ArticlePDF Available

Across the Thin Blue Line: Police Officers and Racial Bias in the Decision to Shoot

Authors:

Abstract and Figures

Police officers were compared with community members in terms of the speed and accuracy with which they made simulated decisions to shoot (or not shoot) Black and White targets. Both samples exhibited robust racial bias in response speed. Officers outperformed community members on a number of measures, including overall speed and accuracy. Moreover, although community respondents set the decision criterion lower for Black targets than for White targets (indicating bias), police officers did not. The authors suggest that training may not affect the speed with which stereotype-incongruent targets are processed but that it does affect the ultimate decision (particularly the placement of the decision criterion). Findings from a study in which a college sample received training support this conclusion.
Content may be subject to copyright.
INTERPERSONAL RELATIONS AND GROUP PROCESSES
Across the Thin Blue Line: Police Officers and Racial Bias in the Decision
to Shoot
Joshua Correll
University of Chicago
Bernadette Park and Charles M. Judd
University of Colorado at Boulder
Bernd Wittenbrink
University of Chicago
Melody S. Sadler
University of Colorado at Boulder
Tracie Keesee
University of Denver
Police officers were compared with community members in terms of the speed and accuracy with which
they made simulated decisions to shoot (or not shoot) Black and White targets. Both samples exhibited
robust racial bias in response speed. Officers outperformed community members on a number of
measures, including overall speed and accuracy. Moreover, although community respondents set the
decision criterion lower for Black targets than for White targets (indicating bias), police officers did not.
The authors suggest that training may not affect the speed with which stereotype-incongruent targets are
processed but that it does affect the ultimate decision (particularly the placement of the decision
criterion). Findings from a study in which a college sample received training support this conclusion.
Keywords: police, race, bias, weapon, training
Inspired in part by high-profile police shootings of unarmed
Black men, a flurry of social psychological research has attempted
to assess the influence of a suspect’s race on the use of force,
specifically in terms of the decision to shoot (Correll, Park, Judd,
& Wittenbrink, 2002; Greenwald, Oakes, & Hoffman, 2003;
Payne, 2001). Although social psychologists have only recently
addressed this question, the impact of suspect ethnicity on police
shootings has long been the focus of researchers in other fields of
study, particularly sociology, political science, and law enforce-
ment. Investigators have consistently found evidence that police
use greater force, including lethal force, with minority suspects
than with White suspects (e.g., Inn, Wheeler, & Sparling, 1977;
Smith, 2004; see Geller, 1982, for a review). Data from the
Department of Justice (2001), itself, indicate that Black suspects
are approximately five times more likely than White suspects, per
capita, to die at the hands of a police officer.
One of the most detrimental consequences of police shootings is
the upheaval they can provoke. Shootings of a minority suspect
may engender a sense of mistrust and victimization among com-
munity members and give rise to conflict between the community
and police. Weitzer and Tuch (2004) present evidence that mem-
bers of ethnic minorities often feel that they are mistreated by the
police, even after statistically controlling for factors like personal
and vicarious experiences with the law, exposure to the media, and
neighborhood disadvantage (see also Sunshine & Tyler, 2003).
The implication is that the police are racist and that officers use
excessive force with minority suspects. In response, Black people
may engage in more belligerent behavior, including “talking back”
to police officers, and—in a vicious cycle—this belligerence may
prompt more severe use of force by police (Reisig, McCluskey,
Joshua Correll, Department of Psychology, University of Chicago;
Bernadette Park, Charles M. Judd, and Melody S. Sadler, Department of
Psychology, University of Colorado at Boulder; Bernd Wittenbrink, Grad-
uate School of Business, University of Chicago; Tracie Keesee, University
of Denver.
Primary support for this work was provided by a grant from the Russell
Sage Foundation. Support for this work also came from National Institute
of Mental Health Grant F31-MH069017 to Joshua Correll and National
Institute of Mental Health Grant R01-45049 to Bernadette Park and
Charles M. Judd.
In the interest of disclosure, we note that Tracie Keesee also serves as
a commander in the Denver Police Department. We thank Chief Gerald
Whitman, the Denver Police Department, Calibre Press, the Denver De-
partment of Motor Vehicles, and (especially) the many officers of the
Denver Police Department and police departments around the country for
their assistance, patience, and participation. We also thank Alinne Barrera,
Heather Coulter, and David M. Deffenbacher for their invaluable assis-
tance with this research and Myron Rothbart for his many helpful com-
ments.
Correspondence concerning this article should be addressed to Joshua
Correll, Department of Psychology, University of Chicago, Chicago, IL
60637. E-mail: jcorrell@uchicago.edu
Journal of Personality and Social Psychology, 2007, Vol. 92, No. 6, 1006 –1023
Copyright 2007 by the American Psychological Association 0022-3514/07/$12.00 DOI: 10.1037/0022-3514.92.6.1006
1006
Mastrofski, & Terrill, 2004). It is equally important to note that, as
a consequence of this tension, officers who see their job as pro-
tecting the community may feel, and to some extent may actually
be, thwarted in their efforts to perform their duty.
Officer-involved shootings, then, can have severe consequences,
not just for the officers and suspects involved, but for the com-
munity at large as well. It is of paramount importance to under-
stand and explain why minority suspects are disproportionately
likely to be shot. The sociological literature offers a number of
explanations. Some research suggests that bias in police shootings
stems, at least in part, from the officers’ role as protectors of the
privileged (predominantly White) classes over the less fortunate
(predominantly minority) members of society (Sorenson, Mar-
quart, & Brock, 1993). Others argue that the racial discrepancy in
officer-involved shootings stems from differential minority in-
volvement in criminal activity (Department of Justice, 2001; Inn et
al., 1977) or from the fact that minorities are disproportionately
likely to live and work in low-income, high-crime communities
(Terrill & Reisig, 2003).
A primary strength of the sociological approach is that it exam-
ines police use of force directly and in its true context. These
researchers study real locations and real officers, and their depen-
dent variable is the number of suspects who are actually shot. They
thus maintain the richness and complexity of the real world when
analyzing relationships between officer-involved shootings and
variables like race or community disadvantage. At the same time,
the preexisting correlations among these variables confound ef-
forts to assess their independent effects. For example, the relation-
ship between the proportion of Black citizens in a community and
perceptions of disorder (Sampson & Raudenbush, 2004) is inex-
tricably tied to, and cannot be fully separated from, racial discrep-
ancies in officer-involved shootings (Terrill & Reisig, 2003). For
this reason, a social psychological analysis of the problem with
experimental methods is useful not to replace but rather to sup-
plement research of a more naturalistic sort.
Over the past several years, social psychological researchers
have examined the effect of race on shoot/don’t-shoot decisions
using videogame-like simulations. In one paradigm, participants
view a series of images (background scenes and people) and are
instructed to respond to armed targets with a shoot response, and
to unarmed targets with a don’t-shoot response as quickly and as
accurately as possible (Correll et al., 2002; Correll, Park, Judd, &
Wittenbrink, 2007; Correll, Urland, & Ito, 2006). The results of
some 20 studies with this task, with a variety of parameters and
manipulations, consistently show racial bias in both the speed and
accuracy with which such decisions can be made. Participants are
faster and more accurate when shooting an armed Black man
rather than an armed White man, and faster and more accurate
when responding “don’t shoot” to an unarmed White man rather
than an unarmed Black man. The bulk of this research has been
conducted with college students, but the effect has been replicated
with community samples of both White and Black participants,
and conceptually similar effects have been obtained by a number
of other labs (Amodio et al., 2004; Greenwald et al., 2003; Payne,
2001; Payne, Lambert, & Jacoby, 2002; Plant, Peruche, & Butz,
2005). These findings, along with reports from sociological and
related literatures, clearly indicate that race can play an important
role in decisions about the danger or threat posed by a particular
person. But experimental data rarely speak directly to police be-
havior.
In our literature review, we discovered only two papers that
examine officers in experimental studies of racial bias. Eberhardt,
Goff, Purdie, and Davies (2004) found that priming the concept of
crime served to orient attention to Black (more than White) faces.
This pattern held for officers and civilians alike. Plant and Peruche
(2005) examined training effects among officers on a task where
images of White and Black men appeared with a gun or nongun
object superimposed on the face. They found that officers showed
racial bias in their errors during the first phase of the study (i.e.,
officers were more likely to mistakenly shoot Black targets who
appeared with nongun objects, and to not shoot White targets who
appeared with a gun in the first 80 trials of the task), but that bias
fell to nonsignificant levels in the second phase (i.e., the last 80
trials of the task). These studies suggest that officers, like under-
graduates, show racial biases in the processing of crime-related
stimuli.
But there is reason to believe that police will differ from citizens
in shoot/don’t-shoot decisions. Most notably, officers receive ex-
tensive experience with firearms during their academy training
(before they are sworn in) and throughout their careers. For ex-
ample, the Denver Police Department requires that new recruits
spend 72 hr in practical weapons training, and officers must
recertify on a quarterly basis. At the firing range, officers and
recruits make shoot/don’t-shoot decisions for target silhouettes
that appear suddenly, either armed or unarmed; in Firearms Train-
ing System simulators (Firearms Training Systems, Inc., Atlanta,
GA), they respond to an interactive video simulation of a poten-
tially hostile suspect; and in simulated searches, they confront live
actors armed with weapons that fire painful but nonlethal ammu-
nition (e.g., paintballs, Simunition, or Air Soft pellets).
An extensive body of research shows that training improves
performance on tasks in which a peripheral cue interferes with a
participant’s response to a central or task-relevant cue. Through
training, participants learn to ignore the irrelevant information and
respond primarily on the basis of the central feature of the stimulus
(e.g., MacLeod, 1998; MacLeod & Dunbar, 1988; Plant & Pe-
ruche, 2005). For example, in a Stroop (1935) task, participants
classify the color in which a word is printed (e.g., red). Color is
thus the central cue. This task becomes more difficult if the word
(a peripheral cue) refers to a different color (e.g., the word “blue”
printed in red). Initially, participants have difficulty with this task,
responding slowly and inaccurately when the central and periph-
eral cues conflict. But with training, judgment improves. Re-
sponses occur more quickly and require less effort and less cog-
nitive control. As a result, experts demonstrate reduced
interference in both latencies and errors. Neuroimaging studies
have even documented the shifting patterns of brain activity that
correspond to the development of automatic task performance
(Bush et al., 1998; Jansma, Ramsey, Slagter, & Kahn, 2001; for a
review, see Kelly & Garavan, 2004). During initial performance on
interference tasks, participants recruit brain regions related to
conflict detection and response control (e.g., the anterior cingulate
and medial prefrontal cortexes). With extensive practice, however,
activation in these regions decreases, presumably because an au-
tomatic task requires less executive supervision.
But automatization may not characterize all learning on inter-
ference tasks. In some cases, training actually promotes controlled
1007
THIN BLUE LINE
processing. For example, when participants are continuously chal-
lenged by variable task requirements or increasing demands, prac-
tice can lead to more extensive recruitment of prefrontal brain
regions (Olesen, Westerberg, & Klingberg, 2004; Weissman,
Woldorff, Hazlett, & Mangun, 2002). Of particular relevance to
shoot/don’t-shoot decisions, this control involves the medial and
middle frontal gyri—areas related to the detection and resolution
of conflicting information and to the maintenance of goal-relevant
representations. In some cases, then, training leads participants to
work harder, in cognitive terms, as they learn to marshal the
attention and control necessary for optimal performance.
When will training promote automaticity in a judgment task, and
when will it promote control? A probable moderator is task com-
plexity (Birnboim, 2003; Green & Bavelier, 2003). On tasks with
simple stimuli (e.g., the words presented in a Stroop task), practice
allows participants to streamline the judgment process, performing
it easily and automatically. Only when the task is difficult (e.g.,
involving visually complex stimuli or ever-changing task require-
ments) does practice seem to promote control. As Birnboim (2003)
wrote, “automatic processing relies on a reduction of stimulus
information to its perceptual and motor features” (p. 29). When
complexity renders this kind of reduction impossible, controlled
processing may be required to “extract more meaningful informa-
tion” (p. 29). Consistent with this argument, Green and Bavelier
(2003) have shown that practice on a visually complex video game
(i.e., Medal of Honor; Electronic Arts, Redwood City, CA) im-
proves performance on attention-demanding tasks, but practice on
a visually simple video game (i.e., Tetris; Electronorgtechnica,
Moscow, Russia) does not.
Task complexity has tremendous relevance for the officer en-
gaged in a potentially hostile encounter. Faced with a range of
irrelevant and confusing factors (e.g., darkness, noise, movement,
bystanders), the officer must determine whether or not a small and
relatively inconspicuous weapon is present. On a reduced scale,
our paradigm attempts to simulate this visual and cognitive chal-
lenge. The task employs a variety of complex and realistic back-
grounds (e.g., parking lots, train stations). By varying backgrounds
and suspect poses (e.g., standing, crouching), as well as the timing
of stimulus onset, we prevent participants from knowing when or
where an object will appear. When the object does appear, it
accounts for roughly 0.2% of the visual field. To respond correctly,
participants must engage in a careful, controlled search for a small
cue amid a complex stimulus array. In contrast to the visually
simple tasks typically employed in research on training, training on
this relatively complex task may not foster automaticity in the
shoot/don’t-shoot decision. In our task—as in a police encounter—
even highly trained experts may need to fully engage executive
control processes to identify the object and execute the appropriate
response (Weissman et al., 2002).
If experts are better able than novices to engage control pro-
cesses, it stands to reason that police officers, whose training and
on-the-job experiences routinely force them to identify weapons in
complex environments, should make fewer errors in our shoot/
don’t-shoot task and should show reduced racial bias in those
errors (i.e., their expertise should minimize stereotypic errors).
This training-based reduction in bias, which we might call a
“police as experts” pattern, serves as our primary hypothesis (H
1
).
But control may not entirely eliminate race-based processing. The
necessity of a slow, effortful, and controlled search for the object
leaves open the possibility that even experts will inadvertently
process racial information. Research suggests that racial cues are
often perceived quickly, whether or not the participant intends to
do so (Cunningham et al., 2004; Ito & Urland, 2003), and accord-
ingly, a slow visual search for the object should glean racial
information. By activating stereotypes, these cues may interfere
with the speed of the decision-making process. By virtue of en-
hanced control, experts may rarely, if ever, shoot an unarmed
Black individual; but because even experts must search (slowly)
for the object, they are likely to perceive the target’s skin color and
facial features, triggering relevant stereotypes. Again, experts may
effectively override this interference and make an unbiased re-
sponse (“don’t shoot”), but because the weapon judgment is not
automatic, the controlled decision to stereotype incongruent targets
may still take more time. This leads us to predict a dissociation,
such that a target’s race may affect the speed of the expert’s
decisions, even though it has no impact on their accuracy.
To examine this possibility, the present research extends past
work in two critical ways. First, we examine bias in both response
times and errors. In past research (e.g., Correll et al., 2002; Payne,
2001), results from these two measures mirrored one another and
were characterized as more or less interchangeable. But the mea-
sures may capture partially distinct aspects of the decision process.
Latency—the time necessary for a participant to respond correctly
to a given target—should depend on the difficulty of processing
the stimulus. The fact that stereotype-incongruent targets (unarmed
Black targets and armed White targets) generally produce longer
latencies suggests that participants have greater difficulty arriving
at a correct decision for these stimuli. Processing difficulty may
also influence error rates, but errors also reflect the participant’s
ultimate decision about which response to make. Particularly from
an officer’s perspective, the distinction between a slow-but-
accurate response (e.g., hesitating and then deciding not to fire)
and an incorrect response (e.g., shooting an unarmed suspect)
assumes great importance.
This research also advances our understanding by comparing
police officers with samples of laypeople drawn from the commu-
nities those officers serve. Community samples provide a crucial
baseline against which we can compare the police. As we have
already discussed, one of the most damaging consequences of
officer-involved shootings in which a minority suspect is killed is
the implication that police inappropriately use race when making
the decision to fire. However, given the prevalence of bias in the
decision to shoot (which has been documented in all types of
people, from White college students to Black community mem-
bers), how can we interpret the magnitude of any bias we might
observe among the police? Inhabitants of the community served by
a given police department provide a critical comparison. As mem-
bers of a common culture, these individuals experience many of
the same influences, whether very global (e.g., national broadcast
media) or very local (e.g., racial and ethnic composition of the
neighborhood, local levels of poverty and crime) in nature. To
fully characterize the presence of any bias among police, it is
therefore critical to examine bias in the communities they serve.
No such comparison is available in existing research. Although we
have elaborated the hypothesis that police will demonstrate less
bias than the community, particularly with respect to their error
rates (H
1
), we note that the comparison between police and com
-
munity presents two other possibilities.
1008
CORRELL ET AL.
Of course, it is also possible that officers will show more
pronounced bias than community members (H
2
) or that police and
civilians will show relatively similar patterns of bias (H
0
). In line
with the former hypothesis, Teahan (1975a, 1975b) presented
evidence that police departments acculturate White officers into
more prejudicial views during their first years on the job. Simi-
larly, the Christopher Commission’s investigation into the Los
Angeles Police Department’s 1991 beating of Rodney King re-
ported that officers who adopted anti-Black attitudes were more
likely to be promoted within the department (Christopher, 1998).
This ostensible culture of bias may find expression in police
officers’ relatively high social dominance orientation (Sidanius &
Pratto, 1999), reflecting support for the group-based (and race-
based) hierarchical structure of society (see Sorenson et al., 1993,
for similar conclusions on the basis of police use of force). Given
these findings, we might reasonably expect a “police as profilers”
pattern, with officers relying heavily on racial information when
making their decisions to shoot.
Finally, police officers and community members may show
equivalent levels of racial bias in decisions to shoot. Inasmuch as
police and community members are subject to the same general
cognitive heuristics (Hamilton & Trolier, 1986) and sociocultural
influences (Devine & Elliot, 1995), the two groups may demon-
strate similar patterns of behavior in the video game simulation.
This prediction would yield a pattern we might call “police as
citizens.”
Our primary hypothesis derives from the possibility that practice
enables police officers to more effectively exert control over their
behavioral choices (relative to untrained civilians). That is, H
1
suggests that officers may more extensively engage in controlled
processing operations during the course of the shoot/don’t-shoot
task. Because of this difference in processing, we predict a diver-
gence between measures of bias that are based on errors and
measures that are based on reaction times. By contrast, H
2
and H
0
offer no clear reason to predict differences between officers and
civilians in terms of cognitive processing, and (accordingly) they
offer no reason to expect a divergence between error-rate and
reaction-time measures.
Study 1
Method
Overview. Three samples of participants completed a 100-trial
video game simulation in which armed and unarmed White and
Black men appeared in a variety of background images. Partici-
pants were instructed that any armed target posed an imminent
threat and should be shot as quickly as possible. Unarmed targets
posed no threat and should be flagged accordingly by pushing the
don’t-shoot button, again as quickly as possible. The speed and
accuracy with which these decisions were made served as our
primary dependent variables, and performance was compared
across three samples: officers from the Denver Police Department,
civilians drawn from the communities those officers served, and a
group of officers from across the country attending a 2-day police
training seminar.
Participants. For the purposes of law enforcement, the city of
Denver is divided into six districts. With the help of the command
staff, officers were recruited for this study from four of these
districts during roll call. Participation was completely voluntary,
and officers were assured that there would be no way to identify
individual performance on the task and that the command staff
would not be informed of who did and did not participate. Officers
were required to complete the simulation during off-duty hours.
Our goal was to recruit primarily patrol officers, and, in this effort,
we were successful: 84% of the sample listed patrol as their job
category. Investigative officers accounted for 9% of the sample,
administrative officers for 2% of the sample, with the remaining
5% of the officers from a mixture of other job categories. A total
of 124 officers participated in the study (9 female, 114 male, 1
missing gender; 85 White, 16 Black, 19 Latina/o, 3 other, 1
missing ethnicity; mean age 37.9 years). Each received $50.
To obtain a companion civilian sample, we enlisted the Depart-
ment of Motor Vehicles (DMV) office in each of the four districts,
recruiting community members to perform the simulation on or
around the same days as the police officers. Several of the DMVs
were in areas with a high concentration of Spanish-speaking citi-
zens. For these areas, a bilingual research assistant recruited and
instructed the participants.
1
A total of 135 civilians participated in
the study. Eight participants were dropped from the analyses: 2
because of a computer malfunction and 6 because they had fewer
than five correct trials for at least one of the four cells of the
simulation design. Thus, the reported results for this sample are
based on 127 civilians (51 female, 73 male, 3 missing gender; 39
White, 16 Black, 63 Latina/o, 9 other; mean age 35.5 years).
Each received $20.
To collect the national police sample, we attended a training
seminar for officers. This was one of several seminars that officers
voluntarily attend to obtain additional training in some particular
area of law enforcement. The seminars are specifically geared for
patrol officers, rather than administrative personnel. The sample of
officers obtained for this study came from 14 different states, and
only 7% worked in some administrative capacity. The remaining
job categories included patrol officers (58%), investigative officers
(14%), traffic officers (7%), SWAT team members (3%), and a
sprinkling of other categories (11%). Although this clearly is not a
random national sample of officers, it offers a greater diversity of
background than the Denver sample. An announcement regarding
the study was made during the seminar, and officers were invited
to participate on one of two evenings after the conclusion of the
seminar for that day. A total of 113 officers participated in the
study (12 female, 100 male, 1 missing gender; 72 White, 10 Black,
15 Latina/o, 13 other, 3 missing ethnicity; mean age 38.4 years).
Each received $50.
Video game simulation. Fifty men (25 Black, 25 White) were
photographed in five poses holding one of a variety of objects,
including four guns (a large black 9 mm, a small black revolver, a
large silver revolver, and a small silver automatic) and four non-
guns (a large black wallet, a small black cell phone, a large silver
Coke can, and a small silver cell phone). For each individual, we
1
Many thanks to Alinne Barrera who tirelessly and happily made time
in her busy schedule to accompany us on these sojourns at the Denver
DMVs.
1009
THIN BLUE LINE
selected two images, one with a gun and one with an innocuous
object, resulting in 100 distinct images (25 of each type: armed
White, armed Black, unarmed White, and unarmed Black), which
served as the principal stimuli, or targets, in the game. Forty of
these images were drawn from previous work (see Correll et al.,
2002, for example stimuli). The others were added in an effort to
diversify the sample of targets. Using Photoshop, we embedded
targets in 20 otherwise unpopulated background scenes, including
images of the countryside, city parks, facades of apartment build-
ings, and so on. Each target was randomly assigned to a particular
background, with the restriction that each type of target should be
represented with equal frequency in each background.
Design. The video game, developed in PsyScope (Cohen,
MacWhinney, Flatt, & Provost, 1993), followed a 2 2 within-
subjects design, with Target Race (Black vs. White) and Object
Type (gun vs. nongun) as repeated factors (see Correll et al.,
2002). On any given trial of the game, a random number (0 –3) of
preliminary backgrounds appeared in slideshow fashion. These
scenes were drawn from the set of 20 original unpopulated back-
ground images. Each remained on the screen for a random period
of time (500 ms– 800 ms). Subsequently, a final background ap-
peared (e.g., an apartment building), again for a random duration.
This background was replaced by an image of a target person
embedded in that background (e.g., an armed White man standing
in front of the building). From the player’s perspective, the target
simply seemed to appear in the scene. The player was instructed to
respond as quickly as possible whenever a target appeared, press-
ing a button labeled shoot if the target was armed and pressing a
button labeled don’t shoot if the target was unarmed. The game
awarded points on the basis of performance. Correctly pressing
don’t shoot in response to an unarmed target earned 5 points, but
shooting earned a penalty of 20 points; pressing shoot in response
to an armed target earned 10 points, but pressing don’t shoot
earned a penalty of 40 points (the implication being that the hostile
target shot the player). Failure to respond to a target within 850 ms
of target onset resulted in a penalty of 10 points. Feedback, both
visual and auditory, and point totals were presented at the conclu-
sion of every trial. The game consisted of a 16-trial practice block
and a 100-trial test block.
Procedure. Officers in the Denver sample were recruited
roughly 1 week prior to the study. Volunteers selected a time and
date to participate. At the scheduled time, each officer was seated
at a small cubicle in a test room equipped with a laptop computer,
button box, and headphones. They completed the simulation and
questionnaire packet. The measures included in the questionnaire
packet are summarized in Table 1. Community members were
approached at one of the various DMV locations, and those who
agreed to participate completed the simulation using the same
equipment as the officers. Community members completed the
same questionnaire as the officers (excluding items specific to
policing). For the national sample of officers, an announcement
was made the first day of the training seminar inviting officers to
participate in the study. Officers completed the simulation and
questionnaire packet on one of two evenings in a room in the hotel
where the conference was held. The equipment was identical to
that used for the Denver officers and civilians. Upon completion,
all participants were debriefed and thanked.
Results
Signal-detection analyses. We began by examining the accu-
racy of responses as a function of trial type and sample. Overall,
participants responded incorrectly on 4.7% of the trials and timed
out on another 4.8% of the trials. Correct and incorrect responses
(i.e., excluding timeouts) were used to conduct a signal-detection
analysis. Applied to the shooter simulation, signal detection theory
(SDT) assumes that armed and unarmed targets vary along some
dimension relevant to the decision at hand (e.g., the threat they
pose). SDT yields estimates of participants’ ability to discriminate
between the two types of target (i.e., sensitivity to the presence of
a weapon, a statistic called d) and the point on that decision-
relevant dimension at which they decide a stimulus is threatening
enough to warrant shooting (i.e., the psychological criterion for the
decision to shoot, a statistic called c). With SDT it is possible to
test whether the race of a target affects discriminability and,
separately, whether target race affects the decision to shoot. Cor-
rell et al. (2002, Study 2) observed no race differences in d but
found that c was lower for Black targets than for White targets.
That is, participants were equally able to differentiate between
armed and unarmed targets regardless of target race, but they used
a more lenient threshold—indicating a greater willingness to
shoot—when the target was Black rather than White.
We calculated d, or the ability to accurately discriminate armed
from unarmed targets, once for the White targets and once for the
Black targets. We also calculated c, or the criterion, assessing the
threshold for making a shoot response separately for Black and
White targets.
2
The SDT estimates were submitted to separate 3
(Sample: national officers vs. Denver officers vs. Denver commu-
nity) 2 (Target Race: Black vs. White) mixed-model analyses of
variance (ANOVAs).
Placement of the criterion for the decision to shoot (c) at zero
indicates no greater tendency to make a shoot response than a
don’t-shoot response. Deviations from zero in a positive direction
indicate a bias favoring the don’t-shoot response, and deviations in
a negative direction indicate a bias to shoot. On average (i.e., for
both officers and civilians and both Black and White targets),
participants demonstrated a bias in favor of the shoot response,
F(1, 361) 4.68, p .03, but the extent to which this was true
depended on sample, F(2, 361) 4.97, p .008. Pairwise
comparisons indicated that the community set significantly lower
criteria than either officer sample, both Fs(1, 361) 4.12, ps
.05. (All pairwise comparisons were tested with the error term
from the full sample.) Indeed, although the mean threshold was
significantly below zero for the community sample, F(1, 126)
10.05, p .002, it did not differ from zero for either of the two
officer samples, both Fs 1, and the two officer samples did not
differ from each other, F(1, 361) 1.22, p .27.
It is important to note that the main effect of target race in the
placement of the decision criterion was significant, F(1, 361)
5.17, p .03, such that c was lower when responding to Black
2
c ⫽⫺0.5 (zFA zH); d⬘⫽zH zFA, where FA is the proportion
of false alarms (relative to correct rejections) and H represents the propor-
tion of hits (relative to misses). The z operator is the translation of these
proportions to z-scores. Both FA and H were assigned a minimum value of
1/2n (where n the total number of no-gun and gun trials, respectively)
and a maximum of 1 (1/2n), to eliminate infinite z-scores.
1010
CORRELL ET AL.
rather than White targets (see the top half of Figure 1 and the
means in Table 2). This discrepancy constitutes bias. Although the
omnibus test of the interaction between target race and sample was
not significant, F(2, 361) 1.87, p .16, pairwise comparisons
indicated a larger target race difference for the Denver community
compared with the national officer sample, F(1, 361) 3.67, p
.056, other Fs 1.49, ps .22. Racial bias in c was significant
among the Denver community sample, F(1, 126) 5.71, p .02,
marginally significant among the Denver officer sample, F(1,
123) 3.28, p .07, and nonsignificant among the national
officer sample, F 1.
It is informative to examine sample differences in c separately
for the White and Black targets. As is clear from Figure 2,
placement of the criterion for the White targets changed very little
across the three samples, and in fact neither the omnibus test of
sample differences, F 1, nor any of the pairwise comparisons, all
Fs(1, 361) 1.54, ps .22, revealed a significant difference on
this measure. Moreover, the criterion for White targets was not
significantly different from zero for any of the three samples, all
Fs 1.49, ps .23. That is, neither officers nor community
members showed a tendency to favor one response over the other
when the target was White. In contrast, the threshold for Black
targets changed substantially and significantly across the three
samples, F(2, 361) 7.03, p .001. The criterion was set lowest
by the Denver community sample, whose mean c was both sig-
nificantly lower than zero, F(1, 126) 15.05, p .001, and
significantly lower than either of the two officer samples, both
Fs(1, 361) 4.42, ps .04. The Denver officers’ mean c value
was also significantly below zero, F(1, 123) 4.04, p .05, and
approached a significant difference when compared to the national
officer sample, F(1, 361) 2.79, p .10. The national officers’
criteria for Black targets did not differ from zero, F 1.
3
3
In each of the three samples, we tested for moderation of bias in
latencies, d, and c by participant ethnicity and gender. Because of the
relatively small number of non-White participants, particularly in the
officer samples, these analyses compared all non-White participants with
White participants. Bias was not moderated by participant ethnicity for any
of the samples ( ps ranged from .76 to .11). The only effect of gender was
moderation of bias in response times for the community sample. Bias was
significantly greater for male than for female community members, F(1,
122) 5.66, p .02, but it is important to note that bias was significant
within each sample, F(1, 50) 11.16, p .002 for female participants,
and F(1, 72) 61.00, p .001 for male participants.
Table 1
Demographic and Psychological Variables Included in Questionnaire Packet and Their
Correlations With Bias in Latencies in Study 1
Variable
Correlation with bias in latencies
National
officers
Denver
officers
Denver
community
Population of city in which officer serves .31
***
——
Population of county in which officer serves .31
***
——
Violent crime in community served .20
**
.09 .05
% African Americans in community served .21
**
.01 .01
% all ethnic minority groups in community served .22
**
.02 .05
Classroom firearms training .01
Firing-range firearms training .03
Video firearms training .02
Live firearms training .02
Total years on the force .09 .17
*
Gender (1 female; 1 male) .13 .13 .21
**
Ethnicity (1 non-White; 1 White) .09 .14 .08
Education .02 .10 .12
Self-rated liberalism (1)–conservatism (13) .04 .21
**
.06
Thermometer rating (warmth toward White people–warmth
toward Black people)
.00 .02 .03
Thermometer rating (warmth toward White people–warmth
toward members of all ethnic minority groups)
.00 .00 .04
Personal stereotype of Black people as dangerous, violent,
and aggressive
.02 .01 .20
**
Cultural stereotype of Black people as dangerous, violent,
and aggressive
.02 .05 .09
Contact with Black people .05 .02 .11
Internal motivation to control prejudice .04 .05 .11
External motivation to control prejudice .16 .12 .20
**
Discrimination scale .13 .04 .08
Note. City and county population have no variance for the Denver police and community samples, and hence
no correlation can be computed. Firearms training data were not collected for the Denver officers, nor for the
community. Ns vary slightly across entries because of missing observations. In the national sample, ns vary from
97–113; in the Denver police sample, they vary from 118 –123; and in the Denver community sample, they vary
from 120 –127. Dashes indicate that data were not collected.
*
p .10.
**
p .05.
***
p .01.
1011
THIN BLUE LINE
With respect to the analysis of d, these data largely replicated
previous work, such that target race did not affect participants’
ability to discriminate armed from unarmed targets. In other
words, the main effect of target race was not significant in the d
analysis, F(1, 361) 1.12, p .29 (see the bottom panel of
Figure 1 and Table 2 for all means and standard deviations).
However, the main effect of sample was significant, F(2, 361)
11.69, p .001. Pairwise comparisons indicated that both officer
samples showed higher discriminability than the community, in-
dicating a greater ability to differentiate armed from unarmed
targets, both Fs(1, 361) 11.01, ps .001. The two officer
samples did not differ from one another, F(1, 361) 1.55, p
.21. The interaction between sample and race of target was mar-
ginally significant, F(2, 361) 2.49, p .085. Pairwise compar-
isons indicated a significant difference only between the Denver
officers and the Denver community, F(1, 361) 4.63, p .04.
The officers showed slightly (but nonsignificant, F 1) greater
sensitivity to weapon detection for Black rather than White targets.
Among the community, d was significantly higher for White
targets than for Black targets, F(1, 126) 4.84, p .03.
Reaction-time analyses. We next examined reaction times. For
each participant, latencies from correct responses were log trans-
formed and averaged separately for each type of target (see Table
2 for means and standard deviations). Averages were analyzed as
a function of sample (national officers vs. Denver officers vs.
Denver community), target race (Black vs. White), and object type
(gun vs. nongun) using a 3 2 2 mixed-model ANOVA with
repeated measures on the latter two factors. Consistent with past
research, we obtained a main effect of object type, F(1, 361)
2,171.27, p .001, such that participants shot armed targets more
Figure 1. Decision criterion placement (c) and sensitivity (d) for Black and White targets as a function of
sample (Study 1).
1012
CORRELL ET AL.
quickly than they decided to not shoot unarmed targets. The target
race main effect was also significant, F(1, 361) 4.90, p .03,
such that, overall, responses were very slightly faster to White
(M 605 ms) than to Black targets (M 608 ms). Moreover, the
sample main effect was significant, F(2, 361) 5.36, p .006.
Contrasts among the samples indicated that both officer groups
responded significantly faster overall than the civilian group, Fs(1,
361) 3.68, ps .056, and the two officer samples did not differ
from each other, F 1.86, p .18 (M
national officers
597 ms,
M
Denver officers
604 ms, M
Denver community
613 ms).
It is important to note that we obtained the Target Race
Object Type interaction, F(1, 361) 239.37, p .001. This effect
reflects racial bias in decisions to shoot (see Figure 2). Notably, the
interaction did not depend on sample, F(2, 361) 1.74, p .18.
Bias was significant for all three samples: for the national sample
of officers, F(1, 112) 68.89, p .001, for the Denver officers,
F(1, 123) 117.29, p .001, and for the Denver community
sample, F(1, 126) 65.29, p .001. Pairwise comparisons
among the samples revealed no differences in the magnitude of
bias between the community sample and either of the officers
samples, Fs 1.17, ps .28, and marginally greater bias among
the Denver than national officer sample, F(1, 361) 3.44, p
.065.
We further examined the simple effects of target race for each
type of object. Again, consistent with previous findings, partici-
pants shot armed targets more quickly when they were Black,
rather than White, F(1, 361) 74.04, p .001, and they indicated
don’t shoot in response to unarmed targets more quickly when they
Figure 2. Response times to Black and White armed and unarmed targets as a function of sample (Study 1).
Table 2
Response Time, Sensitivity, and Decision Criterion Means and Standard Deviations for Studies 1 and 2
Variable
Sample
National officers Denver officers Denver community
Black White Black White Black White
M SD M SD M SD M SD M SD M SD
Study 1
Response time
Gun
ms 552
a
560
b
560
a
572
b
568
a
578
b
log transformed mean 6.31 0.07 6.33 0.07 6.33 0.08 6.35 0.07 6.34 0.09 6.36 0.09
No gun
ms 648
a
635
a
653
a
637
b
663
a
649
b
log transformed mean 6.47 0.06 6.45 0.06 6.48 0.05 6.46 0.06 6.50 0.07 6.48 0.07
Sensitivity (d) 3.42 0.59 3.43 0.50 3.54 0.52 3.50 0.59 3.12 0.78
a
3.24 0.76
b
Threshold (c) .014 0.19 .009 0.21 .032
*
0.18 .006 0.21 .087
*
0.25
a
.026 0.24
b
Study 2
Sensitivity (d) 2.39 0.80 2.17 0.73 1.39 0.84 1.47 1.03
Threshold (c) .072 0.30 .122
*
0.31 .302
*
0.33
a
.185
*
0.39
b
Note. Different row subscripts within each sample indicate a significant Black–White difference at p .05. For the decision criterion, means significantly
different from zero at p .05 are indicated with an asterisk.
1013
THIN BLUE LINE
were White, rather than Black, F(1, 361) 177.27, p .001.
These simple effects did not depend on sample, both Fs 1, ps
.39, and both of the simple target race effects within object type
were significant for each of the three samples, all Fs 15.00, all
ps .001. Pairwise comparisons for the simple effects among the
three samples revealed no significant differences, all Fs 1.85, all
ps .17.
Summarizing the results thus far, we see that officers and
community members differ in the criteria they employ for Black
targets. Community members set a lower, more lenient criterion
for shooting Black targets than either of the two officer samples.
At the same time, officers and community members show similar
levels of bias in terms of the speed with which they can correctly
respond to targets. We have suggested that, by virtue of their
training or expertise, officers may exert control over their behav-
ior, possibly overriding the influence of racial stereotypes. Con-
sistent with the possibility of enhanced control, officers also
showed greater sensitivity than did community members to the
presence of a weapon, regardless of target race. However, we do
not suggest that officers are completely immune to stereotypes. To
the extent that a Black target evokes the concept of danger,
behavioral control should be difficult. Reactions to targets that
violate stereotypic expectancies (i.e., unarmed Black targets and
armed White targets) should be slower than reactions to
stereotype-congruent targets. If officers’ response latencies reflect
the magnitude of racial stereotypes, we might expect greater la-
tency bias for officers exposed to stronger environmental associ-
ations between Black people and crime. Community characteris-
tics, such as crime rates and the proportion of minority residents,
might predict the magnitude of bias among officers in the laten-
cies. It is important to note, however, that if officers can exert
control over their behavior, stereotypic associations should not
produce greater bias in the SDT criteria they employ. We used the
questionnaire data to explore this issue. Because there is very little
variance among the Denver officers on these community charac-
teristics (that is, the population of the city and county served by all
officers in Denver is the same, and racial makeup across commu-
nities varies minimally), the national officer sample affords a more
effective test of these possibilities.
Correlational analyses. We computed indices of racial bias
on the basis of both response times ([RT
unarmed Black target
RT
unarmed White target
] [RT
armed White target
RT
armed Black target
]),
and criteria (c
White
c
Black
). Higher numbers indicate greater
racial bias. We also calculated the effect of target race on discrim-
inability (d
White
d
Black
), with higher numbers representing
greater sensitivity for White targets than for Black targets.
We then conducted exploratory analyses of the relationships
between each of these indices and the questionnaire measures
obtained. We report correlations for all three samples (see Table
1), but again, because the national sample offers greater variability
in terms of community demographics, we focus our discussion on
that sample. Bias in the response times was positively related to the
size (i.e., population) of the city, r(97) .31, p .003, and
county, r(103) .31, p .002, in which the officer served
(population variables were log transformed to normalize their
distributions). This effect suggests that officers in larger commu-
nities showed greater bias in the latency measure. In addition,
officers’ reports of the level of violent crime in their districts
predicted bias in response latencies, r(111) .20, p .03, such
that increases in violent crime were associated with greater racial
bias. Officers rated violent crime levels with respect to FBI sta-
tistics for the average national violent crime rate (500 offenses per
100,000 persons) on a 5-point scale with the endpoints anchored at
much lower than average and much higher than average. Officers
were also asked to estimate the ethnic makeup of the communities
in which they served. The estimated percentage of both African
Americans, r(108) .21, p .03, and ethnic minorities more
generally, r(108) .22, p .03, living in the community posi-
tively predicted racial bias in the latencies. None of the remaining
correlations for the national sample of officers was significant.
Officers serving in districts characterized by a large population,
a high rate of violent crime, and a greater concentration of Black
people and other minorities showed increased bias in their reaction
times. We tentatively suggest that these environments may rein-
force cultural stereotypes, linking Black people to the concept of
violence. The fact that officers from these urban, violent areas
show more pronounced bias in their latencies suggests that stereo-
typic associations may indeed influence police on some level. But
if training enables officers to effectively control their behavior,
such stereotypes should not influence their final shoot/don’t-shoot
decisions. It is interesting that these community demographics,
which systematically predicted latency bias, were completely un-
related to bias in the SDT estimates of decision criteria (rs ranged
from .14 to .13, smallest p value .19). In other words,
environmental variables that increased bias in officers’ latencies
had no effect on the degree of bias in their ultimate decisions.
We also asked participants (community members and officers
alike) to complete several measures of stereotyping and prejudice.
In the past, we have obtained relationships between bias in re-
sponse times and an individual’s awareness of cultural stereotypes
about Black people (Correll et al., 2002, Study 3; Correll, et al.,
2007). In the present study, measures of personally endorsed
stereotypes did correlate with latency bias for the community
members, r(123) .21, p .05, but cultural stereotypes did not.
Moreover, in the officers’ data, neither of these relationships
emerged. It is possible that this difference reflects something
special about the relationship between stereotypes and bias among
officers, but we suspect that the reason has more to do with the
officers’ concerns about going “on the record” with regard to their
attitudes about race. Despite our assurances of anonymity, several
officers were unwilling to complete the measures, and others told
us, rather bluntly, that they would not respond honestly to these
sensitive questions. We therefore view these items with suspicion,
at least for the officer samples.
The effects of target race on the SDT estimates were not related
to any of the demographic variables. As null effects, these results
are difficult to interpret. They may reflect a true lack of corre-
spondence between demographics and performance, but they may
also stem from the relatively low error rates in this task (which
likely reduce the reliability of the SDT estimates).
4
Although
Black–White differences were unrelated to the questionnaire mea-
sures, we did find that the average values of both d and c
(independent of target race) were correlated with training in sim-
ulated building searches. In this type of training, officers interact
with actors, some of whom attack the trainee using weapons
4
We thank an anonymous reviewer for this insight.
1014
CORRELL ET AL.
equipped with nonlethal ammunition. Police with more extensive
training in these encounters were better able to discriminate be-
tween armed and unarmed targets, regardless of the race of the
target, r(113) .20, p .04, and they tended to set a higher
overall criterion in the task, r(113) .17, p .07, reflecting
greater reluctance to shoot. It is interesting that no other type of
training (e.g., classroom training, firing range, interactive video
training) predicted performance in the game. Future researchers
should attempt to replicate these correlations, but the results ten-
tatively suggest that live, interactive training provides officers with
a chance to hone their skills in a manner that improves perfor-
mance.
Discussion
Analyses of the behavioral data showed that the officers’ overall
performance on the video game simulation exceeded that of the
civilians in several ways. First, their response times were faster.
On average, officers were simply quicker to make correct shoot/
don’t-shoot decisions than were civilians. Second, they were better
able to differentiate armed targets from unarmed targets. On av-
erage (i.e., across White and Black targets), d was greater for the
officers than for the community sample. Third, whereas the crite-
rion c for the community was significantly below zero (reflecting
a tendency to favor the “shoot” response), officers adopted a more
balanced criterion. In fact, not only was the officers’ criterion
significantly higher than the community’s, but the officers’ thresh-
old also did not differ significantly from zero. This placement
suggests equal use of the “shoot” and “don’t shoot” responses.
In terms of bias, the SDT results suggest that officers may show
less bias than civilians in their final decisions. Among the com-
munity sample, these data revealed a clear tendency to set a lower
(i.e., more lenient or “trigger-happy”) criterion for Black, rather
than White, targets. But this bias was weaker, or even nonexistent,
for the officers. The reduction in bias seemed to reflect the fact
that, compared with the community members, officers set a higher,
more stringent threshold for the decision to shoot Black targets.
Placement of the criterion for White targets varied minimally
across the three samples.
The response-time data show clear evidence of racial bias for all
samples in this study, the 237 police officers and the community
members alike. Like college students in previous studies, these
individuals seemed to have greater difficulty (indexed by longer
latencies) responding to stereotype-incongruent targets (unarmed
Black targets and armed White targets), rather than to stereotype-
congruent targets. The magnitude of this bias did not differ across
the three samples. It is interesting to note that this equivalence
emerged in spite of the fact that the civilian sample contained
many more ethnic minority members than did the predominantly
White police samples. Although any evidence of racial bias among
police may be cause for concern, there is certainly nothing in the
present data to suggest that officers show greater bias than the
people who live in the communities they serve.
We used correlational analyses to examine officers in the na-
tional sample, and, of all the variables examined, three predicted
bias in reaction times (no variables related to bias in the decision
criteria). Each of the relevant variables reflected some aspect of the
community the officer served. Bias increased as a function of the
community’s size, crime rate, and the proportion of Black resi-
dents and other ethnic minority residents. Police in larger, more
dangerous and more racially diverse environments are presumably
much more likely to encounter Black criminals, reinforcing the
stereotypic association between race and crime. By contrast, of-
ficers with little exposure to Black people may be less likely to
rehearse this association. As a consequence, these officers may
experience less stereotypic interference during the video game
task.
The results from the signal-detection analysis are particularly
provocative. Although police may have difficulty processing
stereotype-inconsistent targets (as evidenced by bias in their re-
sponse times), the SDT results suggest that police do not show bias
in their ultimate decisions. That is, the expertise that police bring
to a shoot/don’t-shoot situation may not eliminate the difficulty of
interpreting a stereotype-inconsistent target, but it does seem to
minimize the otherwise robust impact of target race on the decision
to shoot. Inasmuch as it is the actual decision to shoot (and not the
delay in making that decision) that carries life-and-death conse-
quences for the suspect, bias in the criterion may be considered the
variable of greatest interest to both the police and the community.
However, because of the profound implications of these conclu-
sions, we felt it necessary to replicate these effects. The video
game used in Study 1 imposed an 850-ms timeout window. Al-
though this restriction certainly exerts some pressure on partici-
pants, it offers them sufficient time to respond correctly on the vast
majority of trials. In Study 1, errors and timeouts, together, ac-
counted for only 9.5% of trials. When the total number of errors is
so low, idiosyncratic responses to particular targets may dramati-
cally affect the SDT estimates. In Study 2, therefore, we reduced
the time window in an effort to increase errors and obtain more
stable SDT estimates.
Study 2
Method
Participants. We returned to one police district in Denver and
recruited an additional 33 officers, as well as 52 community
members from a nearby DMV, each of whom completed a version
of the video game simulation with a more restrictive time window.
Several participants experienced great difficulty responding within
this limit, producing few errors and a very high number of time-
outs. Two officers and 7 civilians had an excessive ratio of
timeouts to incorrect trials (more than four timeouts for every
error) and were excluded from the analyses. The results do not
change substantially if they are included. The final sample in-
cluded 31 officers (3 female, 26 male, 2 missing gender; 16 White,
6 Black, 4 Latina/o, 3 other, 2 missing ethnicity; mean age 35.6
years) and 45 community members (20 female, 23 male, 2 missing
gender; 14 White, 18 Black, 10 Latina/o, 3 other; mean age 36.8
years). Officers completed the study while off duty and were paid
$50 in compensation. Community members were paid $20.
Video game simulation and procedure. The video game was
identical to that in Study 1, with the exception that the timeout
window was set to 630 ms. Participants were instructed to respond
as quickly and as accurately as possible, and response latencies
longer than 630 ms were penalized with a loss of 20 points.
Otherwise, the procedures were identical to those in Study 1.
1015
THIN BLUE LINE
Results
Our goal in reducing the timeout window was to induce a
greater number of errors. Our analysis therefore focused on the
parameters derived from the signal-detection analysis. Errors were
substantially greater in this version of the simulation. Overall,
participants made incorrect responses on 16% of the 100 trials and
timed out on 17%. We computed sensitivity (d) and the decision
criterion (c) as in Study 1, using only the correct and incorrect
trials (i.e., excluding timeouts). The estimates were analyzed in a
Sample (officer vs. civilian) Target Race (Black vs. White) 2
2 mixed-model ANOVA, with repeated measures on the latter
factor (see Table 2 for means and standard deviations; see also
Figure 3).
Signal-detection analyses. With respect to the criteria or
estimates of c, we observed that the average criterion was
significantly below zero, F(1, 74) 27.06, p .001. In fact,
the criteria in Study 2 were lower than those in the first study.
Presumably because of the increase in time pressure, partici-
pants showed a greater propensity to shoot (compare Figures 1
and 3). More interesting, the location of the criterion depended
on sample, F(1, 74) 4.95, p .03 (i.e., there was a main
effect of sample). Although the mean value of c was signifi-
cantly below zero for both the officers, F(1, 30) 4.84, p .04
(M ⫽⫺.10), and the community, F(1, 44) 29.38, p .001,
(M ⫽⫺.24), it was significantly lower for the latter. Unlike in
previous work, the main effect of target race in c was not
Figure 3. Decision criterion placement (c) and sensitivity (d) for Black and White targets as a function of
sample (Study 2).
1016
CORRELL ET AL.
significant, F 1, but the Sample Target Race interaction
was, F(1, 74) 3.69, p .059 (see Figure 3). As in Study 1,
the community sample set a lower threshold to shoot Black
targets than to shoot White targets, F(1, 44) 4.24, p .05.
Officers, on the other hand, demonstrated no racial bias, F 1.
Again replicating Study 1, this interaction seems to reflect the
fact that the community set a lower threshold for Black targets
than did the officers, F(1, 74) 9.74, p .003. The two
samples did not differ in the placement of their criteria for
White targets, F 1. It is also interesting to note that all four
of the mean c values in Figure 3 were significantly below zero,
all ts ⬍⫺2.17, ps .04, with the exception of the officers’
criterion for Black targets, t(30) ⫽⫺1.36, p .18.
Turning to sensitivity, we found that d was generally lower in
Study 2 than in Study 1, particularly for the community members,
suggesting that time pressure impaired discriminability (see Payne,
2001). The main effect of sample was significant, F(1, 74)
21.59, p .001. As in Study 1, police officers more effectively
discriminated between armed and unarmed targets (M 2.27) than
did the community members (M 1.43). The police advantage
was evident both for Black targets, F(1, 74) 26.93, p .001,
and for White targets, F(1, 74) 10.54, p .002. There was no
overall effect of target race on d, F 1, suggesting that partici-
pants, in general, were equally able to discriminate White and
Black targets. However, target race did interact marginally with
sample, F(1, 74) 2.81, p .10. Community members were
equally sensitive to both White and Black targets, F 1, but
officers showed marginally greater sensitivity for Black, rather
than White, targets, F(1, 31) 3.53, p .07 (see Figure 3). The
results from Study 1 similarly indicated better sensitivity among
officers than civilians, particularly for the Black targets.
Reaction-time analyses. Previous work has consistently found
that reducing the time window eliminates the race-bias effect in
response times, presumably because it reduces variance in the
latencies (see Correll et al., 2002). Consistent with those findings,
bias in response times was not significant on average in Study 2,
F 1, nor did the magnitude of bias depend on sample, F 1.
Discussion
Like Study 1, Study 2 revealed critical differences between the
performance of police officers and that of civilians. These differ-
ences emerged both in the participants’ ability to discriminate
armed from unarmed targets and in the criterion for the decision to
shoot. Civilians consistently set a lower threshold for the decision
to shoot (c) than did the officers, and this difference was particu-
larly evident for Black targets. In both studies, officers showed
greater sensitivity (d), and again this tended to be particularly true
with Black targets. In sum, then, Study 2 replicated the signal-
detection findings of Study 1, and it did so using a paradigm that
forced participants to respond very quickly, resulting in a greater
number of errors and, so, more stable SDT estimates.
Taken together, the response-time results from Study 1 and the
signal-detection results from both Studies 1 and 2 reveal intriguing
differences between trained police officers and civilians who live
in the communities those officers serve. The latencies suggest that
officers and community members both experienced difficulty pro-
cessing stereotype-incongruent targets. Like community members,
police were slower to make correct decisions when faced with an
unarmed Black man or an armed White man. It is important to
note, however, that the officers differed dramatically from the
civilians in terms of the decisions they ultimately made. Commu-
nity members showed a clear tendency to favor the shoot response
for Black targets (relative to both White targets and relative to a
neutral or balanced criterion of zero). Police, however, showed no
bias in their criteria. Moreover, they showed greater discriminabil-
ity and a less trigger-happy orientation in general (i.e., for both
Black and White targets). These results seem to suggest that
expertise improves the outcome of the decision process (increasing
sensitivity and reducing the unwarranted tendency to shoot, par-
ticularly for Black targets), even though it may not eliminate
processing difficulties associated with stereotype-inconsistent tar-
gets.
We have suggested that this reduction in bias may reflect the
impact of training. In Study 3 we attempted to examine this
possibility more systematically by providing practice on the video
game task to a sample of undergraduates. On the basis of the
results of Studies 1 and 2, we expected that repeated play would
improve sensitivity (facilitating discrimination between armed and
unarmed targets) and reduce racial bias in the placement of the
decision criterion (Plant et al., 2005). But we expected that practice
would not reduce bias in response times. Like the officers, partic-
ipants with more practice on the task should demonstrate improve-
ments in their ultimate decisions in spite of persistent difficulty
with the processing of stereotype-incongruent targets.
Study 3
Method
Participants. Fifty-eight students (29 female, 22 male, 7 miss-
ing gender; 40 White, 1 Black, 3 Asian, 3 Latina/o, 1 Native
American, 2 Other, 8 missing ethnicity) participated in Study 3
either in partial completion of a course requirement or for $15 pay.
Four additional students were included in the original sample but
failed to return for Day 2 and thus are excluded from all analyses.
Video game simulation and procedure. In Study 3, partici-
pants played the video game twice on each of 2 days separated by
48 hr. At each round of play, they completed an 80-trial shoot/
don’t-shoot video game, which was essentially the same as the task
performed in Study 1. This game again used a timeout window of
850 ms. Thus, the design included four factors: 2 (Day) 2
(Round of Play) 2 (Race) 2 (Object), with repeated measures
on all four variables. This design allowed us to examine the
effects of repeated play within a day and also to assess whether
any improvement in performance would carry over from Day 1
to Day 2.
Results
We computed SDT estimates and average reaction times for
correct responses as in Studies 1 and 2.
Signal-detection analyses. We analyzed the SDT estimates as a
function of day (1 vs. 2), round of play (1 vs. 2), and target race (Black
vs. White) using 2 2 2 repeated-measures ANOVAs for both c
and d. Analyses of c revealed that, on average, participants set a
lower criterion to shoot for Black targets than to shoot White targets,
F(1, 57) 10.76, p .002. It is critical, however, that the effect of
1017
THIN BLUE LINE
race depended on round, F(1, 57) 5.08, p .03, such that bias
decreased in the latter round each day. That is, the race difference in
the criterion (i.e., bias) was significant at Round 1 on both Day 1,
t(57) 2.41, p .02, and on Day 2, t(57) 2.53, p .02. But bias
fell to nonsignificant levels at Round 2 on both days: for Day 1,
t(57) 0.17, p .86; for Day 2, t(55) ⫽⫺0.06, p .95 (see Figure
4). Moreover, the Round Race interaction did not depend on day,
F(1, 57) 0.04, p .84. No other effects in this analysis were
statistically significant, all Fs(1, 57) 1.04, ps .31. As predicted
then, practice reduced bias in the decision to shoot, and it did so on
each of the two days. It is interesting that there appeared to be no carry
over in bias reduction from Day 1 to Day 2. We return to this issue in
the Discussion section.
The analysis of sensitivity, or d, revealed only a main effect of
round, F(1, 57) 7.09, p .01, reflecting greater discriminability
during the second game each day. No other effects in this analysis
were statistically significant, all Fs(1, 57) 1.06, ps .30 (see
Figure 4). As predicted, practice enhanced sensitivity and seemed
to have equivalent effects for both Black and White targets. More-
over, the increase in sensitivity occurred each day, and there was
no evidence that the increase carried over from Day 1 to Day 2.
Reaction-time analyses. Latencies were analyzed as a function
of day (1 vs. 2), round of play (1 vs. 2), target race (Black vs.
White), and object type (gun vs. nongun) using a 2 2 2 2
repeated-measures ANOVA. As usual, we observed a main effect
of object, F(1, 57) 409.19, p .001, such that participants
Figure 4. Decision criterion placement (c) and sensitivity (d) for Black and White targets as a function of day
and round of play (Study 3).
1018
CORRELL ET AL.
responded more quickly on gun trials than on nongun trials. This
effect was qualified by an interaction between target race and
object type, F(1, 57) 95.65, p .001, representing significant
racial bias. Our primary concern, however, involved the degree to
which this pattern changed as participants gained experience with
the task. Most interesting, from our perspective, was the question
of whether repeated play altered the magnitude of racial bias in the
speed with which participants could make shoot/don’t-shoot deci-
sions. In stark contrast to the SDT results, bias in reaction times
did not change as a function of round: The three-way interaction
was not significant, F(1, 57) 0.01, p .93. Similarly, neither the
Day Race Object three-way interaction, F(1, 57) 0.01, p
.92, nor the Round Day Race Object four-way interaction
was significant, F(1, 57) 0.00, p .95. In essence, the magni-
tude of this bias did not change over the course of the study.
Further, latency bias was significant in both Round 1, F(1, 57)
33.76, p .001, and Round 2, F(1, 57) 28.52, p .001, on Day
1, as well as Round 1, F(1, 57) 27.04, p .001, and Round 2,
F(1, 57) 17.14, p .001, on Day 2 (see Figure 5).
5
So although
practice decreased racial bias in the decision criteria and improved
overall discriminability (as shown by the SDT analyses), practice
did not attenuate racial bias in reaction times.
Discussion
Participants in Study 3 showed a number of changes as a
function of practice. Most important, practice with the task re-
duced SDT bias and increased sensitivity to the presence or ab-
sence of a weapon. Practice did not, however, affect the magnitude
of racial bias in latencies. Across repeated plays of the video game
simulation, these developing “experts” continued to struggle with
the stereotype-incongruent targets, responding more slowly on
incongruent (compared with congruent) trials.
The effects of training observed in this study with a sample of
undergraduates largely replicate the differences observed between
police officers and civilians in Studies 1 and 2. Undergraduates in
the initial round of Study 3, like members of the Denver commu-
nity, showed bias both in latencies and in their criteria for the
decision to shoot. These effects were evident on both Day 1 and
Day 2. After receiving practice on the shoot/don’t-shoot simulation
task, however, bias in the placement of the criterion diminished,
but bias in reaction times did not change. As a consequence of this
shift, our “expert” participants began to look less like community
members and more like police officers.
However, a single round of practice with our video game task
(which takes roughly 12 min–15 min) differs dramatically from the
training that police receive. As noted above, Denver police recruits
spend approximately 72 hr in weapons training during their time at
the academy. This extended, in-depth practice likely results in
much greater consolidation of the skills necessary to exert control
over their behavior than did the minimal practice afforded to
participants in Study 3. Consistent with this, participants in Study
3 showed pronounced within-day improvements (reductions in
bias and increases in discriminability), but they showed no evi-
dence that this training carried over from Day 1 to Day 2. Upon
entering the lab on Day 2 (48 hr after the Day 1 session), partic-
5
A number of less theoretically interesting effects that did not involve
race and object were present in this analysis. Overall, participants were
faster on Day 2 than Day 1, F(1, 57) 46.94, p .001, marginally faster
at Round 2 than Round 1, F(1, 57) 3.40, p .07, and the Day Round
interaction was significant, F(1, 57) 11.76, p .002, such that the
Round 1 to Round 2 decrease in mean latencies was really only present on
Day 1. (It is interesting that this increase in speed again mirrors sample
differences between the police and community participants in Studies 1 and
2.) The object main effect (faster times to gun trials) was qualified by a
number of interactions. The difference in gun versus no-gun trials was
greater on Day 1 than Day 2, F(1, 57) 15.69, p .001, for the Day
Object interaction, greater at Round 1 than Round 2, F(1, 57) 6.64, p
.02, for the Round Object interaction, and the shift from Round 1 to
Round 2 was really only present on Day 1, F(1, 57) 4.16, p .05, for
the Day Round Object interaction. All of these effects reflect
accelerations in classification speed (for all responses or for the particularly
slow no-gun responses). This acceleration is most pronounced at early
stages of the study and weakens over time, presumably because of a floor
effect.
Figure 5. Response times to Black and White armed and unarmed targets as a function of day and round of
play (Study 3).
1019
THIN BLUE LINE
ipants behaved like novices. On Round 1 of their second day, they
demonstrated racial bias in both response times and SDT criteria.
With additional training on Day 2, this bias dropped once again.
But the reemergence of bias in Round 1 of Day 2 suggests more
extensive training is necessary if participants are to more perma-
nently overcome bias in behavioral responses. The fact that police
officers in Studies 1 and 2 showed no SDT bias during their initial
performance on the video game task may be a testament to their
training and expertise.
General Discussion
We began this research with two primary goals: examining
police officers in a first-person shoot/don’t-shoot task and com-
paring their performance with that of a community sample. This
investigation assessed overall proficiency and the role that a tar-
get’s race plays in the decision-making process. Police differed
from the community members in terms of several critical variables.
On average (ignoring target race), the officers clearly outper-
formed the community sample. They were faster to make correct
responses; they were better able to detect the presence of a weapon
(as measured by d); and they set a significantly higher criterion (c)
for the decision to shoot, indicating a less “trigger-happy” orien-
tation.
Most important for our hypothesis, the officers also differed
from the community sample in the role that a target’s race played
in the placement of SDT criteria for the decision to shoot. This
difference primarily affected Black targets. When the target was
White, all of the samples (Denver community, Denver police, and
national police) set a relatively high criterion, and none of the
samples differed from one another. But when the target was Black,
the community set a significantly lower (more trigger-happy)
criterion than the officers. This was true both in Study 1, which
used a relatively long timeout window, and in Study 2, in which
the timeout window was substantially reduced (yielding much
higher error rates).
In spite of the fact that police showed minimal bias in the SDT
analysis, the officers were similar to the community sample (and to
literally hundreds of past participants in this paradigm) in the
manifestation of robust racial bias in the speed with which they
made shoot/don’t-shoot decisions. Accurate responses to targets
congruent with culturally prevalent stereotypes (i.e., armed Black
targets and unarmed White targets) required less time than did
responses to stereotype-incongruent targets (i.e., unarmed Black
targets and armed White targets). Evidence of bias in response
latencies was consistent and robust across all of the samples
examined in Study 1: the national sample and the Denver sample
of police officers, as well as the Denver community sample, drawn
from the neighborhoods that the Denver officers serve.
The results from Study 3, in which we trained novice college
students on the task, revealed similar effects. Across two rounds of
play, student participants showed a significant decrease in racial
bias, as measured by the decision criterion, accompanied by an
increase in sensitivity. But they showed no change in the magni-
tude of bias as measured by response latencies. An identical
pattern was obtained when students returned for a second day,
during which they again completed two rounds of the video game
task. In the first round of play, student performance mirrored that
of the community; By Round 2, it mirrored that of the police
officers.
The performance of the officers and the expert students in these
studies raises an important set of questions about the processes that
differentiate bias in response times from bias in the threshold to
shoot. Typically, errors and latencies follow a similar pattern, such
that greater difficulty on a given trial increases both response time
and the likelihood of a mistake, as observed in the performance of
community members and novice college students. The officers and
experts, by contrast, showed clear bias in latencies, but target race
had no impact on their ultimate decisions.
To the extent that longer latencies reflect difficulty, the persis-
tent bias in reaction times suggests that even experts have some
trouble processing stereotype-incongruent targets. The visual com-
plexity of the stimuli may essentially require participants to engage
in an effortful, serial search for relevant information about the
object (Shiffrin & Schneider, 1977). At the same time, the salience
and automaticity that generally characterize psychological pro-
cessing of racial cues (Cunningham et al., 2004; Ito & Urland,
2003) suggest that— during the course of that search—participants
are likely to encode target race. In combination with tenacious
racial stereotypes (e.g., Devine & Elliot, 1995), race-based pro-
cessing may impede responses to counterstereotypic targets. In line
with this possibility, Study 1 showed that officers from urban,
high-crime, predominantly minority districts (environments likely
to reinforce stereotypes about Black people) showed greater racial
bias in their latencies.
For officers (and, temporarily, for trained undergraduates), how-
ever, the stereotypic interference ended with reaction times. The
bias evident in their latencies did not translate to the decisions they
ultimately made. This separation of effects may reflect the offic-
ers’ ability to override automatic associations (Kunda & Spencer,
2003), perhaps as a function of their training and expertise. Police
(with extensive training) and “expert” undergraduates (with min-
imal training) were able to reduce bias in the SDT criteria for
Black and White targets. Were these individuals able to avoid snap
judgments on ambiguous trials, such as those posed by counter-
stereotypic targets, and wait for a more complete understanding?
Such a delay when responding to difficult-to-process counterste-
reotypic targets would presumably yield bias in reaction times
(consistent with the data). At the same time, it would minimize
bias in the decision criteria and increase overall accuracy. Anec-
dotally, this explanation matches officers’ intuitions about the
process. In a conversation about the effects reported here, one
officer stated that the findings “make sense” because police are
trained to hold their fire if they are uncertain to wait for greater
clarity.
The possibility that expertise and practice enhance control res-
onates with research beyond the realm of racial stereotyping.
Green and Bavelier (2003) have shown that practice with visually
complex video games enhances visual attention (but practice with
visually simple games does not). And, although practice on a
simple decision task generally promotes automaticity (Bush et al.,
1998; Shiffrin & Schneider, 1977), practice on more complicated
interference tasks or on challenging working-memory tasks can
actually increase control (Olesen et al., 2004; Weissman et al.,
2002). On the basis of functional magnetic resonance imaging,
these studies show that extended practice on difficult tasks leads to
increased activation of the medial and middle frontal gyri—areas
1020
CORRELL ET AL.
associated with control-based conflict resolution and top-down,
rule-based processing. We suggest, then, that police training and
on-the-job experience in complex encounters may allow officers to
more effectively exert executive control in the shoot/don’t-shoot
task, essentially overriding response tendencies that stem from
racial stereotypes. As noted above, the correlational analyses from
Study 1 identified several environmental factors that were associ-
ated with increases in latency bias for the officers (i.e., serving in
urban, high-crime, and predominantly minority districts). It is
interesting to note that these same variables had no impact on the
SDT criteria the officers used.
We do not want to suggest that the minimal training provided in
Study 3 parallels the sort of training that police officers receive.
However, the possibility that police function as highly trained
subjects is intriguing. In the current research, evidence for this
possibility relies on cross-sectional comparisons (Studies 1 and 2)
and on parallels between samples that differ in numerous ways
(i.e., the “expert” students in Study 3 and the police officers). It
would be informative to follow police recruits as they enter the
academy, as they receive training, and as they cope with their first
years of patrol duty. We have begun data collection on such a
project. At present, we have data from 39 recruits in the first weeks
of training at the police academy (prior to any weapons training).
It is striking that these recruits show statistically significant racial
bias in both reaction times and in the decision criteria. Upon
entering the academy, then, recruits behave very much like the
community samples (Studies 1 and 2) and the novice student
sample (Study 3): They set a lower criterion for Black targets than
for White targets. These data are entirely consistent with the
possibility that the reduction in SDT bias among police officers
represents an expertise effect. These data also argue against the
suggestion that police academies or departments indoctrinate their
members into a culture of anti-Black sentiment (Teahan, 1975a), at
least with respect to the sort of judgments studied here.
We must note that our results are only partially consistent with
prior work. Consistent with Eberhardt et al. (2004), we found that
officers orient more quickly to Black people when processing
danger-related stimuli. With respect to reaction times, our results
(like theirs) suggest a bias in attentional focus and processing. But
our data are not consistent with those of Plant and Peruche (2005),
who found that officers showed racial bias in the SDT criteria for
the decision to shoot. Although these officers learned to eliminate
bias over the course of the study, the presence of the initial bias is
inconsistent with our results. Officers in the current studies never
showed significant evidence of bias.
6
This partial correspondence may stem from a variety of factors.
We explore two. First, Plant and Peruche (2005) sampled 50
officers from Florida; in Study 1, we sampled 237 officers from
Colorado and 14 other states. It is possible that the differences
between our findings reflect regional differences between Florida
and other areas of the country. Second, it is possible that the results
reflect differences between the paradigms employed. Plant and
Peruche’s stimuli are, arguably, further removed from the training
and experience of police officers than are the stimuli presented in
our simulation. Plant and Peruche presented Black and White male
faces on which objects (e.g., a gun or wallet) had been superim-
posed. Our stimuli involve full-body images of men holding guns
and other objects. These images are embedded in scenes, such as
parks or cityscapes. To the extent that our stimuli more closely
mirror police training (e.g., Firearms Training System or firing
range encounters) and on-the-job experiences, an officer’s exper-
tise should be more likely to generalize to our task. To the extent
that Plant and Peruche’s paradigm is less similar to the officers’
previous experiences, their participants may have had to learn what
was, in essence, a novel task.
As we discussed in the introduction, sociologists have studied
the question of racial bias in police shootings for many years. The
sociological literature provides a rich, if complicated, context in
which to view the results of the current studies. One account that
has received substantial attention is that police shoot Black sus-
pects more often than White suspects, per capita, because Black
people are disproportionately likely to be involved in crime (par-
ticularly violent crime). The Department of Justice (2001) report
shows that, just as Black suspects are five times more likely than
White suspects to die at the hands of police, police officers are five
times more likely to die at the hands of a Black suspect than a
White suspect. In a similar vein, Reisig et al. (2004) found that the
use of nonlethal force (which seems to depend on suspect race)
may actually reflect race-based differences in the suspect’s pro-
pensity to resist arrest or engage in belligerent behavior toward
officers. It is the suspect’s hostility, they argue—not race—that
prompts a hostile response from the officer. And Inn et al. (1977)
report that the number of Black suspects shot by police is propor-
tionate to the number of Black suspects arrested. They tentatively
conclude that it is the prevalence of criminal activity among Black
people that drives the differential shooting rates. (The authors note,
however, that arrest rates themselves may reflect biases held by the
police and thus do not provide a perfect standard of comparison.)
In line with this reasoning, in Study 1, officers from the national
sample who reported working in communities with (a) high levels
of violent crime and (b) high proportions of minority residents
showed particularly strong patterns of bias in their latencies. Did
their experiences with minority suspects foster associations that
made counterstereotypic trials particularly difficult to process?
The situation is almost certainly more complex. It is clear from
the analysis of Study 1 that officers serving in heavily (more
densely) populated communities also showed greater anti-Black
bias in their reaction times. In combination, these variables seem to
suggest that racial bias in the decision to shoot may reflect the
disproportionate representation of Black people (and perhaps other
ethnic minority groups) in low-income, poverty-stricken, and high-
crime areas. Geller (1982) and Smith (2004) presented evidence
that a greater number of police shootings occur in disadvantaged
neighborhoods and that members of ethnic minorities are more
likely to be killed in these incidents. Using regression models to
predict officer-involved shootings, Terrill and Reisig (2003)
showed that, once neighborhood risk is taken into account, the
6
In light of Plant and Peruche’s (2005) findings, we explored the
possibility that police officers in the current studies showed a decrease in
bias over the course of the shooter task. To examine this possibility, we
reanalyzed the data from Studies 1 and 2, separating the 100 trials into two
50-trial blocks and analyzing SDT estimates (both c and d) as a function
of sample, target race, and block (first half vs. second half). Neither
three-way interaction was significant, and controlling for block did not
alter the findings reported in the text. These data provide no evidence that
police showed less bias than community members because they were better
able to improve their performance over the course of the task.
1021
THIN BLUE LINE
effect of suspect race or ethnicity is no longer statistically reliable.
This research builds on the ecological contamination hypothesis,
first advanced by Werthman and Piliavin (1967), which suggests
that the reputation of a neighborhood distorts perceptions of its
inhabitants. To the extent that a community is seen as a “bad area,”
police may perceive the individuals who live there (or anyone they
happen to encounter there) as a potential threat. If members of
minorities are more likely to live and spend time in disadvantaged
neighborhoods (Sidanius & Pratto, 1999), they may also be more
likely to fall victim to this context-based contamination. As a
consequence, police may be more likely to shoot a Black suspect
because of the context in which the encounter occurs, not because
of racial bias, per se (Fyfe, 1981). In an interesting wrinkle of this
argument, Sampson and Raudenbush (2004) conducted an exten-
sive investigation of the factors that predict perceived community
disorder—the causal variable proposed by ecological contamina-
tion. They found that the mere presence of Black people in a
community is sufficient to evoke the perception of disadvantage.
That is, controlling for objective factors (e.g., prevalence of graf-
fiti, broken windows, and abandoned buildings), the greater the
number of Black people living in an area, the greater the disorder
perceived by both Black and non-Black citizens. If Black people
evoke the perception of neighborhood disadvantage, they may
experience harsher treatment by police—not because the police are
biased to treat Black people in a hostile fashion, but because Black
neighborhoods seem more threatening.
The data presented here suggest that, although police officers
may be affected by culturally shared racial stereotypes (i.e., show-
ing bias in their response times), they are no more liable to this bias
than are the people who live and work in their communities.
Further, at least on the simulation used here, the officers’ ultimate
decisions about whether or not to shoot are less susceptible to
racial bias than are the decisions of community members. The data
suggest that the officers’ training and/or expertise may improve
their overall performance (yielding faster responses, greater sen-
sitivity and reduced tendencies to shoot) and decrease racial bias in
decision outcomes. We feel that this research represents a valuable
melding of basic social psychological processes with an issue of
great importance to our society. By examining the influence of
race in the automatic processing of danger-related stimuli, and the
capacity of expertise to moderate this effect, these findings touch
on a topic of great interest to social psychologists, sociologists,
police, and community groups, alike. The investigation of racial
bias in police use of force presents a unique opportunity to apply
experimental social psychological methods to an issue that is vital
to the members of increasingly diverse neighborhoods and com-
munities.
References
Amodio, D. M., Harmon-Jones, E., Devine, P. G., Curtin, J. J., Hartley,
S. L., & Covert, A. E. (2004). Neural signals for the detection of race
bias. Psychological Science, 15, 88–93.
Birnboim, S. (2003). The automatic and controlled information-processing
dissociation: Is it still relevant? Neuropsychology Review, 13, 19 –31.
Bush, G., Whalen, P. J., Rosen, B. R., Jenike, M. A., McInerney, S. C.,
Rauch, S. L. (1998). The counting Stroop: An interference task special-
ized for functional neuroimaging—Validation study with functional
MRI. Human Brain Mapping, 6, 270 –282.
Christopher, W. (1998). Report of the Independent Commission on the Los
Angeles Police Department. Darby, PA: DIANE Publishing.
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.
Correll, J., Park, B., Judd, C. M., & Wittenbrink, B. (2002). The police
officer’s dilemma: Using ethnicity to disambiguate potentially threaten-
ing individuals. Journal of Personality and Social Psychology, 83,
1314 –1329.
Correll, J., Park, B., Judd, C. M., & Wittenbrink, B. (2007). Stereotypes
and racial bias in the decision to shoot. Manuscript submitted for
publication.
Correll, J., Urland, G. L., & Ito, T. A. (2006). Event-related potentials and
the decision to shoot: The role of threat perception and cognitive control.
Journal of Experimental Social Psychology, 42, 120 –128.
Cunningham, W. A., Johnson, M. K., Raye, C. L., Gatenby, J. C., Gore,
J. C., & Banaji, M. R. (2004). Separable neural components in the
processing of Black and White faces. Psychological Science, 15, 806
813.
Department of Justice. (2001). Policing and homicide, 1976 –98: Justifi-
able homicide by police, police officers murdered by felons (NCJ
180987). Washington, DC: Bureau of Justice Statistics.
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.
Eberhardt, J. L., Goff, P. A., Purdie, V. J., & Davies, P. G. (2004). Seeing
black: Race, crime, and visual processing. Journal of Personality and
Social Psychology, 87, 876 893.
Fyfe, J. J. (1981). Who shoots? A look at officer race and police shooting.
Journal of Police Science and Administration, 9, 367–382.
Geller, W. A. (1982). Deadly force: What we know. Journal of Police
Science and Administration, 10, 151–177.
Green, C. S., & Bavelier, D. (2003, May 29). Action video game modifies
visual selective attention. Nature, 423, 534 –537.
Greenwald, A. G., Oakes, M. A., & Hoffman, H. G. (2003). Targets of
discrimination: Effects of race on responses to weapons holders. Journal
of Experimental and Social Psychology, 39, 399 405.
Hamilton, D. L., & Trolier, T. K. (1986). Stereotypes and stereotyping: An
overview of the cognitive approach. In J. F. Dovidio, & S. L. Gaertner
(Eds.), Prejudice, discrimination, and racism (pp. 127–163). San Diego,
CA: Academic Press.
Inn, A., Wheeler, A. C., & Sparling, C. L. (1977). The effects of suspect
race and situation hazard on police officer shooting behavior. Journal of
Applied Social Psychology, 7, 27–37.
Ito, T. A., & Urland, G. R. (2003). Race and gender on the brain:
Electrocortical measures of attention to the race and gender of multiply
categorizable individuals. Journal of Personality and Social Psychology,
85, 616 626.
Jansma, J. M., Ramsey, N. F., Slagter, H. A., & Kahn, R. S. (2001).
Functional anatomical correlates of controlled and automatic processing.
Journal of Cognitive Neuroscience, 13, 730 –743.
Kelly, A. M. C., & Garavan, H. (2004). Human functional neuroimaging of
brain changes associated with practice. Cerebral Cortex, 15, 1089
1102.
Kunda, Z., & Spencer, S. J. (2003). When do stereotypes come to mind and
when do they color judgment? A goal-based theory of stereotype acti-
vation and application. Psychological Bulletin, 129, 522–544.
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.
1022
CORRELL ET AL.
Olesen, P. J., Westerberg, H., & Klingberg, T. (2004). Increased prefrontal
and parietal activity after training of working memory. Nature Neuro-
science, 7, 75–79.
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. (2002). Best laid plans:
Effects of goals on accessibility bias and cognitive control in race-based
misperceptions of weapons. Journal of Experimental and Social Psy-
chology, 38, 384 –396.
Plant, E. A., & Peruche, B. M. (2005). The consequences of race for police
officers’ responses to criminal suspects. Psychological Science, 16,
180 –183.
Plant, E. A., Peruche, B. M., & Butz, D. A. (2005). Eliminating automatic
racial bias: Making race non-diagnostic for responses to criminal sus-
pects. Journal of Experimental and Social Psychology, 41, 141–156.
Reisig, M. D., McCluskey, J. D., Mastrofski, S. D., & Terrill, W. (2004).
Suspect disrespect toward the police. Justice Quarterly, 21, 241–268.
Sampson, R. J., & Raudenbush, S. W. (2004). Seeing disorder: Neighbor-
hood stigma and the social construction of “broken windows.” Social
Psychology Quarterly, 67, 319 –342.
Shiffrin, R., & Schneider, W. (1977). Controlled and automatic human
information processing: II. Perceptual learning, automatic attending, and
a general theory. Psychological Review, 84, 127–190.
Sidanius, J., & Pratto, F. (1999). Social dominance: An intergroup theory
of social hierarchy and oppression. New York: Cambridge University
Press.
Smith, B. W. (2004). Structural and organizational predictors of homicide
by police. Policing: An International Journal of Police Strategies and
Management, 27, 539 –557.
Sorenson, J. R., Marquart, J. W., & Brock, D. E. (1993). Factors related to
killings of felons by police officers: A test of the community violence
and conflict hypotheses. Justice Quarterly, 10, 417– 440.
Stroop, J. R. (1935). Studies of interference in serial verbal reaction.
Journal of Experimental Psychology, 18, 643– 662.
Sunshine, J., & Tyler, T. R. (2003). The role of procedural justice and
legitimacy in shaping public support for policing. Law & Society Re-
view, 37, 513–548.
Teahan, J. E. (1975a). A longitudinal study of attitude shifts among black
and white police officers. Journal of Social Issues, 31, 47–56.
Teahan, J. E. (1975b). Role playing and group experience to facilitate
attitude and value changes among Black and White police officers.
Journal of Social Issues, 31, 35– 45.
Terrill, W., & Reisig, M. D. (2003). Neighborhood context and police use
of force. Journal of Research in Crime & Delinquency, 40, 291–321.
Weissman, D. H., Woldorff, M. G., Hazlett, C. J., & Mangun, G. R. (2002).
Effects of practice on executive control investigated with fMRI. Cogni-
tive Brain Research, 15, 47–60.
Weitzer, R., & Tuch, S. A. (2004). Race and perceptions of police mis-
conduct. Social Problems, 51, 305–325.
Werthman, C., & Piliavin, I. (1967). Gang members and the police. In D.
Bordua (Ed.), The police: Six sociological essays. New York: Wiley.
Received May 1, 2006
Revision received September 17, 2006
Accepted September 28, 2006
1023
THIN BLUE LINE
... Interestingly, issued exactly this caution in the very first study on the FPST ("it is not yet clear that Shooter Bias actually exists among police officers … there is no reason to assume that this effect will generalize beyond [lay samples]," p. 1328). Yet despite this and later warnings (Cox & Devine, 2016), researchers continued to apply studies from undergraduates to police officers, even as data came to light that police officers did not show the same bias (e.g., Correll et al., 2007. The fact that trained officers may use information in the decision landscape differently than untrained undergraduates represents an important ability difference between the two groups. ...
... If experts attend to different decision components or use these components differently than novices, and this difference changes the effect of target race on the ultimate decision, then the conclusion of widespread race bias in officers' deadly force decisions based on findings from undergraduate participants will be unwarranted. Indeed, sworn officers typically show little to no bias in the behavioral decision to shoot with the standard FPST (e.g., Akinola, 2009;Correll et al., 2007;Ma & Correll, 2011;Sim, Correll, & Sadler, 2013;Taylor, 2011), and this is especially true for studies using immersive shooting simulators such as the one described above (e.g., James et al., 2013James et al., , 2014. Cesario and Carrillo (in press) summarized studies in which sworn officers completed the standard FPST and found that out of 64 possible tests for racial bias, only ∼25% showed anti-Black bias whereas ∼70% showed no bias on the part of officers in one direction or the other. ...
Article
Full-text available
We endorse Cesario's call for more research into the complexities of “real-world” decisions and the comparative power of different causes of group disparities. Unfortunately, these reasonable suggestions are overshadowed by a barrage of non sequiturs, misdirected criticisms of methodology, and unsubstantiated claims about the assumptions and inferences of social psychologists.
... People are then more likely to make a "shoot" decision, which increases errors. Follow-up studies found that exposure to media stories about crimes committed by Black individuals and increasing prevalence of armed Black individuals within the simulations exacerbated shooter bias by confirming the stereotype and increasing the strength of the association with aggression (Correll et al., 2007a(Correll et al., , 2007b. Elsewhere, the shooter bias has been found for West Asian men in European countries with large migrant populations (Essien et al., 2017;Fleming et al., 2010). 2 Schematic association between Black faces and weapons has yielded effects in other relevant decision-making outcomes as well. ...
... Not all studies implicate a shooter bias, however.Correll et al. (2007b) compared trained police officers to a sample of college students and found the police did not exhibit such a bias against unarmed African Americans, andJames et al. (2016) found that officers are less likely to shoot African Americans whether armed or unarmed. ...
Article
Full-text available
Much research has found that implicit associations between Black male faces and aggression affect dispositional judgments and decision-making, but there have been few investigations into downstream effects on explicit episodic memory. The current experiment tested whether such implicit associations interact with explicit recognition memory using an associative memory paradigm in younger and older adults. Participants studied image pairs featuring faces (of Black or White males) alongside handheld objects (uncategorized, kitchenware, or weapons) and later were tested on their recognition memory for faces, objects, and face/object pairings. Younger adults were further divided into full and divided attention encoding groups. All participants then took the race faces implicit association test. Memory for image pairs was poorer than memory for individual face or object images, particularly among older adults, extending the empirical support for the age-related associative memory deficit hypothesis (Naveh-Benjamin in J Exp Psychol Learn Mem Cognit 26:1170–1187, 2000) to associations between racial faces and objects. Our primary hypothesis—that older adults’ associative memory deficit would be reduced under Black/weapon pairings due to their being schematically related stimuli—was not confirmed. Younger adults and especially older ones, who were predominantly White, exhibited an own-race recognition bias. In addition, older adults showed more negative implicit bias toward Black faces. Importantly, mixed linear analyses revealed that negative implicit associations for Black faces predicted increased explicit associative memory false alarm rates among older adults. Such a pattern may have implications for the criminal justice system, particularly when weighting eyewitness testimony from older adults.
... Beside the evaluative priming task and the IAT (Fazio et al., 1986;Greenwald et al., 1998), a wide range of tools were developed in the implicit social cognition field and now falls under the umbrella of indirect evaluative responses. Nowadays, the variety of measures researchers can use is quite substantial: the Go/No-Go Association Task (Nosek & Banaji, 2001), the Shooter Bias Task (Correll et al., 2007), the Extrinsic Affective Simon Task (De Houwer, 2003), the Affective Misattribution Procedure (AMP; Payne et al., 2005), the Stereotype Misperception Task (SMT; Krieglmeyer & Sherman, 2012), the Implicit Relational Assessment Procedure (Barnes-Holmes et al., 2010), or the Relation Responding Task or Approach-Avoidance Task (Chen & Bargh, 1999;Rougier et al., 2018; for a recent review on these measures, see Gawronski & Hahn, 2019). ...
Thesis
In the field of implicit social cognition, indirect evaluative responses represent an opportunity to overcome some of the limitations of self-report. Theoretically capturing something progressively encoded over time and guiding our behaviors, these measures would allow us to determine the attitude that people have towards something, even when these people would not or could not reveal their preferences. For most theoretical models accounting for these behavioral responses, it is through repeated experience that we develop indirect evaluative responses. Recent experimental work, however, highlighted the impact that simple instructions can have on these evaluative responses. Throughout this dissertation, we argue that the effects of repeated experience and simple instructions differ. To investigate this question, we first developed an approach-avoidance training paradigm in which our participants were asked to repeatedly approach and avoid stimuli. After showing that new indirect evaluative responses emerged from this type of training (Exp. 1a–2), we compared this experimental paradigm to an instruction-based procedure (Exp. 3–7). Of these five studies comparing the two procedures on several types of indirect evaluative responses, across different populations, and in different situations, two revealed greater effectiveness of approach and avoidance training (the other three were inconclusive). Two additional experiments addressed the issue of naive theories that individuals might have about this issue (Exp. 8 & 9). Taken together, these results are consistent with recent theoretical advances in the field of implicit social cognition and lead us to recommend paradigms such as approach and avoidance training over paradigms based on simple instructions.
... (Reay et al. 2018;Ågård et al. 2012;Anderson et al. 2018;Bakken and Gilljam 2003;Burrell et al. 2013;Knighton 2004;Larsen 2001;O'Hara et al. 2015;Oosterwold et al. 2018;Vickers and Lewinski 2012). In the presence of perceived immediate personal physical threat, actions are more likely to be self-preservative even when further analysis may have resulted in either the perceived threat being dismissed, or more appropriate courses of action being identified(Correll et al. 2007;Gamble et al. 2018;Harman et al. 2019;Lima and de Araujo 2018; ...
Article
Full-text available
Military and emergency response remain inherently dangerous occupations that require the ability to accurately assess threats and make critical decisions under significant time pressures. The cognitive processes associated with these abilities are complex and have been the subject of several significant, albeit service specific studies. Here, we present an attempt at finding the commonalities in threat assessment, sense making, and critical decision-making for emergency response across police, military, ambulance, and fire services. Relevant research is identified and critically appraised through a systematic literature review of English-language studies published from January 2000 through July 2020 on threat assessment and critical decision-making theory in dynamic emergency service and military environments. A total of 10,084 titles and abstracts were reviewed, with 94 identified as suitable for inclusion in the study. We then present our findings focused on six lines of enquiry: Bibliometrics, Language, Situation Awareness, Critical Decision Making, Actions, and Evaluation. We then thematically analyse these findings to reveal the commonalities between the four services. Despite existing single or dual service studies in the field, this research is significant in that it is the first examine decision making and threat assessment theory across all four contexts of military, police, fire and ambulance services, but it is also the first to assess the state of knowledge and explore the extent that commonality exists and models or practices can be applied across each discipline. The results demonstrate all military and emergency services personnel apply both intuitive and formal decision-making processes, depending on multiple situational and individual factors. Institutional restriction of decision-making to a single process at the expense of the consideration of others, or the inappropriate training and application of otherwise appropriate decision-making processes in certain circumstances is likely to increase the potential for adverse outcomes, or at the very least restrict peak performance being achieved. The applications of the findings of the study not only extend to facilitating improved practice in each of the individual services examined, but provide a basis to assist future research, and contribute to the literature exploring threat assessment and decision making in dynamic contexts.
Article
The sociology of violence has undergone a tremendous change over the past 10 years, increasingly arguing that situational factors are key to violence emergence, rather than context factors. Yet, many key questions regarding this novel situational approach remain unanswered: How can situation and context be conceptually specified? Can context be integrated into a situational explanatory model? And what causal understanding underlies situational approaches? To answer these questions, the paper relies on my empirical studies of officer deadly use of force and of collective violence in protests, as well as other scholars’ empirical work. The article first proposes a specified definition of situation and context. Using these concepts, it then proposes a causal specification of the situational approach through necessary, sufficient, and INUS conditions, as well as context factors as risk factors to violence. Third, in an outlook, it argues that this causal relationship between situation, context, and violence can be theoretically framed through an elaborated symbolic interactionism that integrates context into a situational approach. It also discusses the relevance of the debate for violence avoidance and for other research fields.
Article
It matters how people view the police—and that there is a substantial racial gap in these views. Research has primarily focused on police experiences to explain generally less-positive views among Black Americans. We recommend a subtle but vital shift in focus, seeking instead to explain the remarkably more favorable average views about the police among White Americans. Utilizing comparable data from two 2016 American National Election Studies surveys, we explore the role of contact with the police, politics, and three different dimensions of racial attitudes and views, finding views about the police among White Americans to be shaped in primary ways by concerns about Black Americans. These factors, and racial resentment in particular, explain a significant portion of the average difference in views of the police between Black and White Americans. We discuss the implications of this subtle shift in focus, particularly for work which sets positive views about the police as the goal.
Article
Purpose Limited research focuses on the challenges that exist at the intersection of race and dis/ability for Black men on the autism spectrum in encounters with law enforcement. An interdisciplinary approach is necessary to fully comprehend and mitigate the complex challenges. This conceptual article presents Critical Dis/ability (DisCrit) Theory, a framework usually applied to education, as a lens through which to think about encounters between Black autistic men and police officers. The article concludes with recommendations for collaboration between social scientists, police scholars, law enforcement and the public to improve the outcomes of police encounters involving Black men on the autism spectrum. Design/methodology/approach Using a literature review and analysis of current events, this conceptual article explores the intersections of race and disability, specifically neurodiversity, in policing using Critical Dis/ability (DisCrit) theory, and its predecessors, Intersectionality Theory and Critical Race Theory (CRT). Findings An interdisciplinary approach might be a potential solution to improve police encounters with Black autistic men. Expanding the theoretical frameworks utilized in scholarship about policing might allow for innovative approaches to examine current practices in law enforcement. Collaboration and critical dialogue may yield opportunities for further research and shifts in practice. Originality/value This conceptual article uses two incidents from recent events to highlight the need for increased scholarship around the intersections of autism and race, with a particular focus on Black autistic men. It advocates for the use of social science frameworks, namely DisCrit Theory, as a novel way to approach new research regarding race and dis/ability.
Article
Machine learning findings suggest Eurocentric (aka White/European) faces structurally resemble anger more than Afrocentric (aka Black/African) faces (e.g., Albohn, 2020; Zebrowitz et al., 2010); however, Afrocentric faces are typically associated with anger more so than Eurocentric faces (e.g., Hugenberg & Bodenhausen, 2003, 2004). Here, we further examine counter-stereotypic associations between Eurocentric faces and anger, and Afrocentric faces and fear. In Study 1, using a computer vision algorithm, we demonstrate that neutral European American faces structurally resemble anger more and fear less than do African American faces. In Study 2, we then found that anger- and fear-resembling facial appearance influences perceived racial prototypicality in this same counter-stereotypic manner. In Study 3, we likewise found that imagined European American versus African American faces were rated counter-stereotypically (i.e., more like anger than fear) on key emotion-related facial characteristics (i.e., size of eyes, size of mouth, overall angularity of features). Finally in Study 4, we again found counter-stereotypic differences, this time in processing fluency, such that angry Eurocentric versus Afrocentric faces and fearful Afrocentric versus Eurocentric faces were categorized more accurately and quickly. Only in Study 5, using race-ambiguous interior facial cues coupled with Afrocentric versus Eurocentric hairstyles and skin tone, did we find the stereotypical effects commonly reported in the literature. These findings are consistent with the conclusion that the “angry Black” association in face perception is socially constructed in that structural cues considered prototypical of African American appearance conflict with common race-emotion stereotypes.
Article
Full-text available
The present study examined homicides by police officers, testing threat, community violence, and organizational hypotheses. Using UCR, SHR, Census, and LEMAS data the study extends previous research by examining the relative impact of community violence, inequality and race, and organizational characteristics on the number of killings of felons by police officers in large US cities. The findings show that measures of racial threat and community violence were related to police killings. Measures of organizational policies were largely unrelated to the number of police killings. Overall the study extends research in the area, yet it also points to a more general need for research on the effects of organizational factors on police violence.
Article
Full-text available
Explanations of police coercion have been traditionally embedded within sociological, psychological, and organizational theoretical frameworks. Largely absent from the research are examinations exploring the role of neighborhood context on police use-of-force practices. Using data collected as part of a systematic social observation study of police in Indianapolis, Indiana, and St. Petersburg, Florida, this research examines the influence of neighborhood context on the level of force police exercise during police-suspect encounters using hierarchical linear modeling techniques. The authors found police officers are significantly more likely to use higher levels of force when suspects are encountered in disadvantaged neighborhoods and those with higher homicide rates, net of situational factors (e.g., suspect resistance) and officer-based determinants (e.g., age, education, and training). Also found is that the effect of the suspect's race is mediated by neighborhood context. The results reaffirm Smith's 1986 conclusion that police officers “act differently in different neighborhood contexts.
Article
Full-text available
Research shows that suspects' behavior influences police officers' decisions. Those who are disrespectful to the police are more likely to have that behavior reciprocated. To date, the factors influencing whether suspects will show deference remain largely unexamined. Guided by social interactionist theory and recent developments in urban sociology, we use systematic social observations and census data to investigate. We find that elevated levels of police force can induce suspect disrespect, but more subtle forms tend to have the opposite effect. The size of the audiences observing the encounter also influences whether suspects behave disrespectfully. Hypothesized links between disinhibiting factors such as intoxicants and disrespect are confirmed. Excluding traffic encounters, suspects in disadvantaged neighborhoods are less likely to show deference. This finding helps explain why officers encountering African Americans are the targets of disrespect.
Article
Full-text available
Tested the 2-process theory of detection, search, and attention presented by the current authors (1977) in a series of experiments. The studies (a) demonstrate the qualitative difference between 2 modes of information processing: automatic detection and controlled search; (b) trace the course of the learning of automatic detection, of categories, and of automatic-attention responses; and (c) show the dependence of automatic detection on attending responses and demonstrate how such responses interrupt controlled processing and interfere with the focusing of attention. The learning of categories is shown to improve controlled search performance. A general framework for human information processing is proposed. The framework emphasizes the roles of automatic and controlled processing. The theory is compared to and contrasted with extant models of search and attention. (31/2 p ref) (PsycINFO Database Record (c) 2012 APA, all rights reserved)
Article
The anterior cingulate cortex has been activated by color Stroop tasks, supporting the hypothesis that it is recruited to mediate response selection or allocate attentional resources when confronted with competing information-processing streams. The current study used the newly developed "Counting Stroop" to identify the mediating neural substrate of cognitive interference. The Counting Stroop, a Stroop variant allowing on-line response time measurements while obviating speech, was created because speaking produces head movements that can exceed those tolerated by functional magnetic resonance imaging (fMRI), preventing the collection of vital performance data. During this task, subjects report by button-press the number of words (1-4) on the screen, regardless of word meaning. Interference trials contain number words that are incongruent with the correct response (e.g., "two" written three times), while neutral trials contain single semantic category common animals (e.g., "bird"). Nine normal right-handed adult volunteers underwent fMRI while performing the Counting Stroop. Group fMRI data revealed significant (P < or = 10(-4) activity in the cognitive division of anterior cingulate cortex when contrasting the interference vs. neutral conditions. On-line performance data showed 1) longer reaction times for interference blocks than for neutral ones, and 2) decreasing reaction times with practice during interference trials (diminished interference effects), indicating that learning occurred. The performance data proved to be a useful guide in analyzing the image data. The relative difference in anterior cingulate activity between the interference and neutral conditions decreased as subjects learned the task. These findings have ramifications for attentional, cognitive interference, learning, and motor control mechanism theories.