Content uploaded by Bernadette Park
Author content
All content in this area was uploaded by Bernadette Park
Content may be subject to copyright.
The Police Officer’s Dilemma:
Using Ethnicity to Disambiguate Potentially Threatening Individuals
Joshua Correll, Bernadette Park,
and Charles M. Judd
University of Colorado at Boulder
Bernd Wittenbrink
University of Chicago
Using a simple videogame, the effect of ethnicity on shoot/don’t shoot decisions was examined. African
American or White targets, holding guns or other objects, appeared in complex backgrounds. Participants
were told to “shoot” armed targets and to “not shoot” unarmed targets. In Study 1, White participants
made the correct decision to shoot an armed target more quickly if the target was African American than
if he was White, but decided to “not shoot” an unarmed target more quickly if he was White. Study 2
used a shorter time window, forcing this effect into error rates. Study 3 replicated Study 1’s effects and
showed that the magnitude of bias varied with perceptions of the cultural stereotype and with levels of
contact, but not with personal racial prejudice. Study 4 revealed equivalent levels of bias among both
African American and White participants in a community sample. Implications and potential underlying
mechanisms are discussed.
In February 1999, around midnight, four plain-clothes police
officers were searching a Bronx, New York, neighborhood for a
rape suspect. They saw Amadou Diallo, a 22-year-old West Afri-
can immigrant, standing in the doorway of his apartment building.
According to the police, Diallo resembled the suspect they were
tracking. When they ordered him not to move, Diallo reached into
his pants pocket. Believing he was reaching for a gun, the police
fired a total of 41 shots, 19 of which hit and killed Diallo. Diallo
was in fact unarmed. All four officers were later acquitted of any
wrongdoing in the case.
The police could not have known for certain that Diallo was
harmless. In the dark, they had ordered a potentially dangerous
man to freeze, and that man reached for something. If Diallo had
been armed, their decision to open fire would never have been
questioned. But the decision to shoot a man who later proved to be
unarmed did raise questions, one fundamental question in partic-
ular: Would the police have responded differently if Diallo had
been White? Perhaps Diallo would have been given the benefit of
the doubt, perhaps the order to freeze would have been repeated,
perhaps a slight delay in the decision to fire would have given the
officers time to recognize that this suspect was not reaching for a
gun. Though it is impossible to reach a definitive answer with
respect to Diallo’s case, the dilemma faced by these officers has
important consequences for cities nationwide and warrants a sys-
tematic investigation. It seems crucial to understand whether or not
the decision to shoot is influenced by the target’s ethnicity, and if
so, what this bias represents.
Social psychology has long held an interest in the way that
schemata, including expectancies about social categories like eth-
nicity, guide the interpretation of ambiguous information (Duncan,
1976; Hilton & von Hippel, 1990; Jacobs & Eccles, 1992; Rothbart
& Birrell, 1977; Sagar & Schofield, 1980). The quick and almost
effortless classification of a unique individual into a broad social
category (Brewer, 1988; Fiske, Lin, & Neuberg, 1999; Fiske &
Neuberg, 1990) may lead people to assume that traits generally
associated with the category also apply to this particular member.
Either in the absence of individuating information (Darley &
Gross, 1983; Locksley, Borgida, Brekke, & Hepburn, 1980; see
Hamilton & Sherman, 1994, for a review) or in spite of it (Beckett
& Park, 1995; Krueger & Rothbart, 1988), stereotypic associations
can influence an observer’s perceptions in a top-down fashion. A
stereotype, in essence, can function as a schema to help clarify or
disambiguate an otherwise confusing situation.
Of particular interest to the question of Diallo’s death is the
possibility that the officers’ decision to fire was influenced by the
stereotypic association between African Americans and violence.
The ambiguity of Diallo’s behavior (what was he reaching for?),
which ironically provides a justification for the officers’ decision,
may have set the stage for bias, prompting the officers to draw on
other sources of information, including stereotypes, in an effort to
understand what was happening. Duncan (1976) showed that the
same mildly aggressive behavior is perceived as more threatening
when it is performed by an African American than when it is
performed by a White person. A White person’s light push seems
like a violent shove when performed by an African American.
Sagar and Schofield (1980), following Duncan, presented 6th-
Joshua Correll, Bernadette Park, and Charles M. Judd, Department of
Psychology, University of Colorado at Boulder; Bernd Wittenbrink, Grad-
uate School of Business, University of Chicago.
This material is based on work supported by a National Science Foun-
dation graduate research fellowship awarded to Joshua Correll. Support for
this work also came from National Institute of Mental Health Grant
R01-45049 awarded to Bernadette Park and Charles M. Judd and from a
sabbatical award from the James McKeen Cattell Fund awarded to Berna-
dette Park. We thank Paul G. Davies for his invaluable contributions to the
conceptual development of this project, and Joshua Ingram and David M.
Deffenbacher for their assistance in its implementation.
Correspondence concerning this article should be addressed to Joshua
Correll, Department of Psychology, University of Colorado, Boulder,
Colorado 80309-0345. E-mail: jcorrell@psych.colorado.edu
Journal of Personality and Social Psychology Copyright 2002 by the American Psychological Association, Inc.
2002, Vol. 83, No. 6, 1314–1329 0022-3514/02/$5.00 DOI: 10.1037//0022-3514.83.6.1314
1314
grade boys with line drawings and verbal accounts of ambiguous
dyadic interactions, for example, two boys bumping into one
another in the hallway, or one boy borrowing a pencil from a
classmate without asking. To manipulate the ethnicity of the peo-
ple interacting, the researchers simply shaded in the drawings.
Like Duncan, they found that when an actor was depicted as
African American, rather than White, his behavior seemed more
mean and threatening to the participants. Sagar and Schofield
further found that this bias in perception was similar for both
White and African American participants. That is, the tendency to
see an African American’s behavior as more mean and threatening
than a White person’s did not depend on the observer’s ethnicity.
On the basis of this result, Sagar and Schofield argued that the bias
reflects not the internalization of anti-African American attitudes,
but rather the application of a widely known and cognitively
derived stereotype about the group to the particular target
individual.
Devine (1989) went on to demonstrate that the impact of eth-
nicity on interpretation could occur even without participants’
awareness. She asked participants to rate a target’s ambiguously
hostile behavior after subliminally priming them with words re-
lated to both the social category and the stereotype of African
Americans (but excluding words directly related to violence).
Participants who were primed with a greater number of these
words were more likely to interpret the behavior as hostile, even
though the target’s ethnicity was never mentioned. Lepore and
Brown (1997) primed only the social category of African Ameri-
cans (not the stereotype) and found that the effect of the primes on
interpretation of behavior was only evident among the more prej-
udiced participants. In all of these studies, the association between
the social category, African American, and the concept of violence
seems to lead participants to interpret an ambiguous target as more
dangerous.
Most recently, Payne (2001) demonstrated that participants were
faster and more accurate in distinguishing guns from hand tools
when they were primed with an African American face, as opposed
to a White face. Using Jacoby’s (1991; Jacoby, Toth, & Yonelinas,
1993) Process Dissociation Procedure, Payne then separated par-
ticipants’ errors into automatic and controlled components. The
magnitude of the automatic estimate represents the degree to
which the ethnicity of the prime influences participants’ decisions
when their ability to control that decision fails. Among participants
who were low in motivation to control prejudiced responding,
Payne found that greater prejudice was associated with a greater
automatic effect.
The primary goal of the current research was to carry this line of
inquiry one step further, investigating the effect of a target’s
ethnicity on participants’ decision to “shoot” that target. We
present data from a simplified videogame, which roughly simu-
lates the situation of a police officer who is confronted with an
ambiguous, but potentially hostile, target, and who must decide
whether or not to shoot. In the game, images of people who are
either armed or unarmed, and either African American or White,
appear unexpectedly in a variety of contexts. Unlike previous
research, this game requires participants to make a behavioral
shoot/don’t shoot decision similar to that of a police officer. And
unlike a sequential priming study (such as Payne, 2001), this game
simultaneously presents a target person’s ethnicity and the object
he is holding. A participant need not process ethnicity to determine
whether the target is armed. In spite of these differences, the
research reviewed above strongly suggests that interpretation of
the target as dangerous, and the associated decision to shoot, will
vary as a function of the target’s ethnicity. In Studies 1 and 2, we
test this basic prediction. In Studies 3 and 4, we make an initial
effort to understand the processes underlying this bias in the
decision to shoot.
Study 1
Method
Participants and Design
Forty undergraduates (24 female, 16 male) at the University of Colorado
at Boulder participated in this experiment in return for either $8 or partial
credit toward a class requirement.
1
One of the male participants was
Latino. All other participants were White. The study used a 2 ⫻ 2
within-subject design, with Target Ethnicity (African American vs. White)
and Object Type (gun vs. no gun) as repeated factors.
Materials
Using the PsyScope software package (Cohen, MacWhinney, Flatt, &
Provost, 1993), we developed a simplistic videogame that presented a
series of background and target images. The videogame used a total of 20
backgrounds and 80 target images. Twenty young men, 10 African Amer-
ican and 10 White, were recruited on college campuses to pose as models
for the targets. Each of these models appeared in the game four times, twice
as a target in the gun condition and twice as a target in the no-gun
condition, with a different object and in a different pose each time (five
basic poses were used in the game). There were four non-gun objects (a
silver-colored aluminum can, a silver camera, a black cell phone, and a
black wallet) and two guns (a silver snub-nosed revolver and a black 9-mm
pistol). Each of the objects, within condition, appeared equally often in
each of the five poses. The four target images for each model were
superimposed on randomly determined backgrounds, constrained so that
each background was used once in each of the four conditions and no target
appeared on the same background more than once. Background images
included an intentionally diverse assortment of photographs, such as train
station terminals, parks, hotel entrances, restaurant facades, and city side-
walks. No people appeared in any of the original background scenes.
Examples of the stimuli appear in Figure 1.
In total, there were 80 trials in the videogame, with 20 trials in each cell
of the 2 ⫻ 2 design created by crossing the ethnicity of the target with
whether the target held a gun or a non-gun. Each of the 80 trials began with
the presentation of a fixation point, followed by a series of empty back-
grounds, presented in slide-show fashion. The number of backgrounds on
a given trial was randomly determined, ranging from 1 to 4. The duration
of each was also random, ranging from 500 to 1,000 ms. The final
background in the series was replaced by the target image, created by
superimposing the target on the final background. From the perspective of
the participant, a man seemed to simply appear on the background. The
design of the game was intended to ensure that the participant never knew
when or where the target would appear in the background or when a
response would be required.
To play the game, the participant needed to decide as quickly as possible
whether the object the man was holding was a gun or not. If it was a gun,
1
Gender did not moderate any of the effects we report in this or
subsequent studies. In Study 3, there was a main effect of gender, such that
men had faster reaction times for all targets than did women, t(43) ⫽ 2.31,
p ⫽ .03, but this effect did not replicate in the other studies.
1315
THE POLICE OFFICER’S DILEMMA
the man posed an imminent danger, and the participant needed to shoot him
as quickly as possible by pushing the right button, labeled shoot,ona
button box. If he was holding some object other than a gun, he posed no
danger, and the participant needed to press the left button, labeled don’t
shoot, as quickly as possible. Participants were instructed to use separate
hands for each button and to rest their fingers on the buttons between trials.
Figure 1. Target and background example scenes from videogame. Color originals are available at
psych.colorado.edu/⬃jcorrell/tpod.html
1316
CORRELL, PARK, JUDD, AND WITTENBRINK
The game awarded and deducted points on the basis of performance. A hit
(correctly shooting a target holding a gun) earned 10 points, and a correct
rejection (not shooting a target holding some non-gun object) earned 5
points. A false alarm (shooting a target holding a non-gun) was punished
by taking away 20 points, and a miss (not shooting a target holding a gun)
resulted in our harshest penalty: a loss of 40 points.
2
This payoff matrix
represented an effort to partially, if weakly, recreate the payoff matrix
experienced by police officers on the street, where shooting an innocent
suspect is a terrible mistake (as in the case of Amadou Diallo), but where
the stronger motivation is presumably to avoid misidentifying an armed
and hostile target, which could result in an officer’s death. To minimize
nonresponse, the game assessed a timeout penalty of 10 points if partici-
pants failed to respond to a target within 850 ms. This time window was
selected to force participants to respond relatively quickly, while still
allowing enough time such that errors in the game would be minimized.
Participants’ decisions (“shoot” or “don’t shoot”) and their reaction times
were recorded for each trial. Each trial ended by giving participants
feedback on whether they had made the correct decision on that trial and
by showing them their cumulative point total.
Procedure
Participants, in groups of 1 to 4, were met by a male experimenter who
outlined the study as an investigation of perceptual vigilance, or the ability
to monitor and quickly respond to a variety of stimuli. A detailed set of
instructions for the videogame task followed, including the point values for
each of the outcomes. Participants were also informed that the people with
the first, second, and third highest scores in the study would receive a prize
($30, $15, and $10, respectively) and that 5 others, randomly selected from
participants with scores in the top 30%, would each receive $10. These
prizes were intended to make the payoff matrix personally meaningful.
Finally, participants were asked to pay attention to the faces of the targets,
because they would be tested on their ability to recognize the targets at the
end of the game. Participants then moved to individual rooms to play the
game.
At the conclusion of the game, participants were presented with a series
of 16 recognition trials in a paper-and-pencil task to determine whether
facial characteristics of the targets had been attended to. For each of the 16
faces, participants had to indicate whether they believed it was the face of
one of the targets that had been seen during the game or not. Half of the
presented targets had in fact been seen previously; half had not. Addition-
ally, half of the targets were African American and half were White.
Following the recognition task, participants were given a short question-
naire, which asked whether they valued the monetary incentives, whether
they remembered the point values for hits, misses, false alarms, and correct
rejections. Participants were then fully debriefed, with the experimenter
paying particular attention to alleviate any negative feelings aroused by the
game.
Results and Discussion
To analyze the resulting reaction times, we excluded all trials on
which the participant had either timed-out (i.e., failed to make a
decision in the allotted 850-ms window) or made an incorrect
response (e.g., shooting a target holding a non-gun). This resulted
in the exclusion of data from 7% of the trials across participants,
with a maximum of 20% of the trials for any one participant.
Response latencies on the remaining trials were log-transformed
and then averaged within subject across trials occurring in the
same cell of the 2 ⫻ 2 within-subject research design. An analysis
of variance (ANOVA) of the resulting mean latencies was then
conducted, treating Target Ethnicity (White vs. African American)
and Object Type (gun versus no gun) as within-subject factors.
This analysis revealed a significant main effect for Object, F(1,
39) ⫽ 244.16, p ⬍ .0001, and a significant Object ⫻ Ethnicity
interaction, F(1, 39) ⫽ 21.86, p ⬍ .0001. The resulting cell means
(converted back to the millisecond metric) appear in Table 1. As
these means reveal, participants were significantly faster at making
the correct decision to shoot, when the target held a gun, than the
correct decision to not shoot, when the target did not hold a gun.
More central to our predictions, the interaction suggests that the
speed of responding on gun versus no-gun trials depended on
target ethnicity. We decomposed this interaction by examining the
simple effects of ethnicity separately for the gun and no-gun trials.
Both were significant: Participants fired at an armed target more
quickly if he was African American than if he was White, F(1,
39) ⫽ 10.89, p ⬍ .005, and they decided not to shoot an unarmed
White target more quickly than an unarmed African American
target, F(1, 39) ⫽ 9.77, p ⬍ .005.
2
These point values should, objectively, create a bias to shoot: The two
“don’t shoot” options yield an average reward of –17.5 points, whereas the
“shoot” options yield a less aversive average of –5 points.
Table 1
Means (and Standard Deviations) for Reaction Times and Error Rates as a Function of Target
Ethnicity and Object Type (Studies 1, 2, and 3)
Study
Reaction times Errors per 20 trials
White
targets
Afr. Am.
targets
White
targets
Afr. Am.
targets
Study 1
Armed targets 554 (46) 544 (39) 0.70 (1.07) 0.40 (0.78)
Unarmed targets 623 (38) 634 (39) 1.23 (1.29) 1.45 (1.04)
Study 2
Armed targets 449 (23) 451 (28) 2.46 (1.83) 1.48 (1.38)
Unarmed targets 513 (32) 523 (38) 2.40 (2.76) 3.29 (2.87)
Study 3
Armed targets 550 (40) 539 (45) 0.76 (0.86) 0.49 (0.80)
Unarmed targets 607 (38) 620 (38) 0.33 (0.90) 0.65 (1.24)
Note. Afr. Am. ⫽ African American.
1317
THE POLICE OFFICER’S DILEMMA
We intentionally gave participants a long enough response win-
dow (850 ms) in this study to maximize correct responses to
examine effects on response latencies. And, as we suspected, the
proportions of errors were quite low, averaging 4% of the trials
across participants. Nonetheless, it is possible to examine the error
rates to see if they depended on Target Ethnicity, Object Type, or
their interaction (see mean error rates in Table 1). This analysis
revealed a main effect for Object, F(1, 39) ⫽ 32.31, p ⬍ .0001,
such that errors in the no-gun condition (i.e., false alarms) were
more frequent than errors in the gun condition (i.e., misses). The
interaction between Ethnicity and Object was also significant,
suggesting that the tendency to make more false alarms than
misses was more pronounced for African American targets than
for White targets, F(1, 39) ⫽ 7.68, p ⬍ .01. That is, whereas
participants tended to shoot unarmed targets more frequently than
they decided not to shoot armed targets, in general, this tendency
was stronger when the target was African American than when the
target was White. The simple effects were in the correct direction,
but not statistically significant. Participants were marginally more
likely to miss an armed target when he was White than when he
was African American, F(1, 39) ⫽ 3.66, p ⫽ .06, but errors in
response to unarmed targets did not seem to depend on ethnicity,
F(1, 39) ⫽ 1.68, p ⫽ .20.
Both the latency and error results attest to the role of target
ethnicity in disambiguating potentially threatening stimuli.
Clearly, the responses of participants to these stimuli depended at
some level on the ethnic category of the target, with potentially
hostile targets identified as such more quickly if they were African
American rather than White and benign targets identified as such
more quickly if they were White rather than African American.
Although these results are certainly consistent with our expecta-
tions, they are also somewhat surprising given the fact that the
target ethnicity appeared at exactly the same time as the object that
had to be identified as a gun or not. Certainly participants could
have performed perfectly on the task by attending only to the
object held in the target’s hand and by completely ignoring the
target’s ethnicity or any other individuating information.
To examine whether a target’s features, other than the object he
held, were attended to by participants, we examined their ability to
recognize the faces of the targets they had seen during the game.
A signal detection analysis revealed that sensitivity to old versus
new faces was not above chance level in these recognition data
(mean d⬘⫽0.15), t(39) ⫽ 1.15, p ⫽ .26. Separate analyses within
target ethnicity revealed that participants were unable to recognize
African American targets at a better than chance level (mean d⬘⫽
⫺0.08), t(39) ⫽⫺0.48, p ⫽ .63, although recognition sensitivity
for the White targets did exceed chance levels (mean d⬘⫽0.33),
t(39) ⫽ 2.26, p ⬍ .05. Our data suggest, then, that target ethnicity
affected participants’ judgments even while participants remained
largely incapable of recognizing the faces of the targets they had
seen.
Study 2
Our first study allowed participants a sufficient response win-
dow so that they made correct decisions in the case of nearly all
targets. That is, error rates were very low. As a result, the strongest
results from the first study were found with decision latencies on
correct responses, with faster decisions to armed African American
targets than to armed Whites, and faster decisions to unarmed
White targets than to unarmed African Americans. Although sig-
nificant, the interactive effects of Target Ethnicity and Object Type
on response errors were substantially weaker (and the relevant
simple comparisons were not significant).
In the second study, we sought to replicate the basic pattern of
results from the first study, but this time to make the task substan-
tially harder by shortening the amount of time during which
participants had to respond. Clearly, if the effects that we are
exploring are to be relevant to more real-life scenarios, such as
those encountered by police officers, then we would like to show
our effects on actual responses (and errors in responses) rather than
simply on the speed with which correct responses are made.
Additionally, to increase the importance of performance in the
task, we recruited participants exclusively for pay in this study and
we offered them incentives directly tied to the quality of their
performance, paying up to $20 for a study taking well less than an
hour.
Method
Participants and Design
Forty-four undergraduates (33 female, 11 male) participated in this
experiment in return for a minimum payment of $10, with the opportunity
to earn additional money (up to a total of $20) by scoring points in the
game. This incentive was intended to increase the personal significance of
the rewards and penalties. One male participant was Latino, and 1 female
was Asian. All other participants were White. We used the same 2 ⫻ 2
design, with Target Ethnicity (African American vs. White) and Object
Type (gun vs. no gun) as within-subject factors.
Materials and Procedure
The materials and procedure were identical to those of Study 1, with the
exception of the following modifications. First, we made clear to partici-
pants that they would be paid as a function of their performance. They were
told that they started with an initial sum of $14 to their credit. Each point
earned or lost (according to the same payoff matrix used in Study 1) was
worth 1 cent. It was made clear that if they performed perfectly across
all 80 trials, they would earn $20. If they lost points, they could lose up to
$4, but they were guaranteed a base pay of $10. Second, we adjusted the
game’s response window from 850 ms to 630 ms to force participants to
make decisions more quickly, with the goal of increasing error rates.
Although a 630-ms response window may provide ample time to process
simple stimuli such as faces or isolated objects, our images were fairly
complex, and the shortened window proved a challenge for our partici-
pants. A pretest indicated that the shortened response window had the
desired effect, increasing errors, but also dramatically increasing the pro-
portion of trials on which participants failed to respond in time. Because
the meaning of a timeout is ambiguous, a third change we made was to
discourage timeouts by increasing their associated penalty from 10 to 50
points (i.e., 50 cents) and stressing the importance of responding quickly in
the instructions. As participants’ point totals directly affected the amount
they were paid, this provided a considerable incentive. We also set an a
priori limit, such that any participant with more than 10 timeouts would be
excluded from the analysis. A final change we made was to the program
used to record participants’ data. In Study 1, for each trial, the program
only recorded the response, response latency, target ethnicity, and target
object (gun vs. no-gun), but the exact target and background for the trial
were not recorded. We modified this in Study 2 so that we could identify
particular stimuli that were associated with a greater number of errors.
1318
CORRELL, PARK, JUDD, AND WITTENBRINK
Results and Discussion
Before conducting the primary analysis of error rates, we elim-
inated 5 participants (all female) who exceeded our a priori thresh-
old of 10 timeouts (one eighth of all trials). Additionally, we
examined error rates for particular targets to determine if correct
responses were particularly difficult for some. In fact, there were
a number of targets that were outliers in the overall distribution,
inducing many more errors than the other targets. For instance, one
unarmed African American target was shot by more than 90% of
our participants. Additionally, one armed African American target
and four unarmed White targets resulted in errors for more than
one third of the participants. In each of these target images, some
detail seemed potentially misleading. For example, one target
had a stripe in his shorts that could be mistaken for a gun given
the position of his arm. We suspect the substantially shorter
time-out window was responsible for producing the unusually
high error rates for these six targets. To deal with these outliers,
we conducted all analyses twice, once with the full dataset
and once deleting the six outlying targets. The analyses that we
report are based on the partial dataset. However, with only one
exception, as noted below, the results were unaffected by their
inclusion/exclusion.
Participants’ error rates (number of errors divided by the total
number of valid trials) were subjected to a 2 ⫻ 2 ANOVA, with
Target Ethnicity (White vs. African American) and Object Type
(gun vs. no gun) as the independent variables. The relevant cell
means are given in Table 1. The analysis revealed a significant
effect for Object, such that the proportion of errors when a gun was
present (i.e., misses) was lower than the proportion of errors when
a gun was absent (i.e., false alarms), F(1, 38) ⫽ 6.42, p ⬍ .02. We
also found the predicted interaction between Ethnicity and Object,
F(1, 38) ⫽ 17.83, p ⬍ .0001. A test of the simple effects revealed
that, when the target was unarmed, participants mistakenly shot
him more often if he was African American than if he was White,
F(1, 38) ⫽ 6.53, p ⬍ .02, though this effect was not significant
when all targets were analyzed. When the target was armed,
however, participants mistakenly decided not to shoot more often
if he was White than if he was African American, F(1,
38) ⫽ 13.31, p ⬍ .001.
In addition to the analyses of the error rates, we also analyzed
the decision latencies for correct responses, as in Study 1. Not
surprisingly, given the considerably shorter response window in
this study, there were no effects in the latencies. It seems that
Study 1’s interaction in response speed was, in this study, pushed
over into error rates, due to the tightened response window.
As in Study 1, participants were unable to recognize presented
targets above chance level. An analysis of the mean sensitivity to
old versus new faces revealed a nonsignificant overall d⬘⫽⫺0.02,
t(38) ⫽⫺0.15, p ⫽ .88. Sensitivity was not above chance for
either the White targets (mean d⬘⫽0.12), t(38) ⫽ 0.74, p ⫽ .46,
or for the African American targets (mean d⬘⫽⫺0.16), t(38) ⫽
⫺0.97, p ⫽ .34.
To understand the error rate results in greater detail, further
analyses were conducted using the signal detection model (Green
& Swets, 1966/1974; MacMillan & Creelman, 1991). Applied to
the present context, the signal detection analysis assumes that
targets encountered, both those with a gun and those without a gun,
vary on a judgment-relevant dimension. For example, in the
present studies, the extent to which the targets appeared to be
threatening might have served as a critical dimension. On average,
targets with guns are more threatening than targets who possess
other objects (to the extent that they are discriminated at all), but
nevertheless, there is a distribution of targets within each set, and
these vary in how threatening they subjectively appear to be. Thus,
we have two distributions of targets, one of targets with guns and
one of targets without guns, and the signal detection model as-
sumes that these are both normal distributions with equal vari-
ances. To some extent, of course, these two distributions overlap
and the question of sensitivity is the question of the extent to which
this is true. That is, if participants are relatively sensitive or
accurate, shooting those targets who have guns and not shooting
those targets who don’t have guns, then the two distributions are
largely separated from each other.
Additionally, because participants make a choice between
shooting and not shooting a target on the basis of the subjective
sense of how threatening the target appears to be, they set a
decision threshold somewhere along the continuum that underlies
the two distributions. Above that threshold, they shoot the target;
below threshold they do not. Where that threshold is set is com-
monly referred to as the decision criterion.
From the two kinds of errors (false alarms: shooting an unarmed
target; misses: not shooting an armed target), one can derive
estimates of both sensitivity, commonly defined as d⬘, and decision
criterion, in this case defined as c. We estimated both of these
parameters for our participants, once for the White targets and
once for the African American targets. Unsurprisingly, given the
relatively low percentages of errors, participants showed consid-
erable accuracy (i.e., high levels of d⬘) for both the White and
African American targets (White M ⫽ 2.47 [SD ⫽ 0.87]; African
American M ⫽ 2.48 [SD ⫽ 0.85]). A test of differential sensitivity
between the two kinds of targets failed to reject the null hypoth-
esis, F(1, 38) ⫽ 0.01, p ⫽ .93. There were differences, however,
between the two kinds of targets in the response criterion (White
M ⫽ 0.03 [SD ⫽ 0.30]; African American M ⫽⫺0.24
[SD ⫽ 0.31]), such that a significantly lower decision criterion to
shoot the target was found for African American targets, F(1,
38) ⫽ 22.21, p ⬍ .0001. These results are depicted graphically in
Figure 2. In sum, from the perspective of the signal detection
model, the differences between responses to the African American
and White targets arose not from differences in the underlying
accuracy with which the two kinds of targets, those with a gun and
those without a gun, can be discriminated. Rather, in the case of
the African American targets, participants simply set a lower
threshold for the decision to shoot, being willing to shoot targets
who seemed less threatening.
3
3
The same pattern of signal detection results emerges both for Study 1
and for Study 3. For Study 1, sensitivity did not differ: African American
d⬘⫽3.30, White d⬘⫽3.28, F(1, 39) ⫽ 0.10, p ⫽ .75; but the decision
criterion did: African American c ⫽⫺0.17, White c ⫽⫺0.09, F(1, 39)
⫽ 10.07, p ⫽ .003. In Study 3, sensitivity did not differ: African American
d⬘⫽3.54, White d⬘⫽3.56, F(1, 44) ⫽ 0.12, p ⫽ .73; but the decision
criterion did: African American c ⫽⫺0.02, White c ⫽ 0.07, F(1, 44)
⫽ 6.96, p ⫽ .02.
1319
THE POLICE OFFICER’S DILEMMA
Study 3
Studies 1 and 2 provide evidence that the decision to shoot an
armed target is made more quickly and more accurately if that
target is African American than if he is White, whereas the
decision not to shoot is made more quickly and more accurately if
the target is White. This pattern of results is fundamentally con-
sistent with research suggesting that participants may use ethnicity
to interpret an ambiguously threatening target. When ambiguous
behavior is performed by an African American, it seems more
hostile, more mean, and more threatening than when it is per-
formed by a White person (Duncan, 1976; Sagar & Schofield,
1980). Participants also recognize a weapon more quickly and
more accurately after seeing an African American face, rather than
a White face (Payne, 2001). Here, we have shown that ethnicity
can also influence a behavioral judgment with serious conse-
quences for both target and shooter.
Simply documenting the existence of this bias does not clarify
the mechanism by which ethnicity influences the decision to shoot.
We suggested earlier that participants may use the stereotypic
association between the social category, African American, and
concepts like violence or danger as a schema to help interpret
ambiguous behavior on the part of any given African American
target. Through deductive inference, traits associated with the
category may be applied to the individual category member. It is
important to recognize that the proposed process does not require
a participant to dislike African Americans, or to hold any explicit
prejudice against them, nor does it require that the participant
endorse the stereotype; it simply requires that, at some level, the
participant associates the two concepts “African American” and
“violent.” Previous research is equivocal in its support of this
possibility, suggesting that bias in the interpretation of an ambig-
uous stimulus may depend on both stereotypic associations and on
prejudice. Sagar and Schofield (1980), for example, provide evi-
dence for a stereotype-driven effect. Recall that these researchers
found that both White and African American participants inter-
preted behavior as more threatening if it had been performed by an
African American target. Reasoning that bias among the African
American participants is not likely to reflect prejudice against
African Americans, they concluded that it reflects instead a com-
mon belief, or stereotype, that African Americans are more violent
than Whites. A culturally communicated stereotypic association
may influence interpretations even if the observer does not per-
sonally endorse the stereotype or hold a prejudiced attitude (De-
vine, 1989). Data presented by both Lepore and Brown (1997) and
Payne (2001), however, have shown that more prejudiced partic-
ipants show greater bias in their interpretations of ambiguous
stimuli (for Payne, 2001, this relationship was moderated by
motivation to control prejudice). Of course, the effect of prejudice
on perceptions may be indirect, operating chiefly through the
stronger negative stereotypic associations that accompany preju-
diced attitudes. The question is whether stereotypic associations
predict bias over and above prejudice. To be clear, we hypothesize
that although the magnitude of the bias evident in our videogame
may covary with participants’ prejudice against African Ameri-
cans, it is not a function of that prejudice, per se, but rather reflects
the deductive application of stereotypic associations (often asso-
Figure 2. Hypothetical normal distributions representing unarmed and armed targets for signal detection
analyses: White (top panel) and African Americans (bottom panel) targets.
1320
CORRELL, PARK, JUDD, AND WITTENBRINK
ciated with prejudice) between African Americans and violence.
Because participants can use traits associated with the group to
disambiguate a particular African American target, they may in-
appropriately perceive that target as threatening or hostile.
Study 3 represents a first attempt to test these predictions. After
playing the videogame, participants completed a questionnaire
designed to measure prejudice and two forms of the association
between African Americans and violence. The first measure of this
association assessed stereotypes that the participant personally
endorses or believes. We refer to this as the personal stereotype.
The second measure, called the cultural stereotype, is designed to
assess the participant’s awareness that a stereotype of African
Americans as violent is present in U.S. culture, generally. Though
we use the terms personal stereotype and cultural stereotype, this
distinction maps cleanly on to the endorsement/knowledge distinc-
tion suggested by Devine (Devine, 1989; Devine & Elliot, 1995),
who has shown that, although people often personally disavow
negative stereotypes about African Americans, they are well aware
that those stereotypes exist. Because this knowledge represents a
psychological link between the social category and the trait, acti-
vating the concept of the group may predispose a participant to
make use of the stereotypic trait in interpretations of an ambiguous
target—even if he or she does not personally endorse the stereo-
type. Both personal and cultural forms of the stereotypic associa-
tion, then, may influence interpretation of an ambiguous target.
Method
Participants and Design
Forty-eight undergraduates (26 female, 22 male) participated in this
experiment in return for either $10 or partial credit toward a class require-
ment. Two male participants were Latino, and 1 female was Asian. Another
female was African American and was excluded from our analyses. All
other participants were White. Two White females were also removed from
the dataset, one because the game’s shoot and don’t shoot labels were
reversed, and one because she was working as a research assistant on a
different study of African American stereotypes. The final sample in-
cluded 45 students. This study used the same 2 ⫻ 2 within-subject design,
with Target Ethnicity (African American vs. White) and Object Type (gun
vs. no gun) as repeated factors.
Materials
Videogame. In this study, we used the videogame parameters we had
used in Study 1. The response window was set at 850 ms, and we expected
effects primarily in the latency of correct responses, rather than in error
rates.
Questionnaire. Study 3 added a battery of individual difference mea-
sures. First, participants completed the Modern Racism Scale (MRS;
McConahay, Hardee, & Batts, 1981), the Discrimination (DIS) and Diver-
sity Scales (DIV) (both from Wittenbrink, Judd, & Park, 1997), all of
which are designed to measure prejudice against African Americans, as
well as the Motivation to Control Prejudiced Responding Scale (MCP;
Dunton & Fazio, 1997; Fazio, Jackson, Dunton, & Williams, 1995), which
assesses participants willingness to express any prejudice they may feel.
Items from these scales were intermixed (presented in a single, randomly
determined order) and responses were given on 5-point scales, ranging
from strongly disagree to strongly agree. The items were intermingled with
filler items from the Right-Wing Authoritarianism Scale (RWA, Alte-
meyer, 1988) and the Personal Need for Structure Scale (PNS; Thompson,
Naccarato, Parker, & Moskowitz, 2001), which are addressed below.
Second, to examine the degree to which participants endorsed a negative
stereotype of African Americans as aggressive and dangerous, we asked
them to estimate, on the basis of their personal beliefs, the percentage of
both African Americans and Whites who are dangerous, violent, and
aggressive (separate estimates were made for each trait by filling in a value
from 0% to 100%). Third, we included a measure of participants’ percep-
tions of the cultural stereotype that African Americans are aggressive and
dangerous. Participants were asked to again consider the three attributes
(dangerous, violent, and aggressive), giving prevalence estimates, not on
the basis of their own personal beliefs, but rather on the basis of their
perceptions of what most White Americans would estimate. These esti-
mates were made by marking a 130-mm line anchored with the adjective
(e.g., dangerous) on the right, and its negation (e.g., not dangerous)onthe
left.
In addition to these primary measures, the questionnaire included several
exploratory components. We included the RWA scale (Altemeyer, 1988),
which measures an individual’s predisposition to think of social relations in
terms of dominance and submission; the PNS scale (Thompson et al.,
2001), which measures differences in the desire for a simple structure; and
a five-item measure of contact with African Americans. These measures
were included partly as filler items designed to mask the questionnaire’s
focus on prejudice and stereotyping, but also because these constructs have
been shown to be related to prejudice or stereotyping in previous research
(Neuberg & Newson, 1993; Pettigrew & Tropp, 2000; Pratto, Sidanius,
Stallworth, & Malle, 1994). Responses to the contact items were made on
7-point scales. The first contact question asked participants to rate how
many African Americans they know, using a scale anchored with don’t
know any African Americans and know a lot of African Americans. The
second item asked for a rating of how well they know their African
American acquaintances on a scale from don’t know well to know very well.
The third item asked about the degree of contact with African Americans
in their neighborhood, when growing up. The fourth item asked about the
number of African American friends they had while growing up. And the
fifth item asked about the number of African Americans who had attended
their high school. The last three items used a scale ranging from none to
many.
Procedure
As before, a male experimenter greeted participants, in groups of 1 to 4,
and introduced the study as an investigation of perceptual vigilance. He
went on to note that, because the vigilance task did not require the entire
time period, participants would work on a separate questionnaire study
afterward. After learning about the rules of the game, participants moved
to computer terminals in private rooms and played the videogame. As each
participant completed the game, the experimenter moved him or her to a
table (still in the private room) and administered the short questionnaire,
from Studies 1 and 2, assessing basic reactions to the game. The experi-
menter subsequently announced that the videogame study was over and
provided another consent form, ostensibly for the separate questionnaire
study. After collecting the consent form, he handed the participant an
envelope containing the questionnaire. We made every effort to stress the
confidentiality of the responses on the questionnaire. The experimenter told
participants not to put any identifying information on the forms, not even
a code number, and to seal the packet in the envelope when they had
finished. He then left them alone to complete the questions. As in Studies 1
and 2, participants were fully debriefed. During this process, the experi-
menter probed for suspicion about the relationship between the game and
the subsequent questionnaire.
Results and Discussion
In the debriefing, 6 participants reported that they had noticed
that both the game and the questionnaire involved ethnicity, and
1321
THE POLICE OFFICER’S DILEMMA
that this awareness had prompted them to wonder if the two were
related. Two of the 6 reported strong suspicion. The following
results are based on the complete dataset, but exclusion of the 6
participants does not affect the analyses in either direction or
significance. To analyze the videogame data, we submitted the
log-transformed reaction times from correct trials to a 2 ⫻ 2
ANOVA, with Target Ethnicity and Object Type as the indepen-
dent variables (see Table 1 for means converted back to millisec-
onds). The targets that had proved problematic in Study 2 were
excluded from this analysis, though their inclusion does not sub-
stantially affect the results. Replicating the results from the first
study, we found both a pronounced effect for Object, such that
armed targets were responded to more quickly than unarmed
targets, F(1, 44) ⫽ 171.33, p ⬍ .0001, and an Ethnicity ⫻ Object
interaction, F(1, 44) ⫽ 22.44, p ⬍ .0001. Simple effects tests
revealed that, when the target was armed, participants, on average,
fired more quickly if he was African American than if he was
White, F(1, 44) ⫽ 4.15, p ⬍ .05. When presented with an unarmed
target, participants chose the “don’t shoot” alternative more
quickly if he was White than if he was African American, F(1,
44) ⫽ 22.72, p ⬍ .0001.
Mean scores on the error rates were largely consistent with those
from Study 1. The Ethnicity ⫻ Object interaction was significant,
F(1, 44) ⫽ 7.20, p ⫽ .01. Simple effects tests showed an ethnicity
effect only among targets without guns, F(1, 44) ⫽ 5.76, p ⫽ .02,
such that these were incorrectly shot more often if they were
African American. The simple effect for armed targets was not
significant, F(1, 44) ⫽ 2.31, p ⫽ .14. A test of the mean recog-
nition sensitivity for the presented targets was significant in this
study (mean d⬘⫽0.25), t(44) ⫽ 2.51, p ⫽ .016. As in Study 1,
however, sensitivity was above chance only for the White targets
(mean d’ ⫽ 0.62), t(44) ⫽ 4.71, p ⬍ .0001, and not for the African
American targets (mean d⬘⫽⫺0.15), t(44) ⫽⫺1.14, p ⫽ .26.
Having replicated the Ethnicity ⫻ Object interaction in the
response latency scores, we wanted to examine its correlates.
Accordingly, for each participant we computed a within-subject
contrast score, assessing the magnitude of the Ethnicity ⫻ Object
interaction for that particular participant. Higher scores on this
variable, which we refer to as Shooter Bias, indicate faster re-
sponses to unarmed White than to unarmed African American
targets, and to armed African American than armed White targets.
Table 2 reports the correlations between this Shooter Bias
measure and the various questionnaire measures. Table 2 also
reports the means, standard deviations, and internal consistency
(coefficient alpha) statistics for the various attitude scales in our
data. With the exception of contact and the personal and cultural
stereotype measures, all measures were collected on 5-point scales
with higher numbers indicating greater endorsement of the con-
struct. None of the explicit prejudice scales—MRS, DIS, and
DIV—show significant correlations with the Shooter Bias from the
videogame. That is, those who reported higher levels of prejudice
on these scales did not show a stronger ethnicity bias in the
videogame. Because these three measures are highly intercorre-
lated, we also combined them, averaging all items together. This
composite scale was similarly uncorrelated with Shooter Bias.
To compute the personal stereotype measure of African Amer-
icans as aggressive, we calculated the degree to which participants
rated African Americans as more violent than Whites, more dan-
gerous than Whites, and more aggressive than Whites. These three
difference scores were averaged together to form the personal
stereotype index. The measure reflects perceptions of the preva-
lence of the negative stereotypic attributes among African Amer-
icans relative to Whites. Because this measure is based on per-
centage estimates, it can potentially range from ⫺100 to 100. One
participant chose not to complete the relevant items, so all tests of
the personal stereotype are based on a sample of 44, rather than 45.
The same process was followed in computing the extent to which
participants believed there is a negative cultural stereotype of
African Americans as dangerous and aggressive. Because the raw
scores on the cultural stereotype items were made on 130-mm
lines, the index potentially ranges from ⫺130 to 130. As is clear in
Table 2, the measure of personal endorsement of the negative
stereotype of African Americans as aggressive and violent did not
correlate with the Shooter Bias. However, the perception of a
parallel negative cultural stereotype did correlate with the magni-
tude of the Shooter Bias in the videogame.
Of the exploratory measures (RWA, PNS, and contact), only
contact was related to Shooter Bias. Contact scores were calculated
Table 2
Correlations of Shooter Bias in Videogame With Questionnaire Measures (Study 3)
Variable MSD
␣
Shooter
Bias MRS DIS DIV
Prejudice
comp.
Personal
stereo.
Cultural
stereo. MCP Contact RWA
MRS 1.63 .66 .86 .15
DIS 2.09 .73 .87 .16 .80**
DIV 2.43 .64 .64 .05 .46** .59**
Prejudice comp. 2.09 .60 .91 .14 .85** .95** .78**
Personal stereo. 1.43 7.56 .54 .05 .38** .38** .38** .43**
Cultural stereo. 41.37 24.15 .88 .37** ⫺.06 ⫺.07 ⫺.21 ⫺.12 ⫺.06
MCP 3.23 .48 .72 .03 ⫺.35* ⫺.29* ⫺.27† ⫺.34* ⫺.31* .06
Contact 2.56 1.00 .72 .38** ⫺.18 ⫺.02 .11 ⫺.03 ⫺.07 .09 .15
RWA 2.16 .61 .72 ⫺.04 .26† .24 .37** .33* .02 ⫺.25† .25† .11
PNS 2.78 .55 .77 ⫺.15 .16 .00 ⫺.16 ⫺.01 ⫺.04 .15 .17 ⫺.06 .29*
Note. For all measures except personal stereotype, n ⫽ 45. All comparisons involving personal stereotype are based on n ⫽ 44. MRS ⫽ Modern Racism
Scale; DIS ⫽ Discrimination Scale; DIV ⫽ Diversity Scale; Prejudice comp. ⫽ prejudice composite; stereo. ⫽ stereotype; MCP ⫽ Motivation to Control
Prejudiced Responding Scale; RWA ⫽ Right-Wing Authoritarianism Scale; PNS ⫽ Personal Need for Structure Scale.
† p ⱕ .10. * p ⱕ .05. ** p ⱕ .01.
1322
CORRELL, PARK, JUDD, AND WITTENBRINK
by averaging participants’ responses to the five 7-point contact
items. This measure showed a significant and somewhat surprising
correlation with the bias: Participants who reported more contact
with African Americans exhibited a more pronounced Shooter
Bias in the videogame. We discuss this intriguing effect in the
General Discussion when we consider potential mechanisms that
may give rise to Shooter Bias.
We suggested that the Shooter Bias evident in this videogame
might be a consequence of participants using stereotypic associa-
tions about African Americans to help interpret ambiguous African
American targets. The data from Study 3 suggest that the magni-
tude of the bias was related to participants’ perceptions of the
cultural stereotype about African Americans. The bias was not,
however, related to either personally endorsed stereotypes or to
prejudice. This is somewhat surprising, because, to the extent that
people personally endorse the violent stereotype or hold prejudices
against African Americans, we might suppose the negative asso-
ciations to be stronger and more likely to influence their interpre-
tations of, and behavior toward, an ambiguous target.
There are well-documented social desirability concerns associ-
ated with expressing prejudice or negative stereotypic beliefs
about African Americans (Dunton & Fazio, 1997; McConahay et
al., 1981; Plant & Devine, 1998), so it may be that participants
simply refused to express their personal views. In his research,
Payne (2001) found no zero-order correlation between prejudice
(as measured by the MRS) and the automatic component in his
weapon identification task. He did find a moderated relationship
between the two variables, though, such that a positive correlation
emerged only among participants who were low in MCP. A similar
test in our data yielded no significant interaction between MRS
and MCP, F(1, 41) ⫽ 0.00, p ⬍ .95, or between personal stereo-
type and MCP, F(1, 40) ⫽ 0.95, p ⬍ .34, when predicting Shooter
Bias.
Unlike prejudice and personal stereotypes, our measure of cul-
tural stereotype should be generally free from social desirability
concerns. It involves participants’ estimates of the stereotype held
by American society. The fact that cultural stereotype correlates
with Shooter Bias suggests that awareness of the stereotype, itself,
even though a person may not believe that stereotype, can be
sufficient to produce bias. One might argue, however, that our
cultural stereotype measure was just another way of measuring
personal prejudice, in a manner that allowed participants to express
their own prejudices relatively free from normative constraints.
That is, by attributing prejudicial beliefs to others, participants
were now able to express more freely the prejudice that they
themselves felt.
The bivariate correlation between the cultural stereotype mea-
sure and our composite personal prejudice scale was ⫺.12 (p ⫽
.41), suggesting that this cultural stereotype measure is not a
simple proxy for personal prejudice levels. However, it might be
the case that the relationship between the cultural stereotype mea-
sure and personal prejudice depends on the participant’s level of
motivation to control prejudice, again following the theoretical
arguments of Fazio et al. (1995). To examine this possibility, we
regressed the cultural stereotype measure on our composite per-
sonal prejudice measure, MCP, and their interaction. The interac-
tion proved to be a significant predictor, F(1, 41) ⫽ 4.67, p ⬍ .05.
The direction of this interaction was as predicted: there was a more
positive relationship between personal prejudice levels and the
cultural stereotype measure among those who were lower in mo-
tivation to control prejudice.
We were interested in whether cultural stereotype would con-
tinue to predict Shooter Bias once we removed the extent to which
the cultural stereotype variable is a measure of personal prejudice,
particularly among those low in motivation to control prejudice.
Accordingly, we estimated a model with Shooter Bias as the
criterion, regressing it on the cultural stereotype measure while
controlling for our personal prejudice composite, MCP, and the
interaction between personal prejudice and MCP. In this model,
again, only the cultural stereotype measure related significantly to
bias in the videogame, F(1, 40) ⫽ 5.24, p ⬍ .03. Thus, even
removing personal prejudice levels from the cultural stereotype,
and controlling for the fact that personal prejudice levels were
more strongly related to the cultural stereotype among those low in
MCP, the cultural stereotype measure continued to predict bias in
our videogame.
4
This suggests that it is truly knowledge of the
cultural stereotype that is at work here, rather than simply an
indirect measure of personal prejudice. We consider this a sobering
prospect because it suggests that the bias may be endemic in
American society.
A number of studies have shown that cultural stereotypes can be
automatically activated even when a perceiver does not endorse
them (Banaji & Greenwald, 1995; Devine, 1989; Gilbert & Hixon,
1991; Macrae, Milne, & Bodenhausen, 1994). Cultural influences,
including television, movies, music, and newspapers provide a
constant barrage of information that often depicts African Amer-
icans as violent (Cosby, 1994; Gray, 1989), and those depictions
may shape our understanding of the world (Gerbner, Gross, Mor-
gan, & Signorielli, 1986). Popular culture, including Gangsta Rap
songs like the Notorious B.I.G.’s “Somebody’s Gotta Die,” Snoop
Dogg’s “Serial Killa,” or Dr. Dre’s “Murder Ink,” and movies like
Colors or Training Day may foster bias by enhancing detrimental
stereotypic associations, in spite of the fact that the audience
knows the characters and events are fictitious.
If cultural stereotypes associating African Americans with vio-
lence do, in fact, lead to Shooter Bias, any person exposed to
American culture should be liable to demonstrate the bias, regard-
less of his or her personal views about African Americans. Re-
search suggests that the very people who are targeted by cultural
stereotypes are influenced by the media representations they see
(Berry & Mitchell-Kernan, 1982; Stroman, 1986; SuberviVelez &
Necochea, 1990), know full well that the stereotypes exist (Steele
& Aronson, 1995), and even activate those stereotypes automati-
cally (Banaji & Greenwald, 1995). Sagar and Schofield (1980), as
noted above, found similar levels of bias among their African
American and White participants using their interpretation task. To
examine further the possibility that knowledge of the cultural
stereotype may, in and of itself, lead to Shooter Bias, we sought to
4
The attempt to control for the prejudice composite measure, MCP, and
their interaction only removes variance based on personal prejudice to the
extent that these scales reliably measure that variance. There is reason to
assume that these measures only partially assess prejudice, particularly for
participants high in MCP. Thus, although the analysis represents our best
attempt to examine the effects of cultural stereotypes over and above
prejudice in the current dataset, it is nonetheless imperfect. Our thanks to
Keith Payne for this insight.
1323
THE POLICE OFFICER’S DILEMMA
test for bias in a more diverse sample that included African
American participants.
Study 4
Method
Participants and Design
Fifty-two adults from bus stations, malls, and food courts in Denver,
Colorado, were recruited to participate in this study in return for $5. The
study followed the same 2 ⫻ 2 within-subject design used in Studies 1–3,
with Target Ethnicity (African American vs. White) and Object Type (gun
vs. no gun) as repeated factors, but in Study 4 we added a between-subject
factor, namely Participant Ethnicity (African American vs. White). The
final sample included 25 African Americans (6 females, 19 males) and 21
Whites (8 females, 13 males). One Asian and 4 Hispanic or Latino
participants, and 1 participant who did not indicate his ethnicity, were
excluded from the analyses, though the results do not differ if they are
included in the White sample.
Materials
In this study, we used the videogame parameters from Studies 1 and 3.
The response window was set at 850 ms and, again, we expected effects in
the latency of correct responses, rather than in error rates. Before beginning
this study, the targets identified as problematic in Study 2 were edited in
Photoshop to clarify the object in the picture.
Procedure
At each location, two male experimenters set up 2–3 laptop computers
equipped with the videogame program and earphones, to minimize dis-
tractions inherent in the nonlaboratory environment. Without a button box,
participants pressed the k key on the laptop keyboard to indicate shoot, and
the d key to indicate don’t shoot. While one experimenter circulated and
recruited participants, the other oversaw the experiment, giving instruc-
tions to each participant individually. After completing the videogame,
participants were paid and debriefed. In this study, we did not include
instructions to attend to target faces, nor did we test for recognition after
the game.
Results and Discussion
Before analyzing the videogame data, we reexamined the targets
that were problematic in Study 2. The targets no longer induced
unusually high numbers of errors, and they were therefore included
in the analyses reported below. The results reported do not change
in direction or magnitude if the targets are excluded. We submitted
the log-transformed reaction times from correct trials to a 2 ⫻
2 ⫻ 2 mixed-model ANOVA, with Participant Ethnicity as a
between-subject factor, and Target Ethnicity and Object Type as
within-subject factors (see Table 3 for means converted back to
milliseconds). Across all participants, we again found a pro-
nounced effect for Object, such that armed targets evoked re-
sponses more quickly than unarmed targets, F(1, 45) ⫽ 347.82,
p ⬍ .0001. The Target Ethnicity ⫻ Object interaction, or Shooter
Bias, was also significant, F(1, 45) ⫽ 14.75, p ⬍ .001. Crucially,
though, the magnitude of the bias did not depend on Participant
Ethnicity, F(1, 44) ⫽ 0.10, p ⫽ .75. Examining the African
American and White samples separately, we found that the Target
Ethnicity ⫻ Object interaction was significant for both, F(1,
24) ⫽ 6.55, p ⫽ .017 and F(1, 20) ⫽ 8.01, p ⫽ .01, respectively.
Simple effects tests again showed that, when the target was
armed, participants decided to shoot more quickly if he was
African American than if he was White, F(1, 45) ⫽ 7.62, p ⫽ .008.
When the target was unarmed, participants pressed the don’t shoot
button more quickly if he was White than if he was African
American, resulting in an identical test statistic, F(1, 45) ⫽ 7.62,
p ⫽ .008. Neither simple effect depended on Participant Ethnicity,
F(1, 44) ⫽ 0.07, p ⫽ .79, for the unarmed targets, and F(1,
44) ⫽ 0.42, p ⫽ .52, for the unarmed targets.
An analysis of the error rates revealed that the Target Ethnic-
ity ⫻ Object interaction was only marginal, F(1, 45) ⫽ 3.24, p ⫽
.08, and its magnitude did not depend on Participant Ethnicity,
F(1, 44) ⫽ 0.66, p ⫽ .42.
General Discussion
In four studies, we attempted to recreate the experience of a
police officer who, confronted with a potentially dangerous sus-
pect, must decide whether or not to shoot. Our goal was to examine
the influence of the suspect’s ethnicity on that decision. We used
a simplified videogame to present African American and White
male targets, each holding either a gun or a nonthreatening object.
Participants were instructed to shoot only armed targets. We rea-
soned that participants might use the stereotype, or schema, that
African Americans are violent to help disambiguate the target
Table 3
Means (and Standard Deviations) for Reaction Times and Error Rates as a Function of Target
Ethnicity, Object Type, and Participant Ethnicity (Study 4)
Participants
Reaction times Errors per 20 trials
White
targets
Afr. Am.
targets
White
targets
Afr. Am.
targets
White participants
Armed targets 590 (43) 578 (36) 1.38 (1.36) 0.76 (0.77)
Unarmed targets 652 (40) 665 (41) 1.19 (0.93) 1.29 (1.49)
Afr. Am. participants
Armed targets 578 (42) 567 (47) 2.00 (1.53) 1.52 (1.58)
Unarmed targets 645 (47) 659 (41) 1.64 (1.80) 1.44 (1.47)
Note. Afr. Am. ⫽ African American.
1324
CORRELL, PARK, JUDD, AND WITTENBRINK
stimuli, and would therefore respond with greater speed and ac-
curacy to stereotype-consistent targets (armed African Americans
and unarmed Whites) than to stereotype-inconsistent targets
(armed Whites and unarmed African Americans).
In Study 1, participants fired on an armed target more quickly
when he was African American than when he was White, and
decided not to shoot an unarmed target more quickly when he was
White than when he was African American. In Study 2, we
attempted to increase error rates by forcing participants to make
decisions very quickly. Participants in this study failed to shoot an
armed target more often when that target was White than when he
was African American. If the target was unarmed, participants
mistakenly shot him more often when he was African American
than when he was White. A signal detection analysis of these data
revealed that, although participants’ ability to distinguish between
armed and unarmed targets did not depend on target ethnicity,
participants set a lower decision criterion to shoot for African
American targets than for Whites. That is, if a target was African
American, participants generally required less certainty that he
was, in fact, holding a gun before they decided to shoot him. In
Study 3, we returned to an analysis of reaction times, replicating
the Ethnicity ⫻ Object Type interaction (Shooter Bias) obtained in
Study 1, and examining individual difference measures associated
with the magnitude of that effect. Shooter Bias was more pro-
nounced among participants who believed that there is a strong
stereotype in American culture characterizing African Americans
as aggressive, violent and dangerous; and among participants who
reported more contact with African Americans. Prejudice and
personal endorsement of the stereotype that African Americans are
violent failed to predict Shooter Bias in the simple correlations,
and their predictive power was no stronger among participants low
in motivation to control prejudice. The fact that Shooter Bias in
Study 3 was related to perceptions of the cultural stereotype, rather
than prejudice or personally endorsed stereotypes, suggests that
mere knowledge of the stereotype is enough to induce this bias. In
Study 4, we obtained additional support for this prediction. Testing
both White and African American participants, we found that the
two groups display equivalent levels of bias.
The results of these studies consistently support the hypothe-
sized effect of ethnicity on shoot/don’t shoot decisions. Both in
speed and accuracy, the decision to fire on an armed target was
facilitated when that target was African American, whereas the
decision not to shoot an unarmed target was facilitated when that
target was White. This Shooter Bias effect is consistent with the
results reported by Payne (2001). Payne primed participants with
African American and White faces, and asked them to identify
subsequent target objects as either hand tools or weapons. His
results suggest that responses to hand tools were faster (and, in a
second study, more accurate) when preceded by White, relative to
African American, primes, whereas responses to weapons were
faster (but no more accurate) when preceded by African American
primes. This priming effect maps nicely onto our results. The
consistency between our results and those obtained by Payne is
particularly striking given methodological differences between the
two paradigms. Four primary differences stand out. Payne used
small, decontextualized and relatively simple images of faces (the
center portion of the face) and objects, whereas our stimuli were
very complex, with target individuals appearing against realistic
backgrounds. Payne used a sequential priming task, whereas we
used simultaneous presentation of ethnicity and object. A conse-
quence of Payne’s priming task, which used a constant 200-ms
stimulus onset asynchrony, is that the appearance of a prime in his
task should have clearly indicated to participants that a target was
imminent. Our task, however, presented targets at random inter-
vals, with no prime, so that participants were never certain about
when they would appear. Finally, whereas Payne asked his partic-
ipants to identify a target object as a tool or a weapon, we asked
our participants to decide whether or not to shoot a target person.
Although both decisions depend on the presence of a weapon, the
psychological implications of the two tasks are quite different.
Payne’s task was framed as a categorization judgment, whereas
our task was characterized as a behavioral response. In spite of
these distinctions, both paradigms reveal a pronounced effect of
target ethnicity on reactions to weapons.
In line with Sagar and Schofield (1980), we have argued that
ethnicity influences the shoot/don’t shoot decision primarily be-
cause traits associated with African Americans, namely “violent”
or “dangerous,” can act as a schema to influence perceptions of an
ambiguously threatening target. The relationship between cultural
stereotype and Shooter Bias obtained in Study 3 provides support
for this hypothesis. The subsequent finding that African Americans
and Whites, alike, display this bias further buttresses the argument.
It is unlikely that participants in our African American sample held
strong prejudice against their own ethnic group (Judd, Park, Ryan,
Brauer, & Kraus, 1995), but as members of U.S. society, they are,
presumably, aware of the cultural stereotype that African Ameri-
cans are violent (Devine & Elliot, 1995; Steele & Aronson, 1995).
These associations, we suggest, may influence reactions to the
targets in our videogame. Though ambient cultural associations
may impact most members of U.S. society, it is certainly plausible
that personal endorsement of stereotypes, and perhaps prejudice,
will lead to even stronger negative associations with African
Americans, potentially magnifying bias. (Though the data in
Study 3, specifically the lack of a relationship between Shooter
Bias and personal stereotype, offer little support for this argument,
at present.)
It seems appropriate at this juncture to speculate on mechanisms
that may underlie Shooter Bias. Our basic findings indicate that a
target’s ethnicity, though technically irrelevant to the decision task
at hand, somehow interferes with participants’ ability to react
appropriately to the object in the target’s hand. This interference
seems roughly analogous to a Stroop effect, and research on this
extensively studied phenomenon may provide a useful perspective
from which to consider our results. The common Stroop experi-
ment presents participants with a word, and requires them to
identify the color of the ink in which that word is written (e.g.,
green ink). Performance on this simple task can be disrupted when
the word, itself, refers to a different color than the ink (e.g., RED
printed in green ink), relative to performance when the color of the
ink and the referent of the word are the same (e.g., GREEN printed
in green ink) or when the word does not refer to a color at all (e.g.,
EGGS printed in green ink). The Stroop paradigm, like our video-
game, simultaneously presents participants with information that is
relevant to the judgment at hand (ink color and object, respec-
tively) as well as information that is irrelevant (word name and
ethnicity, respectively). Participants need not process the irrelevant
information to perform the task, but in both cases, the presence of
incongruent information on the irrelevant dimension interferes
1325
THE POLICE OFFICER’S DILEMMA
with participants’ ability to process the relevant information. Re-
searchers have suggested that, because we so frequently read the
words that we see, reading occurs quickly. Ink naming, though, is
an unusual and relatively cumbersome task. If these two processes
occur in parallel, the quicker word reading may produce interfer-
ence by winning a kind of horse race, getting to the finish line and
influencing responses ahead of the slower ink-naming process,
which eventually provides the definitive answer (Cohen, Dunbar,
& McClelland, 1990; Posner & Snyder, 1975). Similarly, the
speedy categorization of people into ethnic categories, described
by Brewer (1988) and Fiske and Neuberg (1990), should quickly
activate stereotypes and interfere with the unfamiliar and less
automatic gun/no-gun judgment (see Figure 3). This analogy is not
perfect, of course. Although it may be natural to read the word
RED when it appears, the typical day-to-day response to an Afri-
can American does not involve gunfire. However, to the extent that
a person spontaneously associates an African American target with
violence, the ethnicity of the target should conflict with the judg-
ment that he is unarmed, and it may therefore inhibit the “don’t
shoot” response.
Cohen et al. (1990) characterized Stroop interference as an
interaction between two variables: attention to the irrelevant di-
mension and the strength of the association between the incongru-
ent information and the incorrect response. Both of these variables
can moderate Stroop effects independently (see Walley, McLeod,
& Khan, 1997; Walley, McLeod, & Weiden, 1994, for research on
attention; see Lu & Proctor, 2001, for research on the strength of
association). Though it is only speculation at present, we suggest
that the two significant predictors of Shooter Bias in Study 3,
cultural stereotype and contact, are important because they capture
these two components of Stroop interference. We have already
presented the argument that a cultural stereotype represents an
associative link between African Americans and traits related to
violence and danger. We further suggest that the role of contact in
predicting Shooter Bias may reflect, at least in part, the other
component of Stroop interference: attention to irrelevant ethnic
cues. People who have had extensive contact with African Amer-
icans may have, over the course of that experience, learned to
naturally parse the world in terms of ethnic categories. They may
be essentially schematic for ethnicity. Greater attention to ethnicity
combined with an association between African Americans and
violence should, from the Stroop perspective, magnify Shooter
Bias. In line with this prediction, Payne (Payne, Lambert, &
Jacoby, in press) has shown that asking participants to use ethnic
cues in their judgments (like a person engaged in racial profiling)
increases the magnitude of the automatic component in error
responses in his task, relative to control participants who receive
no special instructions. Of greater interest, asking participants to
avoid using ethnicity in their judgments also increases the magni-
tude of the automatic component. This suggests that attention to
the irrelevant ethnic cue may produce interference.
The Stroop conceptualization offers another, perhaps more
hopeful, prediction. If Shooter Bias is, in part, a function of the
automaticity with which ethnic cues are processed relative to the
automaticity of the object cues (i.e., ethnicity’s ability to win the
horse race against the relevant dimension), the bias should be
minimized by interventions that speed up the gun/no-gun decision.
As the relevant decision becomes more automatic, the effect of
the irrelevant dimension should weaken. Experimental research
(MacLeod, 1998; MacLeod & Dunbar, 1988) as well as computer
simulations (Cohen et al., 1990) have demonstrated that repeated
training on ink-naming tasks, which should render that process
more automatic, reduces Stroop interference. Similarly, training
participants to quickly and effortlessly distinguish guns from cell
phones may reduce Shooter Bias.
Though we have characterized Shooter Bias as a result of
distorted interpretations of an ambiguous target, there are several
stages at which this bias may actually be functioning. Before
shooting, a participant must (a) perceive the object, (b) interpret
the object as a gun with some degree of certainty, and (c) decide
to press the “shoot” button once a criterion of certainty has been
reached. Stereotypic schemata may theoretically affect any or all
of these processes, and it is difficult to disentangle them theoret-
ically, let alone empirically. Figure 3 depicts the three processing
stages and how faster, more automatic processing along the irrel-
evant dimension (as suggested by the Stroop research) might bias
each stage of relevant processing (the solid arrows). Throughout
this article, we have argued that bias impacts the second stage of
Figure 3. Faster, more automatic processing on the irrelevant ethnic dimension may bias participants’ (a)
perception of targets, (b) interpretation of targets, or (c) the criterion of certainty required for the “shoot”
response. Afr Amer ⫽ African American.
1326
CORRELL, PARK, JUDD, AND WITTENBRINK
this process, changing the interpretation of an ambiguous object. A
participant who catches a glimpse of some elongated shape in the
target’s hand may draw on stereotypic associations, interpreting
the shape as a gun if the target is African American but as a cell
phone if he is White. Participants may almost “see” different
objects.
One problem with this perspective is that we took pains to
ensure that the objects presented in our target images were clearly
identifiable. Even under time pressure, is it fair to characterize
these objects as ambiguous? Certainly, very few of our participants
actually misperceived the objects: our primary effects were in
reaction times, not errors, and errors were consistently quite low.
It is possible that the bias in reaction time represents the effects of
stereotypes on actual perception of the object, not on its interpre-
tation. von Hippel and his associates (von Hippel, Jonides, Hilton
& Narayan, 1993) showed that when a participant has a relevant
schema, he or she can infer the gist of a stimulus with very few
perceptual details. Without a schema, though, more detailed per-
ceptual encoding may be necessary. In the context of the current
research, the stereotype of African Americans may influence the
number features a participant must process to make the correct
object identification. An African American target provides a
schema relevant to guns, so participants who see just a few features
of a gun quickly identify it and decide to shoot. A White target,
perhaps, provides no useful schema, and participants must attend
to more features of the gun in his hand before they recognize it,
causing them to respond more slowly. von Hippel’s research
provides an elegant rationale for the differential speed required to
shoot an armed target, but we are less confident that perceptual
differences underlie reactions to the unarmed targets. Perceptual
processes can only account for the simple effect of ethnicity
among unarmed targets if we assume that White people stereo-
typically carry cell phones, wallets, coke cans, and cameras, and
that this stereotype reduces the number of perceptual cues neces-
sary to identify these objects relative to African Americans, where
no cell phone stereotype exists. Empirically, it should be possible
to test the viability of a perceptual encoding account of Shooter
Bias. If perceptual differences drive ethnic bias, then memory for
trivial details, such as the kind of gun or the color of the cell phone,
which should reflect the extent of perceptual encoding, should
differ as a function of target ethnicity.
Another, more macroscopic, alternative to our interpretation-
based account is that Shooter Bias may reflect changes in the
decision criterion that participants use. Bias would clearly emerge
if participants require one level of certainty that the object is a gun
when deciding to shoot African American targets, but have an-
other, more stringent, criterion for Whites. Even if the perception
and interpretation of an object do not differ as a function of target
ethnicity (e.g., the participant is 75% certain that the object is a gun
for both African American and White targets), a participant who
requires 60% certainty for African American targets, but 80%
certainty for Whites, will show Shooter Bias. Unfortunately, the
current studies do not allow us to discern between the interpreta-
tion and decision criterion explanations. Though the signal detec-
tion terms, sensitivity and criterion, might foster an expectation
that Study 2 should be able to resolve this question, that is not the
case. Study 2 suggested that sensitivity was equal for African
American and White targets, and that only the criterion differed.
The criterion may differ, though, either because the certainty
needed to make the shoot/don’t shoot decision differs with target
ethnicity (bias in the decision stage), or because a given object in
the hand of an African American target is simultaneously more
likely to be perceived as a gun and less likely to be perceived as a
non-gun, than the same object in a White target’s hand (bias in the
interpretation stage). The signal detection theory figure from
Study 2 assumes that the average armed White target and the
average armed African American target seem equally threatening,
that the two gun distributions fall at the same point on the x-axis.
As we have graphed it, the figure suggests that the criterion to
shoot shifts down for African American targets. However, it is also
possible that participants use the same criterion for White and
African American targets, but generally perceive African Ameri-
cans as more threatening. If this were the case, the criterion line in
the chart for African American targets would have the same
x-coordinate as the White criterion line, but the mean of the two
African American distributions (both armed and unarmed) would
seem to shift up on the dimension of perceived threat. Even using
signal detection theory, we have no way to statistically disentangle
these two possibilities in the current data.
Bias in the decision-making stage may be seen as consistent
with ideomotor effects. Bargh, Chen, and Burrows (1996, Study
3), for example, found that participants primed with African Amer-
ican faces exhibited more aggressive behavior in response to a
rude request from an experimenter. It is possible that the partici-
pants’ behavior was still driven by bias in their interpretation, that
those primed with African American actually perceived the exper-
imenter as more hostile (along the lines of Devine, 1989). But
Bargh and others (e.g., Chartrand & Bargh, 1996; Dijksterhuis,
Aarts, Bargh, & van Knippenberg, 2000; Stapel & Koomen, 2001)
have demonstrated direct behavioral priming effects in a number
of situations designed to preclude interpretation-based bias. It is
not unreasonable to suppose that participants in our studies were
cued by a target’s ethnicity to behave aggressively toward African
American targets, shooting them more often and more quickly than
Whites. Payne (2001), though, did not require a behavioral re-
sponse. Because his task required that participants classify objects
as guns or hand tools, rather than react violently, ideomotor effects
cannot account for his findings. In the absence of more defini-
tive evidence, and given the consistency between Payne’s results
and ours, parsimony argues for an interpretation-based explana-
tion of Shooter Bias, rather than a criterion-based or ideomotor
explanation.
These studies have demonstrated that the decision to shoot may
be influenced by a target person’s ethnicity. In four studies, par-
ticipants showed a bias to shoot African American targets more
rapidly and/or more frequently than White targets. The implica-
tions of this bias are clear and disturbing. Even more worrisome is
the suggestion that mere knowledge of the cultural stereotype,
which depicts African Americans as violent, may produce Shooter
Bias, and that even African Americans demonstrate the bias. We
understand that the demonstration of bias in an African American
sample is politically controversial given the nature of this task, and
we offer two considerations. First, the results of a single study are
not definitive. Our findings should be replicated by researchers in
other labs with different materials before generalizations are made.
Second, our goals as psychologists include understanding, predict-
ing, and controlling behavior. Ultimately, efforts to control (i.e.,
reduce or eliminate) any ethnic bias in the decision to shoot must
1327
THE POLICE OFFICER’S DILEMMA
be based on an accurate understanding of how target ethnicity
influences that decision, even if that understanding is politically or
personally distasteful.
Though these studies suggest that bias in the decision to shoot
may be widespread, it is not yet clear that Shooter Bias actually
exists among police officers. The studies we report use exclusively
lay samples, and there is no reason to assume that this effect will
generalize beyond this population. There is even a possibility,
suggested by literature on the Stroop effect, that police training
may actually reduce Shooter Bias by rendering the gun/no-gun
decision more automatic for officers. If this is the case, police
might show less bias than the average college sophomore. Exam-
ining these sorts of effects in a sample of police officers is of the
utmost importance.
The studies reported here suggest that Shooter Bias is present
among White college students (Studies 1–3) and among a com-
munity sample that consists of both Whites and African Americans
(Study 4). The effect is robust and clearly a cause for concern, no
matter the underlying cause. On the basis of our data, though, bias
does not seem to simply reflect prejudice toward African Ameri-
cans, and there is reason to believe the effect is present simply as
a function of stereotypic associations that exist in our culture. That
these associations can have such potentially profound conse-
quences for members of stigmatized groups is a finding worthy of
great concern. Since the death of Amadou Diallo, New York has
witnessed a number of similar, though less publicized, cases, and
Cincinnati, Ohio, has added Timothy Thomas’s name to the list of
unarmed African American men killed by police officers. Social
psychological theory and research may prove invaluable in the
effort to identify, understand and eventually control processes that
bias decisions to shoot (and possibly kill) a person, as a function
of his or her ethnicity.
References
Altemeyer, B. (1988). Enemies of freedom: Understanding right-wing
authoritarianism. San Francisco: Jossey-Bass.
Banaji, M. R., & Greenwald, A. G. (1995). Implicit gender stereotyping in
judgments of fame. Journal of Personality and Social Psychology, 68,
181–198.
Bargh, J. A., Chen, M., & Burrows, L. (1996). Automaticity of social
behavior: Direct effects of trait construct and stereotype activation on
action. Journal of Personality and Social Psychology, 71, 230–244.
Beckett, N. E., & Park, B. (1995). Use of category versus individuating
information: Making base-rates salient. Personality and Social Psychol-
ogy Bulletin, 21, 21–31.
Berry, G. L., & Mitchell-Kernan, C. (1982) Introduction. In G. L. Berry &
C. Mitchell-Kernan (Eds.), Television and the socialization of the mi-
nority child (pp. 1–11). New York: Academic Press.
Brewer, M. B. (1988). A dual process model of impression formation. In
T. K. Srull & R. S. Wyer (Eds.), Advances in social cognition (Vol. 1,
pp. 1–36). Hillsdale, NJ: Erlbaum.
Chartrand, T. L., & Bargh, J. A. (1996). Automatic activation of impres-
sion formation and memorization goals: Nonconscious goal priming
reproduces effects of explicit task instructions. Journal of Personality
and Social Psychology, 71, 464–478.
Cohen, J. D., Dunbar, K., & McClelland, J. L. (1990). On the control of
automatic processes: A parallel distributed processing account of the
Stroop effect. Psychological Review, 97, 332–361.
Cohen, J. D., MacWhinney, B., Flatt, M., & Provost, J. (1993). PsyScope:
An interactive graphic system for designing and controlling experiments
in the psychology laboratory using Macintosh computers. Behavior
Research Methods, Instruments & Computers, 25, 257–271.
Cosby, C. O. (1994). Television’s imageable influences: The self-
perceptions of young African-Americans. Lanham, MD: University
Press of America.
Darley, J. M., & Gross, P. G. (1983). A hypothesis-confirming bias in
labeling effects. Journal of Personality and Social Psychology, 44,
20–33.
Devine, P. G. (1989). Stereotypes and prejudice: Their automatic and
controlled components. Journal of Personality and Social Psychol-
ogy, 56, 5–18.
Devine, P. G., & Elliot, A. J. (1995). Are racial stereotypes really fading?
The Princeton trilogy revisited. Personality and Social Psychology Bul-
letin, 21, 1139–1150.
Dijksterhuis, A., Aarts, H., Bargh, J. A., & van Knippenberg, A. (2000). On
the relation between associative strength and automatic behavior. Jour-
nal of Experimental Social Psychology, 36, 531–544.
Duncan, B. L. (1976). Differential social perception and attribution of
intergroup violence: Testing the lower limits of stereotyping of blacks.
Journal of Personality and Social Psychology 34, 590–598.
Dunton, B. C., & Fazio, R. H. (1997). An individual difference measure of
motivation to control prejudiced reactions. Personality and Social Psy-
chology Bulletin, 23, 316⫺326.
Fazio, R. H., Jackson, J. R., Dunton, B. C., & Williams, C. J. (1995).
Variability in automatic activation as an unobtrusive measure of racial
attitudes: A bona fide pipeline? Journal of Personality and Social
Psychology, 69, 1013⫺1027.
Fiske, S. T., Lin, M., & Neuberg, S. L. (1999). The continuum model: Ten
years later. In S. Chaiken & Y. Trope (Eds.), Dual-process theories in
social psychology (pp. 231–254). New York: Guilford Press.
Fiske, S. T., & Neuberg, S. L. (1990). A continuum of impression forma-
tion, from category-based to individuating processes: Influences of in-
formation and motivation on attention and interpretation. In M. P. Zanna
(Ed.), Advances in experimental social psychology (Vol. 23, pp. 1–74).
New York: Academic Press.
Gerbner, G., Gross, L., Morgan, M., & Signorielli, N. (1986). Living with
television: The dynamics of the cultivation process. In J. Bryant & D.
Zillman (Eds.), Perspectives of media effects (pp. 14–40). Hillsdale, NJ:
Erlbaum.
Gilbert, D. T., & Hixon, J. G. (1991). The trouble of thinking: Activation
and application of stereotypic beliefs. Journal of Personality and Social
Psychology, 60, 509⫺517.
Gray, H. (1989) Television, Black Americans, and the American dream.
Critical Studies in Mass Communication, 6, 376–386.
Green, D. M., & Swets, J. A. (1966). Signal detection theory and psycho-
physics. New York: Wiley. (Reprinted 1974, Huntington, NY: Krieger).
Hamilton, D. L., & Sherman, J. W. (1994). Stereotypes. In R. S. Wyer &
T. K. Srull (Eds.), Handbook of social cognition (pp. 1–68). Hillsdale,
NJ: Erlbaum.
Hilton, J. L., & von Hippel, W. (1990). The role of consistency in the
judgment of stereotype-relevant behaviors. Personality and Social Psy-
chology Bulletin 16, 430–448.
Jacobs, J. E., & Eccles, J. S. (1992). The impact of mothers’ gender-role
stereotypic beliefs on mothers’ and children’s ability perceptions. Jour-
nal of Personality and Social Psychology 63, 932–944.
Jacoby, L. L. (1991). A process dissociation framework: Separating auto-
matic from intentional uses of memory. Journal of Memory and Lan-
guage, 30, 513⫺541.
Jacoby, L. L., Toth, J. P., & Yonelinas, A. P. (1993). Separating conscious
and unconscious influences of memory: Measuring recollection. Journal
of Experimental Psychology: General, 122, 139⫺154.
Judd, C. M., Park, B., Ryan, C. S., Brauer, M., & Kraus, S. (1995).
Stereotypes and ethnocentrism: Diverging interethnic perceptions of
1328
CORRELL, PARK, JUDD, AND WITTENBRINK
African American and White American youth. Journal of Personality
and Social Psychology, 69, 460–481.
Krueger, J., & Rothbart, M. (1988). The use of categorical and individu-
ating information in making inferences about personality. Journal of
Personality and Social Psychology, 55, 187–195.
Lepore, L., & Brown, R. (1997). Category and stereotype activation: Is
prejudice inevitable? Journal of Personality and Social Psychology, 72,
275–287.
Locksley, A., Borgida, E., Brekke, N., & Hepburn, C. (1980). Sex stereo-
types and social judgment. Journal of Personality and Social Psychol-
ogy, 39, 821–831.
Lu, C.-H., & Proctor, R. W. (2001). Influence of irrelevant information on
human performance: Effects of S–R association strength and relative
timing. Quarterly Journal of Experimental Psychology: Human Exper-
imental Psychology, 54(A), 95–136.
MacLeod, C. M. (1998). Training on integrated versus separated Stroop
tasks: The progression of interference and facilitation. Memory & Cog-
nition, 26, 201–211.
MacLeod, C. M., & Dunbar, K. (1988). Training and Stroop-like interfer-
ence: Evidence for a continuum of automaticity. Journal of Experimen-
tal Psychology: Learning, Memory, and Cognition, 14, 126–135.
MacMillan, N. A., & Creelman, C. D. (1991). Detection theory: A users
guide. Cambridge, England: Cambridge University Press.
Macrae, C. N., Milne, A. B., & Bodenhausen, G. V. (1994). Stereotypes as
energy-saving devices: A peek inside the cognitive toolbox. Journal of
Personality and Social Psychology, 66, 37–47.
McConahay, J. B., Hardee, B. B., & Batts, V. (1981). Has racism declined
in America? It depends on who is asking and what is asked. Journal of
Conflict Resolution, 25, 563⫺579.
Neuberg, S. L., & Newson, J. T. (1993). Personal need for structure:
Individual differences in the desire for simpler structure. Journal of
Personality and Social Psychology, 65, 113–131.
Payne, B. K. (2001). Prejudice and perception: The role of automatic and
controlled processes in misperceiving a weapon. Journal of Personality
and Social Psychology, 81, 181⫺192.
Payne, B. K., Lambert, A. J., & Jacoby, L. L. (in press). Best laid plans:
Effects of goals on accessibility bias and cognitive control in race-based
misperceptions of weapons. Journal of Experimental Social Psychology.
Pettigrew, T. F., & Tropp, L. R. (2000). Does intergroup contact reduce
prejudice: Recent meta-analytic findings. In S. Oskamp (Ed.), Reducing
prejudice and discrimination (pp. 93–114). Mahwah, NJ: Erlbaum.
Plant, E. A., & Devine, P. G. (1998). Internal and external motivation to
respond without prejudice. Journal of Personality and Social Psychol-
ogy, 75, 811–832.
Posner, M. I., & Snyder, C. R. R. (1975). Attention and cognitive control.
In R. L. Solo (Ed.), Information processing and cognition: The Loyola
symposium (pp. 55–85). Hillsdale, NJ: Erlbaum.
Pratto, F., Sidanius, J., Stallworth, L. M., & Malle, B. F. (1994). Social
dominance orientation: A personality variable predicting social and
political attitudes. Journal of Personality and Social Psychology, 67,
741–763.
Rothbart, M., & Birrell, P. (1977). Attitude and perception of faces.
Journal of Research in Personality 11, 209–215.
Sagar, H. A., & Schofield, J. W. (1980). Racial and behavioral cues in
black and white children’s perceptions of ambiguously aggressive acts.
Journal of Personality and Social Psychology 39, 590–598.
Stapel, D. A., & Koomen, W. (2001). The impact of interpretation versus
comparison mindsets on knowledge accessibility effects. Journal of
Experimental Social Psychology, 37, 134–149.
Steele, C. M., & Aronson, J. (1995). Stereotype threat and the intellectual
test performance of African Americans. Journal of Personality and
Social Psychology, 69, 797–811.
Stroman, C. A. (1986). Television viewing and self-concept among Black
children. Journal of Broadcasting and Electronic Media, 30, 87–93.
SuberviVelez, F. A., & Necochea, J. (1990). Television viewing and
self-concept among Hispanic American children—A pilot study.
Howard Journal of Communications, 2, 315–329.
Thompson, M. M., Naccarato, M. E., Parker, K. C. H., & Moskowitz, G. B.
(2001). The personal need for structure and personal fear of invalidity
measures: Historical perspectives, current applications, and future direc-
tions. In G. B. Moskowitz (Ed.), Cognitive social psychology: The
Princeton symposium on the legacy and future of social cognition (pp.
19–39). Mahwah, NJ: Erlbaum.
von Hippel, W., Jonides, J., Hilton, J. L., & Narayan, S. (1993). Inhibitory
effect of schematic processing on perceptual encoding. Journal of Per-
sonality and Social Psychology, 64, 921–935.
Walley, R. E., McLeod, B. E., & Khan, M. (1997). Tests of a translational
model of Stroop interference: Translation or attention? Canadian Jour-
nal of Experimental Psychology, 51, 10–19.
Walley, R. E., McLeod, B. E., & Weiden, T. D. (1994). Increased attention
to the irrelevant dimension increases interference in a spatial Stroop task.
Canadian Journal of Experimental Psychology, 48, 467–492.
Wittenbrink, B., Judd, C. M., & Park, B. (1997). Evidence for racial
prejudice at the implicit level and its relationship with questionnaire
measures. Journal of Personality and Social Psychology, 72, 262⫺274.
Received February 11, 2002
Revision received July 5, 2002
Accepted July 5, 2002 䡲
1329
THE POLICE OFFICER’S DILEMMA
