic evaluations in general, has typically relied on the existence of significant Prime � Target interactions, as opposed to absolute comparisons of response latencies across different types of trials, to indicate the existence of an evaluative bias because such individual comparisons can be compromised by confounding factors that differentially influence responses to specific types of stimuli (see Bargh, Chaiken, Govender, & Pratto, 1992; Duckworth, Bargh, Garcia, & Chaiken, 2002; Fazio et al., 1995; Glaser & Banaji, 1999; Klauer, Rossnagel, & Musch, 1997).

Volume 15—Number 5 321

D. DeSteno et al.

at UNIV TEXAS AT TYLER on August 22, 2016pss.sagepub.comDownloaded from

 

 

Dasgupta, McGhee, Greenwald, & Banaji, 2000; Fazio, Sanbonmatsu,

Powell, & Kardes, 1986; Greenwald, McGhee, & Schwartz, 1998).

However, there are procedural differences between these two tasks.

Thus, replicating Experiment 1 using the IAT would demonstrate the

robustness of the predicted effect. Second, some data suggest that

compared with priming techniques, the IAT may be more sensitive to

individual and group differences and somewhat more reliable across

time (Bosson, Swann, & Pennebaker, 2000). We therefore expected

that this task might be better able to capture the predicted pattern of

emotion-induced moderation of intergroup bias.

Method

Participants

Eighty-one students (51 females, 30 males) participated in this ex-

periment in partial fulfillment of requirements for a psychology

course.

Procedure and Measures

The procedure and measures used were identical to those of Experi-

ment 1 with two exceptions: Automatic attitudes were assessed using

an IAT instead of evaluative priming, and participants’ self-reported

attitudes toward the ingroup and outgroup were also measured to

ensure the success of the minimal-group manipulation.

In the IAT task, participants first completed three practice blocks

during which they categorized four types of stimuli (pictures rep-

resenting in- and outgroup members and positive and negative

words) using two designated response keys. Specifically, participants

classified (a) valenced words for 20 trials, (b) pictures of in- and

outgroup members for another 20 trials, and then (c) all four types of

stimuli simultaneously (20 trials). These practice tasks were coun-

terbalanced such that half the participants learned to categorize

ingroup and good stimuli using the same key and outgroup and bad

stimuli using a different key. The remaining participants learned the

opposite stimulus pairing. Participants then completed the first round

of emotion induction, which was followed by a data-collection block

of the IAT that was identical to the last practice block, only longer

(50 trials).

Next, additional practice was given so that participants could learn

to categorize stimuli in the combination opposite to what they had

learned before. During this practice, they first classified pictures of in-

versus outgroup members using response keys opposite to those they

had used previously (20 trials). Next, they classified all four types of

stimuli simultaneously such that, for example, those participants who

had previously paired ingroup with good and outgroup with bad

learned to associate ingroup with bad and outgroup with good (20

trials). Participants then completed another round of emotion induc-

tion to reinstantiate their feeling state, followed by a second data-

collection block of the IAT (50 trials). Next, participants reported their

attitudes toward the groups using a five-item, 7-point semantic dif-

ferential scale (unintelligent-intelligent, bad-good, unpleasant-pleas-

ant, dishonest-honest, awful-nice); their responses were averaged into

a single attitudinal index (a5.87). Participants’ emotional states were assessed at the end as in Experiment 1.

Results and Discussion

Manipulation Checks

The emotion manipulations were successful, as indicated by the

Emotion Induction � Emotion Rating interaction, F(2, 79) 5 13.14, p < .001. Participants in the sad condition reported more sadness

(M 5 3.63) than anger (M 5 2.98), t(25) 5 2.89, p < .01, d 5 0.57;

participants in the angry condition reported more anger (M 5 3.64)

than sadness (M 5 3.21), t(27) 5 2.46, p 5 .02, d 5 0.46. Neutral

Fig. 1. Reaction time in the evaluative priming task (Experiment 1) as a function of emotion, prime, and target. Error bars represent standard errors.

322 Volume 15—Number 5

Prejudice From Thin Air

at UNIV TEXAS AT TYLER on August 22, 2016pss.sagepub.comDownloaded from

 

 

participants reported low levels of both emotions (Msadness 5 1.57,

Manger 5 2.05).

Analysis of participants’ self-reported attitudes verified the success

of the group-assignment procedure. As in previous minimal-group

research, participants reported more positive attitudes toward their

ingroup (M54.90) than their outgroup (M54.46), F(1, 79)511.21,

p 5 .001, d 5 0.75. 7

Automatic Attitudes Toward Social Groups

Automatic attitudes were measured as the differential speed with

which participants classified outgroup with good stimuli and ingroup

with bad stimuli compared with the reverse combinations; larger

difference scores correspond to stronger bias against the outgroup

relative to the ingroup. Participants’ emotional states differentially

biased their intergroup evaluations, F(1, 79) 5 4.03, p < .05, d 5

0.54. 8 As shown in Figure 2, only participants in the angry condition

showed strong automatic prejudice against the outgroup and relative

preference for the ingroup, t(27)53.32, p < .01; those in the sad and

neutral conditions showed no intergroup bias (both ts < 1). Further

analyses revealed that the interaction effect was driven by slower

responses to outgroup 1 good/ingroup 1 bad classifications for par-

ticipants in the angry condition compared with those in the neutral

and sad conditions, F(1, 79) 5 3.91, p 5 .05. Response latencies for

ingroup 1 good/outgroup 1 bad classifications did not differ signifi-

cantly across emotion conditions (F < 1). Thus, as in Experiment 1,

anger created automatic outgroup bias where none had previously

existed.

GENERAL DISCUSSION