University of Southern California Dissertations and Theses

Neural correlates of learning and reversal of approach versus withdraw responses to in- and out-group individuals

(USC Thesis Other)

Neural correlates of learning and reversal of approach versus withdraw responses to in- and out-group individuals

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content
i

NEURAL CORRELATES OF LEARNING AND REVERSAL OF APPROACH
VERSUS WITHDRAW RESPONSES TO IN- AND OUT-GROUP INDIVIDUALS

by

Jaclyn Mae Ronquillo

A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(PSYCHOLOGY)

August 2010

Copyright 2010 Jaclyn Mae Ronquillo

ii

Acknowledgments
I owe my deepest gratitude to those who have put in their valuable time, effort,
and support throughout this dissertation. First and foremost, I would like to thank my
chair, Dr. John Monterosso, for making this research possible. His guidance, clever
insight, and generosity are appreciated beyond what words can express. I would also like
to thank my other mentor, Dr. Brian Lickel, who has provided me with unwavering
support and encouragement throughout the years.
Furthermore, I would like to express my appreciation to my committee members,
Dr. Norman Miller, Dr. Jennifer Overbeck, and Dr. Stephen Read, for their time,
feedback, and for simply inspiring me by setting great examples. I am also extremely
thankful for the countless hours and hard work put into this dissertation project by Cody
Ashe-McNally, Candis Clark, Xochitl Cordova, Louise Cosand, Dr. Dara Ghahremani,
Lisa Giragosian, Shan Luo, Katie Reyes, Deborah Samson, Jodi Stone, and Dr. Andrew
Ward. Finally, to my husband, Patrick Adachi, for his understanding and patience
throughout my doctoral training.

iii

Table of Contents

Acknowledgments ii
List of Tables iv
List of Figures v
Abstract vi
Introduction 1
Method 17
Results 26
Discussion 45
Conclusion 55
References 57
Appendix: Instructions for RRLT Task 63

iv

List of Tables
Table 1. Mean response times and standard deviations for race 31
by evaluative association on the race reversal learning task

Table 2. Correlations among race reversal learning task and individual 34
difference measures

Table 3. Significant activations for whole brain analyses 37
Table 4. Correlations between individual differences measures and 44
functional regions of interest

v

List of Figures
Figure 1. Graphical representation of the design of race reversal 21
learning task

Figure 2. Accuracy during learning (across conditions) 27

Figure 3. Mean response time as a function of race and evaluative 29
association during learning

Figure 4. Mean response time as a function race and evaluative 30
association during reversal

Figure 5. Bias scores as a function of time during learning for race 32
reversal learning task

Figure 6. Race sensitive brain regions as indexed by significant BOLD 38
activity contrasting Black > White faces, collapsing across valence

Figure 7. Significant BOLD activity in occipital pole contrasting Black 39
approach to White approach

Figure 8. Significant BOLD activity in left operculum / insula contrasting 40
Black withdraw to White withdraw during learning

Figure 9. Significant BOLD activity in prefrontal cortex contrasting Black 41
withdraw to White withdraw during reversal

Figure 10. Signal change in anatomical regions of interest previously linked 43
to reversal learning

vi

Abstract
The present study examined the underlying neural mechanisms involved in the
learning and reversal of evaluative associations towards in- and out-group individuals.
While undergoing functional magnetic resonance imaging (fMRI), White participants
were asked to engage in a novel race reversal learning task (RRLT) in which they would
attempt to maximize points by either approaching or withdrawing (by means of a
joystick) from pictures of White and Black individuals, in correspondence to whether the
individual was a potential for positive points (gains) or negative points (losses). After six
acquisition trials, half of the stimuli would subsequently “reverse” such that good
individuals switched to bad, and bad individuals switched to good. Additionally,
individual difference measures assessing general executive function (i.e., Stroop Task),
general behavioral inhibition and reward activation, category-level implicit and explicit
race bias, and the motivation to control prejudiced responses were included to correlate
with brain activity and performance on the RRLT. Reaction times during the RRLT
indicated race bias (Race x Valence interaction) that was largely (though not completely)
driven by responses to initial presentations of faces. An interaction between Race and
Valence was not observed during the reversal phase of the task; however, a main effect of
Race emerged, such that participants were slower to respond to Black than White targets
that had reversed valence, regardless of whether they were positive or negative.
Neuroimaging data indicated increased activation in response to Black faces relative to
White faces in brain regions associated with visual processing (occipital cortex),
perceived salience (e.g., striatum), and with cognitive control (including the frontal

vii

operculum). Furthermore, we observed heightened activation in the insula for Black faces
relative to White faces when associated with a withdrawal response during learning. Post
hoc analyses indicated that endorsing category-level symbolic racism was associated with
increased controlled processing (frontal operculum) when withdrawing from Black faces
relative to White faces during learning. Taken together, the data suggest that race bias
attenuated rapidly in the present context with repeated exposure to Good and Bad
individuals. However, the overall slowness of responding to out-group faces during
reversal as well as persisting differences in brain response to Black vs. White targets
suggests that this individuation did not completely eliminate activation of the
superordinate race category.

1

Introduction
According to Allport (1954, p. 332), when individuals experience an inner
conflict, "people put the brakes upon their prejudices...[this] inner check operates
differently in different circumstances." Indeed, this notion has largely fueled the extant
research on the cognitive control of stereotypes and prejudice. More specifically,
Allport's (1954) contention highlights three important points that social psychologists
have long studied. First, the "inner conflict" suggests that there is an underlying
mechanism for the detection of race-related conflict. Second, the notion of "brakes"
implies that people have the ability to exert inhibitory control over their attitudes,
evaluations, and behavior towards members of other racial categories. And finally, that
self-regulatory processes (the "inner check") operate "differently in different
circumstances" implies that context and goals are an important influence on the self-
regulation of prejudice and stereotyping.
Though Allport (1954) argued that prejudice is controllable, more contemporary
research has distinguished between explicit (or controlled) versus implicit (or automatic)
racial bias (Greenwald & Banaji, 1995; Bargh, 1989; Bargh & Chartrand, 1999). Whereas
explicit race bias is under the conscious control of the perceiver, implicit race bias is
thought to be uncontrollable and possibly beyond the perceiver's threshold of awareness
(Greenwald & Banaji, 1995). Since Allport's time, normative standards of egalitarianism
have been on the rise, thus making it socially undesirable to express negative racial
attitudes. However, recent research has suggested that despite deliberate attempts to

2

appear non-prejudiced, it may still linger on an implicit level (e.g., Fazio, Jackson,
Dunton, & Williams, 1995).
In the past, some social psychologists have argued that due to their automatic
nature, implicit prejudice and stereotypes are inevitable (Bargh, 1999; Devine, 1989).
That is, automatic processes are spontaneously activated due to triggering conditions
regardless of the perceiver's thoughts, intentions, or goals (Dovidio & Fazio, 1992;
Bargh, 1989; 1997). For example, in a now classic study, Devine (1989) presented
subliminal prejudice-congruent cues to individuals with high- and low- self-reported
prejudice. Results indicated that regardless of level of prejudice, participants exhibited
automatic stereotyping when exposed to prejudice-congruent cues. Furthermore, when
given the opportunity to deliberately control prejudice, only participants low in self-
reported prejudice appeared non-prejudiced. These results were taken as evidence that a
perceiver's conscious efforts to control implicit racial biases are highly unsuccessful.
In stark contrast, much empirical evidence now attests to the malleability of
implicit race bias (for a review, see Blair, 2002). Specifically, a perceiver's goals as well
as contextual factors have been shown to successfully influence implicit race bias.
Sometimes (and only sometimes) optimal behavior requires people to respond to
outgroup members in ways contrary to implicit race bias. For example, engaging in a
cooperative manner with an outgroup member has been shown to attenuate implicit race
bias (e.g., Lowery, Hardin & Sinclair, 1998). Though some progress has been made, the
underlying mechanism through which this is accomplished is still left open for debate.
One hypothesis is that context that requires individuation is sufficient to remove implicit

3

race bias (Brewer, 1988; Fiske & Taylor, 1991; Wheeler & Fiske, 2005). Alternatively, it
may be that executive function plays a role through either overriding bias or through
maintenance of attention on aspects of the stimuli that are relevant to the perceiver's
goals.
Effortful Control and Attention as Moderators of Implicit Race Bias
Kawakami, Dovidio, Moll, Hermsen, and Russin (2000) demonstrated that
participants showed reduced implicit race bias (indexed by a sequential priming task)
after being trained extensively to negate stereotypic stimuli (i.e., a picture of a Black or
White individual paired with a stereotype) and affirm non-stereotypic stimuli (e.g., a
picture of Black or White individual paired with a non-stereotype). Similarly, Olson and
Fazio (2006) used an evaluative conditioning paradigm to reduce automatic racial
prejudice. Specifically, White participants were repeatedly exposed to conditioned
stimulus-unconditioned stimulus (CS-US) pairings of "Black-Good" and "White-Bad."
Results showed that the procedure was effective in attenuating implicit race bias scores,
and interestingly, the effect persisted for two days after the evaluative conditioning
session. Likewise, Kawakami, Phills, Steele, and Dovidio (2007) demonstrated that
training individuals to approach photographs of Blacks (by pulling a joystick) and avoid
photographs of Whites (by pushing a joystick) subsequently resulted in lower implicit
race bias (as indexed by an IAT) when compared to individuals trained to do the
opposite. While these findings suggest that race bias can be attenuated by repeated
stimulus pairings contrary to implicit race bias (i.e., Black/positive and White/negative),
it is still unclear from these data whether effortful learning of such "incongruent"

4

associations reduces implicit race bias via inhibiting or updating an existing association.
This is an important (but overlooked) nuance particularly for understanding the nature of
the malleability of implicit attitudes.
Although associative learning has been used to successfully attenuate implicit
race bias, the picture is not one-sided. Olsson, Ebert, Banaji, and Phelps (2005) exposed
Black and White individuals to photographs of unfamiliar Black and White faces paired
with either an aversive stimulus (i.e. shock) or alone. Results demonstrated that for both
Black and White subjects, conditioned fear responses to racial ingroup faces rapidly
extinguished when they were no longer paired with an aversive stimulus. However,
conditioned fear responses to racial outgroup faces reached extinction at a much slower
rate. This finding suggests that the process of extinguishing negative evaluative
associations towards outgroup members is not as rapidly accomplished relative to
ingroup members.
A common feature of existing measures of implicit race bias is that they assess
category-level associations that are activated automatically in response to a given
stimulus. However, some have argued that if a perceiver does not attend to stimulus-
associated cues that trigger implicit stereotypes and prejudice, then underlying category-
level representations are not automatically activated. According to Kunda and Spencer
(2003), stereotype activation can occur automatically through well-learned associations to
stereotypic cues or via goal-driven inference to satisfy perceiver goals (Sinclair & Kunda,
1999). Following this line of reasoning, researchers have investigated the role of a
perceiver's focus of attention on implicit race bias. For example, Gilbert and Hixon

5

(1991) demonstrated that White participants under high cognitive load exhibit less
stereotypic completions on a word fragment task when exposed to a member of another
racial category. Others have shown that manipulating the salience of a particular social
category influences implicit race bias. More specifically, Mitchell, Nosek, and Banaji
(2003) varied whether participants attended to race or gender when completing an IAT
and race Go/No go Association Task (GNAT). Results indicated that when gender was
salient, Black females were evaluated more positively than White males; on the other
hand, when race was salient, the reverse was true. Similarly, when race was made salient
compared to occupation, participants showed a bias in favor of White politicians relative
to Black athletes. However, when occupation was salient, the reverse was true.
Taken together, these studies suggest that task demands and goals that manipulate
a perceiver's focus of attention play a fundamental role in the activation of implicit race
bias. Early models of impression formation posited that the primary determinant of
category activation is the search requirements imposed by the perceiver’s goals and
motivations (Bruner, 1957). Likewise, Brewer and colleagues (e.g., Brewer, 1998; N.
Miller, Brewer, & Edwards, 1985) have argued that impression formation is a dual-
process in which category-level representations influence individual-level associations as
a function of category accessibility, contextual cues, and processing goals. For instance,
when seeing people as distinct individuals rather than as members belonging to a
particular racial category, category-based responding is de-emphasized (N. Miller et al.,
1985). Furthermore, past research has shown that whether memory for information is
organized at the individual or category level is also a function of perceiver goals, so that

6

mere salience of group categorization is not sufficient to elicit category-based processing
above and beyond the effects of visual cues alone. Furthermore, category-level
representations have been argued to be used more for minority social categories than
majority ones (Mullen, 1991).
In sum, behavioral evidence has well documented that both effortful control (e.g.,
explicit training of race-incongruent stimulus-response associations) and manipulating a
perceiver’s goals and processing strategies can moderate implicit race bias. Still, the
question remains of how this is accomplished. One possibility is that executive function
plays a role in inhibiting relatively automatic racial associations that may become salient.
Alternatively, it might be that task demands and perceiver motivation are enough to
eliminate implicit racial biases in a given context, thereby not utilizing cognitive control.
Perhaps both automatic and controlled processes interact to reduce implicit race bias. To
help address this question, social neuroscience has investigated brain processes
underlying these moderating conditions.
Neuroscientific Evidence for the Self-Regulation of Implicit Racial Attitudes
Social neuroscience researchers have used functional magnetic resonance imaging
(fMRI) to explore the neural correlates of race evaluation (for a review, see Eberhardt,
2005). Until recently, the primary focus has been on differential activity within the
amygdala, a sub-cortical structure that "reflects arousal triggered by fast unconscious
assessment of potential threat elicited by sensory, social, and emotional stimuli"
(Adolphs, Tranel, Damasio & Damasio, 1994). Numerous fMRI studies have
demonstrated greater amygdala response to Black faces than White faces by mostly

7

White perceivers (e.g., Phelps, O’Connor, Cunningham, Funayama, Gatenby, Gore, and
Banaji, 2000; Cunningham, Johnson, Raye, Gatenby, Gore, and Banaji, 2004; Lieberman
et al., 2005). Among Whites, Phelps et al. (2000) found differential blood oxygen-level-
dependent (BOLD) amygdala activity to photos of racial outgroup versus ingroup faces,
suggesting that amygdala responses to human faces are affected by the perceived race of
the stimulus and that of the subject. Moreover, it was demonstrated that the strength of
this activation was correlated with implicit, but not explicit, measures of race bias. In
sum, the evidence suggests that the amygdala is responsive to potential threat associated
with unfamiliar members of a racial outgroup. Interestingly, these effects occurred even
when face stimuli were presented so briefly that conscious awareness of the content of
the stimuli was not possible (Cunningham et al., 2004).
Although the majority of social neuroscience findings on race-related biases focus
on the amygdala, other researchers have turned their attention to identifying limbic and
cortical structures involved in the self-regulation of racial attitudes (for a review, see
Stanley, Phelps, & Banaji, 2008). According to Miller and Cohen (2001, 171),"The PFC
serves a specific function in cognitive control: the active maintenance of patterns of
activity that represent goals and the means to achieve them." Thus, at a broad level, it has
been hypothesized that the PFC is involved in top-down, goal-directed processes
recruited during race-related evaluation. For example, the Iterative Reprocessing (IR)
model has highlighted the neural systems involved in the activation and regulation of
attitudes and evaluation (Cunningham & Zelazo, 2007). According to the IR model,
perception of a stimulus (e.g. Black face) sends sensory information to the thalamus,

8

which in turn activates the amygdala. In this initial phase, the evaluation is automatic in
nature. The amygdala may send information to other limbic structures, such as the
hypothalamus, and/or it might also send information to the somatosensory cortex (e.g.,
anterior insula) and/or the orbitofrontal cortex (which is associated with affective
processing). At this point, the evaluation is still relatively automatic. However, over time,
cortical structures that are implicated in more controlled processes are recruited. For
example, the anterior cingulate cortex (ACC) may be activated upon detection of conflict.
The ventrolateral prefrontal cortex, including the frontal operculum, has been implicated
in response inhibition (Aron & Poldrack, 2006), which may be required to suppress fast-
initiated responses that are inconsistent with the actor’s goals. In addition, other areas of
the prefrontal cortex are activated during reflective processes, and particularly implicated
in maintenance of attention on goal-relevant stimuli (through working memory) while
simultaneously exerting inhibitory control.
As proposed by the IR model, evaluative processing occurs on a continuum from
automatic to controlled processing over time. Importantly, the key feature of this model is
that initial encounters with a given stimulus may evoke “quick and dirty” evaluations, but
that information gathered over time may serve to evoke more nuanced evaluations that
may also handle ambivalence (typically defined as “simultaneous activation of positive
and negative associations to the same stimulus”; Cunningham, Zelazo, Packer, & Van
Bavel, 2007). For example, imagine that a White individual who holds negative category-
level race bias encounters a Black individual for the first time in a non-threatening
context with the goal of getting to know whether, in general, the person holds good or

9

bad personal attributes. The IR model would predict that the initial encounter may call
upon pre-existing category-level attitudes, but that upon getting to know whether or not
the person really is good or bad, reflective processes override the initial response. If the
White individual learns that the Black individual is bad, then the initial response may
hold, but context and/or personal motivations may exert an influence on the initial
response. So, for instance, if the person has self-presentational concerns at the time, he or
she may try to avoid appearing prejudiced and would recruit controlled processes to
accomplish this goal. On the other hand, if the Black individual becomes known as
having positive attributes, then an ambivalent attitude possibly arises, in which case the
pre-potent evaluation is moved to the background and the new information is
foregrounded. According to Cunningham et al., “While ambivalent attitudes represent a
challenge when an individual is forced to direct behavior in a binary approach/avoid
fashion, they also allow for the representation of complexity and behavioral flexibility.”
This has important implications for the present study, and we will return to this model
later.
As preliminary evidence for the IR model, Cunningham et al. (2004)
demonstrated that White perceivers' amygdala activity to Black faces decreased as a
function of the length of exposure to the stimuli from subliminal (30 ms) to supraliminal
(525 ms). Furthermore, amygdala activity predicted an increase in the dlPFC and in the
ACC, areas involved in detection and regulation, in general. Together, these findings
provide early support for the IR model by showing an interaction between limbic (i.e.,
amygdala) and cortical structures (i.e., dlPFC and ACC) and by showing that negative

10

racial evaluation (as indexed by the amygdala) declines as a function of opportunity to
control implicit race biases. While “time” in this preliminary study was operationalized
as longer viewing time, the model allows for time to be interpreted as opportunity to
engage in control processing (e.g., repeated exposures).
Similarly, a study conducted by Richeson, Baird, Gordon, Heatherton, Wyland,
Trawalter, & Shelton (2003) provided converging evidence for the involvement of the
ACC and the dlPFC in the self-regulation of implicit racial attitudes. In their study,
participants' performance on tasks requiring executive function (i.e., Stroop task) was
assessed following an interracial interaction and later correlated with activity in the ACC
and the dlPFC during exposure to White and Black faces. They found that activity in the
dlPFC, and not the ACC, was correlated with performance on the Stroop task when
controlling for individuals' implicit racial attitudes. Based on these findings, the authors
suggest that the dlPFC is recruited during interracial interactions to control implicit racial
attitudes while the ACC is involved in detecting the need for such inhibitory control.
Amodio, Kubota, Harmon-Jones, and Devine (2006) further examined whether
alternative mechanisms underlie the regulation of implicit bias as a function of internal
versus external cues. Whereas internal cues reflect self-standards that are internalized for
self-regulation, external cues reflect normative pressures to respond without prejudice.
Internal versus external cues were manipulated by varying the ostensible privacy of
participants’ responses on a measure of race bias. Half of the participants were told their
responses would “remain confidential and [they] should not be concerned with external
pressures to appear non-prejudiced,” while the other half were informed that the

11

experimenter would be “paying attention to their responses to determine whether they
showed signs of racial prejudice.” The authors predicted that externally driven regulation
of implicit race bias should involve a distinct mechanism for processing external,
normative (e.g., social desirability), apart from a mechanism involved in processing
internal cues. Indeed, using electroencephalography (EEG), they found that brain activity
(as indexed by error-related negativity ERN) involved with conflict monitoring in the
dorsal ACC (dACC) was associated with response regulation according to internal cues,
while a response regulation according to external cues was associated with brain activity
involved with error perception in the rostral ACC (error-related positivity; P
e
), namely,
rACC. These findings suggest that the underlying neural mechanisms for the regulation
of personal versus normative cues are distinct from one another.
Finally, Beer et al. (2008) conducted a study in which participants completed the
Implicit Association Test (IAT) while undergoing fMRI. When contrasting incongruent
to congruent pairings in the IAT, Beer et al. (2008) found that the ACC, dlPFC,
operculum, and precuneus were active. Furthermore, when using the automatic
associations parameter (AC; Black/unpleasant, White/pleasant) from the quad-model of
implicit task performance (Conrey, Sherman, Gawronski, Hugenberg, & Groom, 2005) as
a predictor of brain activity, negative associations with outgroup members were related to
left lateral orbitofrontal cortex (OFC) activity, whereas positive associations with ingroup
members were related to amygdala, medial OFC, and right lateral OFC activity.
In conclusion, recent research has identified distinct neural areas involved in the
regulation of implicit racial biases. While the amygdala has been implicated in more

12

automatic processing of race-related stimuli (among other things), the ACC and areas of
the lateral prefrontal cortex (namely, dlPFC and OFC) have been involved in the self-
regulation of implicit racial bias. However, social neuroscience has looked thus far at
control over race bias at the category level. At present, there is no research examining
neural processes associated with the interplay between category level race-bias and
learning about individual members of different races.

Overview of the Present Research
Although it was previously thought that automatic attitudes were uncontrollable
(Bargh, 1999; Devine, 1989), there is ample evidence to suggest that they are malleable
(Blair, 2002). Returning to Allport's (1954) line of reasoning, the present study addresses
the nature of the "brakes" involved in the regulation of implicit racial attitudes. We are
specifically interested in the potential role played by control-processing in learning about
in- and out-group individuals that are associated with outcomes that are either consistent
or that are inconsistent with category-level race bias.

Individuation, updating, and inhibition
Consider the situation in which an individual with strong out-group bias
repeatedly encounters an out-group individual paired with positive associations. Based
on work cited above, we anticipate that markers of race bias would attenuate over time,
but the mechanism of the attenuation is unclear. One possibility is that individuation

13

directly renders category-level race bias inactive. Perhaps repeated exposure to an out-
group member paired with some hedonically significant outcome leads directly to
attenuation of the activation of category-level associations. On this view, it is not clear
that executive control processing plays any role in the attenuation of race-bias in response
to known individuals, since category level associations may not become activated when
context and experience provide a basis for individual-level association. Alternative to
this “pure individuation” account, the shift away from category level associations to
individual-level associations (as required to respond adaptively in some situations) may
at least in part depend on control processes. Perhaps individual out-group members
inherit category-level associations. If so, the updating demands will be greater when
targets are associated with outcomes incongruent with the actor’s bias. And apart from
any role control processes play in updating, if out-group members continue to activate
category-level associations, control processes may also be called upon to actively inhibit
category level race bias. That is, the learning that occurs when one repeatedly encounters
a positive out-group individual might include active inhibition of category bias.

The Race Reversal Learning Task (RRLT)
In order to examine the issues raised above, we utilized an adaptation of an
existing “reversal learning” task. Reversal learning tasks assess the ability to update
previously learned responses. In the initial phase of a typical reversal learning task,
participants are exposed to repeated pairings of one or more stimulus-response
associations. This acquisition is followed by a reversal phase in which the association of

14

some or all of the stimuli used in the acquisition phase require a different response than
what was initially learned. Performance in this reversal phase is dramatically affected by
damage to the lateral prefrontal cortex. Butter (1969) reported that after bilateral lesions
to the ventrolateral prefrontal cortex, macaque monkeys remained proficient at acquiring
novel stimulus-response pairings, but were markedly deficient at altering responses when
associations reversed. A similar tendency towards perseverative errors is observed
among humans with bilateral lesions to homologous brain areas. Using a variant of the
task that is well-suited to fMRI, researchers were able to distinguish neural correlates of
updating that occurs during reversal learning, from correlates of inhibitory control over
prior learned associations (inhibition of previously learned response; Ghahremani,
Monterosso, Jentsch, Bilder, & Poldrack, 2009). No study to our knowledge has ever
applied reversal-learning tasks to an intergroup context..
RRLT predictions
It is clear from prior findings that race category-level bias in attitudes is
conceptually orthogonal to the issue of the extent to which individual level
information/learning trumps race (makes category-level attitudes inactive). However, this
has received disproportionately little attention relative to category level attitudes, at least
in terms of implicit measurement. Our race reversal learning task (RRLT) is different
from existing implicit race bias measures in that task demands encourage individuation of
targets rather than categorization, thereby possibly making the effects of category-level
bias less salient and presumably less potent (Brewer, 1988).

15

Since the RRLT task, which we describe in detail below, requires the participant
to learn about individuals, it encourages individuation of targets. Nevertheless, while the
context is one of individuation, it has been argued that category-level race biases might
become active at first encounter with an individual. But with repeated exposure, these
biases may become attenuated by task goals and/or perceiver motivations (Brewer, 1988;
Kunda, Davies, Adams, & Spencer, 2002). Following from this line of reasoning, we
expect to see (at the behavioral level) differential response latencies when comparing first
trials to later trials, where bias scores (the difference score of stereotype-incongruent to
stereotype-congruent trials) would be higher in first trials compared to later trials. Our
rationale for this hypothesis is based on the idea that at first encounter, participants have
no evaluative information about the individual; thus, it is possible that category level bias
may become active (e.g., Bargh, 1999). However, on later trials, participants are given
evaluative information about each individual and task demands move attention away
from categories and toward individuals, presumably attenuating implicit race bias. But
critically, if race bias does attenuate over repeated exposures to an individual target, the
presence of reversal trials allows a strong test of the direct “individuation” interpretation
of the attenuation. If race bias disappears because sufficient exposure eliminates the
tendency for race-category associations to be activated by the target, bias would not be
expected to re-emerge during reversal, since additional encounters with an individual
should further individuation, even if specific associations with that individual must be
updated or inhibited.

16

Furthermore, the present study will allow us to correlate category-level race bias
with person-level associations by examining the relationship between the Race GNAT
(Nosek & Banaji, 2001) and the RRLT performance. Because task demands call for
categorization in the former and individuation in the latter, we would expect little
correlation between the GNAT and RRLT, provided that variance in RRLT relates
primarily to variance in processes involved in learning about individual members of in-
and out-groups. Alternatively, if category-level biases mediate individual level
associations, we would expect to see a correlation between the GNAT and the RRLT.
A nice feature of the RRLT is that it presents the opportunity to specifically
examine inhibitory control of previously learned and newly acquired associations in the
reversal phase of the task. Thus, not only would we be able to correlate category-level
bias during learning, we would also be able to observe how category-level bias plays a
role in reversal of pre-potent responses (e.g., Black-Bad) and newly acquired race
associations (e.g., Black-Good). At the neural level, we are broadly interested in whether
networks associated with controlled processing would be preferentially recruited when
responses were inconsistent with race bias in both learning and reversal. If category-level
associations are inherited by individual category members, then brain activity during
acquisition of bias-incongruent associations might resemble that observed during reversal
learning (increased metabolism in lateral OFC, rACC, DLPFC and rIFG; see Stanley et
al., 2008.) If repeated exposure results in individuation (attenuation and eventual
elimination of activation of category-level associations) then race-bias should not be
observed during reversal. If, however category-level bias remains active or potentially

17

active despite repeated exposure to individual associations, then brain response in regions
implicated in controlled processing might indicate less activity when the reversal is from
race-incongruent to race-congruent associations, relative to when the reversal is from
race-congruent to race-incongruent associations. Alternatively, vigilance related to race
and either internally or externally driven preoccupation with race bias might result in
general heightened control-processing of out-group members, regardless of whether
associations are congruent or incongruent with bias.
Method
Participants
Thirty-eight right-handed Caucasian participants (7 males, 28 females, mean age
= 21.77, SD = 4.89) with normal to corrected-to-normal vision participated in this study.
A subset of participants (N = 18) underwent functional magnetic resonance imaging
(fMRI). Prior to scanning, participants passed MRI safety screening and provided written
informed consent according to the USC Institutional Review Board. Behavioral and
neuroimaging data from two participants were discarded due to unacceptable accuracy on
the RRLT (near chance level). Neuroimaging data from two additional participants were
discarded due to excessive artifact (e.g., motion) in the acquired images.
Materials
Twenty-four photographs (12 Black faces, 12 White faces) were used in the
RRLT and 24 additional images were used in a post-scan memory test. All stimuli were
pre-tested by a separate sample of White individuals (N=10) for neutral expression,
stereotypicality, and approachability. Stimulus presentation and timing of all stimuli and

18

response events was achieved using Matlab (Mathworks, Natick, MA) and the
Psychtoolbox (www.psychtoolbox.org) on an Apple MacBook Pro running Mac OSX
(Apple Computers, Cupertino, CA).
Design and Procedure
The experimental design consisted of a 2 (Race of face: White, Black) x 2
(Evaluative association: Good, Bad) x 2 (Learning Phase: Acquisition, Reversal) within-
subjects design. Participants who completed only the behavioral portion of the study
reported to the lab individually, and were led by the experimenter into a testing room.
These participants completed the battery of measures in full counter-balanced order (with
the exception of the Face Recognition Task, which always followed the RRLT to
minimize time to keep faces in working memory). Each session lasted approximately one
hour, and participants were paid $10 in addition to bonus earnings from the RRLT and
given course credit (if applicable).
For the subset of participants who underwent fMRI scans, a training session was
required prior to scanning. In this session, participants practiced the RRLT with a
separate set of stimuli to become familiarized with the task. We made every effort to
ensure that participants understood the task and reached an acceptable threshold of
accuracy prior to scanning. Half of the participants completed the battery of individual
difference measures in this session prior to scanning, whereas the other half completed it
after the scan. Also, implicit and explicit measures were counterbalanced, with half
completing the GNAT before explicit race bias questionnaires, and half completing it
after. Explicit questionnaires were administered in random order. The entire experiment

19

(training and fMRI scan) lasted approximately 1.5 hours, and participants were paid $30
plus bonus RRLT earnings.
Race reversal learning task (RRLT). We modified the original reversal learning
task (RLT) used in Ghahremani et al. (2009), which consisted of a deterministic,
feedback-driven discrimination task. Ghahremani and colleagues developed this novel
measure so as to allow observation of brain responses to two major behavioral processes
involved in reversal learning; namely, inhibition of a pre-potent association and
formation of a new alternative association. The RLT includes extended acquisition
periods in which a stable pre-potent response could be established. Contrasting initial
stages of learning and relearning sheds light on the brain processes involved in relearning
of stimulus-response associations during reversal as distinguished from those generally
involved in learning.
In the RRLT, participants completed three separate runs of 96 trials each, for a
total of 288 trials. There were a total of 12 stimulus repetitions per face (6 learning, 6
reversal, or 12 repeated exposures for non-reversing stimuli). On initial trials, participants
were presented with either a Black or White face and asked to randomly guess whether it
is associated with an approach (i.e., pull joystick toward you) or withdraw (i.e., push
joystick away from you) behavior. The face was presented for 1 s, during which
participants were required to indicate their response. Instantaneously with response, the
face either grew larger with approach responses or smaller with withdrawal responses to
simulate the feeling of genuine approach and withdraw behaviors. Also simultaneous
with response, feedback about evaluative association appeared in the form of a colored

20

square frame around the stimulus for 1 s. A green frame indicated the corresponding face
was a "good guy," whereas a red frame indicated that the face was a "bad guy." If
participants did not respond within the 1 s response deadline, the phrase “no response
recorded” appeared above the image and no feedback about evaluative association was
provided.
Participants received one point if they approached a Black-Good or White-Good
target, and lost a point if they approached a Black-Bad or White-Bad target.
Withdrawing always resulted in no points. Since the task included an equal number of
Good and Bad targets, this ensured that if a participant were completely guessing, both
approaching and withdrawing would have the same expected value (i.e., 0). A running
total of points appeared beneath the stimulus during feedback presentation. Following
presentation of feedback, a fixation cross was displayed for a variable duration delay
(inter-stimulus interval, ISI) of 0.5 to 16 s (with a mean of 2s) before the start of the next
trial (see Figure 1).

21

Figure 1. Graphical representation of the design of race reversal learning task

Ghahremani et al. (2009) designed the original RLT so as to minimize the
potential for participants to predict reversals via rule-following in order to emphasize
stimulus-response associative learning. Similarly, in the RRLT, some stimuli never
reversed, leaving in question for participants whether a given encountered stimulus would
eventually require response reversal. As an incentive to maximize points, participants
were informed that they would be given $0.10 for each point earned. Participants were
encouraged to respond as quickly and as accurately as possible and were told that their
goal should be to accrue as many points as possible. Although reaction times did not
affect money earned, the participant was required to respond within 1 second of the
appearance of the target.

22

Behavioral Measures. To assess explicit race bias, the Symbolic Racism Scale,
Motivation to Control Prejudice, and feeling thermometers were administered. The
Symbolic Racism scale (Henry & Sears, 2002) was designed to have scaling properties to
help prevent response biases and general mindless response patterns. Example items
include, " It’s really a matter of some people not trying hard enough; if Blacks would
only try harder they could be just as well off as Whites", and "Over the past few years,
blacks have gotten less than they deserve (reverse-scored)." The Motivation to Control
Prejudiced Responses scale (Dunton & Fazio, 1997) assesses individual differences in the
extent to which people attempt to control the expression of prejudice, both for internal
and external reasons, separately. Feeling thermometers ask participants to gauge their
level of warmth on a scale from 0 to 100 towards European Americans and African
Americans. A difference score is taken as an index of ingroup bias.
Participants also completed a computerized task that measures race bias indirectly
(Race Go/NoGo task (GNAT); Nosek & Banaji, 2001). In the GNAT, strength of
association is assessed by the degree to which items belonging to the target category and
attribute (e.g., Black or White and good) can be discriminated from distracter items that
do not belong to those concepts. One condition requires simultaneous identification of
stimuli that represent the target category (Black or White) and an attribute (good). A
second condition requires simultaneous identification of stimuli that represent the same
target category and an alternative attribute (bad). The GNAT works by presenting target
(signal) and distracter (noise) stimuli for brief periods of time. The GNAT requires the
same response - “go” (press the space bar)—to items that belong to instances of a

23

category (i.e., Black or White) or a particular evaluative attribute (i.e., good or bad), both
of which, for this purpose, serve as the signal. No response “no-go” (do not press any
key) is called for when items appear that do not belong to the target category or attribute
(noise).
The Behavioral Approach and Behavioral Inhibition Scales (BIS/BAS; Carver &
White, 1994) were also be included to assess individual differences in inhibition and
approach toward reward. Though exploratory, it is reasonable to think that individuals
high in behavioral inhibition may also have higher brain activity in areas involved in
controlled processing (i.e., ACC, dLPFC) as well as report a higher motivation to control
prejudice.
We also wanted to ensure that the cross-race recognition effect (enhanced
uncertainty for Black targets independent of bias, known as the own-race bias or cross-
race effect; for a review, see Meissner & Brigham , 2001) was not an unexplained source
of variance in our study. Recently, Lebrecht, Pierce, Tarr, and Tanaka (2009) examined
the relationship between the own-race bias and implicit race bias. They found that
training White individuals to individuate Black faces results in reduced own-race bias and
reduced implicit race bias relative to White individuals trained to categorize Black faces
according to race. Furthermore, they found that for the individuation condition only, the
degree to which an individual's implicit race bias decreased was significantly correlated
with the degree of improvement that the individual showed in their ability to discriminate
between African American faces. Because the RRLT calls for individuation (rather than
categorization) of targets, we can be reasonably confident that the own-race bias would

24

be less of a concern in our study. However, to account for this possibility, we included a
face recognition task in the battery of individual difference measures.
Finally, the Stroop Color Interference task was included as a measure of general
interference control. The extant research is largely based on the use of implicit measures
that by nature assess category-level race bias. For example, in the IAT, participants are
asked to categorize a series of pictures or words into four categories (e.g., Black, White,
bad, good). An implicit negative racial attitude would be defined as faster response
latencies when stimuli are stereotype-congruent (Black + bad, White + good) versus
stereotype-incongruent (Black + good, White + bad). Another popular measure, the
Sequential (Evaluative) Priming Task (Banaji & Hardin, 1996; Fazio, Jackson, Dunton,
& Williams, 1995; Wittenbrink, Judd, & Park, 1997), asks participants to complete a
series of trials in which a prime (e.g., photograph of African-American or Caucasian
face) appears briefly followed by a target stimulus (e.g., positive or negative adjective).
Following each target, participants are required to make an evaluative judgment about the
stimulus that appeared (e.g., good versus bad). Again, response latencies are taken as an
index of implicit associations, such that faster responses to targets that are preceded by
stereotype-congruent primes versus stereotype-incongruent primes would be considered
implicit bias.
Although successful applications provide preliminary support for the validity of
these measures (for a review, see Fazio & Olson, 2003), their underlying mechanisms are
still controversial (Wittenbrink & Schwarz, 2007). Gawronski, Deutsch, LeBel, and
Peters (2008) argue that implicit measures do not reflect the activation of mental

25

associations, but instead provide only an indirect proxy. Specifically, they argue that the
impact of activated associations on task performance is mediated by mechanisms that are
specific to the task, and that outcomes of implicit race bias measures are driven by a
response interference (RI) component. For example, in Fazio et al.’s (1995) evaluative
priming task, the valence of the prime stimulus may trigger a pre-potent response
tendency that can be compatible or incompatible with the response tendency triggered by
the target word. If the prime stimulus and the target word share the same valence, the two
response tendencies have synergistic effects. If, however, the prime stimulus and the
target word have a different valence, the two response tendencies have antagonistic
effects. In this sense, priming effects are driven by two competing response tendencies,
thereby implying an RI mechanism. The evidence suggests that pre-potent response
tendencies elicited by activated associations depend on participants’ attention to
association-relevant stimulus features, which in turn can influence the reliability and the
construct validity of these measures. Thus, the Stroop task was included to compare
performance on a general response interference task to performance on the RRLT and to
help us understand the construct validity of the RRLT.
Image Acquisition. A Siemens 3T Magnetom MRI scanner was used for brain
image acquisition at the USC Dana and David Dornsife Cognitive Neuroimaging Center.
Consistent with the scanning protocol used by Ghahremani et al. (2009), we acquired 240
functional T2*-weighted echoplanar images (EPI) [slice thickness, 4 mm; 34 slices;
repetition time (TR), 2 s; echo time (TE), 30 ms; flip angle, 90°; matrix, 64 x 64; field of
view (FOV), 200 mm]. Two additional volumes were discarded at the beginning of each

26

run to allow for T1 equilibrium effects. In addition, a T2-weighted matched-bandwidth
high-resolution anatomical scan (same slice prescription as EPI) and magnetization-
prepared rapid-acquisition gradient echo (MPRAGE) scan was acquired for each
participant for registration purposes (TR, 2.3; TE, 2.1; FOV, 256; matrix, 192 x 192;
sagittal plane; slice thickness, 1 mm; 160 slices). The orientation for matched bandwidth
and EPI scans was oblique axial so as to maximize full brain coverage and to optimize
signal from ventromedial prefrontal regions. Furthermore, images were subjected to
prospective acquisition correction (PACE), which helps reduce head motion during data
acquisition.
Results
Behavioral Analyses. To assess difficulty of the RRLT, we examined accuracy
across the 6 repetitions prior to reversal (acquisition) and found that approximately 70%
of participants reached 100% accuracy by the third repetition (see Figure 2).

27

Furthermore, approximately 67% of participants reached full accuracy by the second trial
post-reversal. On average, fewer than 10% of the post-first trials were errors. Since
accuracy was near ceiling, behavioral analysis of RRLT focused on reaction times. All
analyses of reaction time during the task are based on correct trials only.
We were interested in whether mean response latencies of approach versus
withdrawal responses differed as a function of race, valence, and trial number. Mean
response times were submitted to a 2 (race: white, black) x 2 (evaluative association:
good, bad) x 6 (trial number) repeated measures ANOVA, and this analysis was
performed separately for learning and reversal. For learning, we found a significant main
effect of trial, F(1, 24) = 84.50, p < .001, η
2
= .78, with faster response times during later
trials. We also found a significant main effect of evaluative association, F(1, 24) = 41.90,
p < .001, η
2
= .64, where mean response latencies were significantly faster for
Figure 2. Accuracy during learning (across conditions)

28

approaching rewarding stimuli (i.e., good faces; M = 0.68, SE = .014) than withdrawing
from non-rewarding stimuli (i.e., bad faces; M = 0.71, SE =.013). Furthermore, the
expected race by evaluative association interaction was significant, F(1, 24) = 10.75, p =
.003, η
2
= .31. For good faces, mean response latencies were faster for White faces (M=
0.65, SE= .013) than Black faces (M = .68 SE = .016). However, the opposite pattern
emerged for bad faces, such that mean response latencies were faster for Black faces (M
= 0.70, SE= .013) relative to White faces (M = 0.72, SE = .015; see Figure 3). The three-
way interaction of Valence X Race X Trial (a test of the hypothesized attenuation of race
bias with repetition) was not significant, F(5,20) = 2.08, p = .11)

29

Figure 3. Mean response time as a function of race and evaluative association during
learning

Because only half the trials reversed, there were fewer data during reversal than
during acquisition. In addition to the implied relative reduction in statistical power
during reversal, this led to considerable occurrence during reversal of empty cells for
some subjects at one or more levels of a repeated measure ANOVA that included trial
number as a within subjects factor. Therefore, for analysis of performance during
reversal, mean response times were collapsed across trial number for each of the four
categories. We submitted the mean composites to a 2 (Race: Black, White) x 2 (Valence:
Good, Bad) repeated-measures ANOVA and found a main effect of race, F(1, 33) =

30

6.205, p < 0.02, η
2
= 0.16, such that mean response latencies were significantly faster for
White faces (M = 0.67, SE = 0.017) than Black faces (M = 0.70, SE = 0.015). We also
found a main effect of valence, F(1,33) = 8.79, p < .007, η
2
= 0.21, such that mean
response latencies were significantly faster for Bad faces (M = 0.67, SE = .016) than
Good faces (M = 0.70, SE = 0.17; see Figure 4). Table 1 presents the means and standard
deviations the four categories (Black-Good, Black-Bad, White-Good, White-Bad) for
both learning and reversal.
Figure 4. Mean response time as a function race and evaluative association during
reversal

31

Table 1. Means and standard deviations for race by evaluative association during
learning and reversal

Learning
Black/Good Black/Bad White/Good White/Bad
Mean 0.70 0.72 0.67 0.74
SD 0.08 0.07 0.07 0.07
Reversal
Black/Good Black/Bad White/Good White/Bad
Mean 0.72 0.69 0.70 0.66
SD 0.11 0.09 0.10 0.11

Although the repeated measures analysis did not indicate a significant interaction
between race bias and trial number (p = .11 for the 3-way interaction), further inspection
of the data suggested that there was a non-linear shift in race-bias in which bias
attenuated dramatically after the first response. Post-hoc analyses were carried out to test
whether bias declined as a function of time (as would be predicted by the IR model;
Cunningham et al., 2007). We created a bias score at each time point during learning and
reversal, separately. We created bias scores by taking the difference between stereotype-
incongruent associations (Black Good – White Bad) and stereotype-incongruent
associations (Black Bad – White Good). Using paired-samples t-tests, we compared bias
at the first trial to later trials collapsed and found that bias at trial 1 (M = 0.15, SD =
0.257) was significantly higher than later trials (see Figure 5).

32

To assess whether our expected Race x Valence interaction was driven by first
trials only, we submitted response times to 2 (race: white, black) x 2 (evaluative
association: good, bad) x 5 (trial number) repeated measures ANOVA, excluding first
trials. Presumably, if the interaction were being driven by first trials only, we would no
longer see evidence of this after dropping first trials. However, we still observed main
effects of valence F(1,35) = 49.47, p < .001, η
2
= .59 and trial F(4, 140) = 27.15, p <
.001, η
2
= .44, as well as a marginally significant Race x Valence interaction, F(1,35) =
4.07, p = .052, η
2
= .104. We observed the same pattern of results as when first trials
were included, where participants were faster for White faces (M = 0.64, SE = 0.012 )
than Black faces (M = 0.66, SE= 0.014 ) when valence was good, but faster for Black
Figure 5. Bias scores as a function of time during learning for race reversal learning
task

33

faces (M = .69, SE = .014) than White faces (M = .70, SE =.014) when valence was bad.
It is also worth noting that there was no indication of a Race X Valence X Trial
interaction in the analysis with first trial removed F(4,21) = .93, p = .45).
Race Go/No Go Association Task. To assess implicit category-level bias, mean
response times on the GNAT were submitted to a 2 (Race: Black, White) x 2 (Valence:
Good, Bad) repeated measures ANOVA (again, accuracy was near ceiling, thus analyses
are for correct trials only). No significant main effects emerged, but a significant
interaction was found, F (1, 35) = 54.98, p < .001, η
2
= .61. More specifically,
participants were faster to correctly classify White Faces and Good words when paired
together (M = 544. 46 SE = 6.91) than when Black Faces and Good words were paired
together (M = 576.26, SE = 7.95). Conversely, participants were faster to correctly
classify Black Faces and Bad words when paired together (M = 546.34, SE = 9.10)
relative to when White Faces and Bad words were paired together (M = 586.29, SE =
8.16).
Interestingly, although the nature of the RRLT and the GNAT are different in
terms of task demands (i.e., RRLT is individuation-based whereas GNAT is
categorization-based), the patterns of findings for the Race x Evaluative Association
interaction were quite similar (i.e., cross-over interactions). Presumably, if the RRLT and
GNAT are measuring the same underlying construct, they should be highly correlated.
On the other hand, it may be the case that differences in task demands are sufficient to
render almost no correlation between the two measures. To test these competing
hypotheses, we first created bias scores for each measure separately by taking the

34

difference between incongruent (Black Good + White Bad) and congruent (Black Bad +
White Good) trials. We observed no evidence of an association between these measures
r(33) = -0.009. Although null findings are not a statistical basis for concluding the
measures have no association, we calculated the 95% confidence interval for rho to be
between -0.35 and +0.34, which provides a statistical basis for concluding that any true
association between these measures is likely to be modest. This is not surprising,
however, given that test-retest reliability and correspondence between various implicit
measures has been demonstrated to be low to moderate (Cunningham, Preacher, &
Banaji, 2001). We currently do not have any measures of reliability concerning
the RRLT.
Explicit Questionnaires. Table 2 presents the correlations between all individual
difference measures and the RRLT.
Table 2. Correlations among race reversal learning task and individual
difference measures

35

Consistent with past research (e.g. Olson & Fazio, 2006), there was little correlation
between the GNAT and explicit questionnaires. Similarly, the RRLT did not correlate
with any explicit measures. In fact, the only significant correlations that came out of these
analyses were that general behavioral inhibition correlated moderately with both internal
motivation to control prejudiced responses r
s
(33) = 0.398 and restraint motivation to
control prejudiced responses r
s
(33) = .433, p < 0.05. Also, symbolic racism was
significantly and negatively correlated with external motivation to control prejudice,
r
s
(33) = -0.464, p < .01. Finally, the general response inhibition (i.e. Stroop) was
significantly and negatively correlated with external motivation to control prejudice,
r
s
(33) = -0.404, p < .05, such that higher one’s self-reported external motivation to
control prejudice, the lower one’s capacity for general response inhibition.
Stroop Task and Face Recognition. The Stroop task did not correlate with the
RRLT or GNAT, even at a relaxed threshold; thus, we did not include it as a covariate in
any analyses. To account for cross-race recognition deficits (for a review, see Meissner
and Brigham , 2001), we tested for differential accuracy in face recognition between
Black and White faces. We found no differences in face recognition for Black versus
White faces, χ
2
(1) = .735, p = .391. Thus, we can be reasonably certain that our findings
are independent of perceptual cross-race recognition deficits.
fMRI Analyses. Functional MRI image analysis was performed using the FSL
(4.1.5) toolbox from the Oxford Centre for fMRI of the Brain (www.fmrib.ox.ac.uk/fsl).
Specifically, image processing and analysis was conducted using FEAT (fMRI Expert
Analysis Tool) version 5.98. For preprocessing, the head movement that was not captured

36

by PACE was corrected by MCFLIRT (Motion Correction using FMRIB’s Linear Image
Registration Tool) (Jenkinson et al., 2002). Data were temporally filtered by a high-pass
filter with 100 s cutoff and spatially smoothed by a Gaussian kernel of full- width at half-
maximum of 5 mm. The preprocessed data were then submitted to a general linear model
(GLM) with blood oxygen level-dependent (BOLD) responses as our dependent variable.
Within-subject statistical analyses were performed in native image space and statistical
maps were normalized into Montreal Neurological Institute (MNI) space prior to higher-
level (group) analyses (Jenkinson and Smith, 2001).
Our primary neuroimaging analyses were conducted using whole brain analysis
with a cluster height threshold of Z > 2.3 and a corrected p-value threshold set to p <
0.05. Significant activations are reported in Table 3.

37

Table 3. Significant Activations for Whole Brain Analyses
Region of activation

MNI
Coordinates
Number
of Voxels
Z p
Black > White
R Frontal pole 12 54 30 11294 3.95 <.0001
L Occipital pole -6 -8 0 9146 5.06 <.0001
R Orbitofrontal cortex 34 36 -22 1612 3.79 <.0001
L Lateral Occipital Cortex -50 -60 44 1062 3.34 <.0001
L Precuneus -12 -52 26 496 3.26 .005
R Lateral Occipital Cortex 54 -60 30 471 3.18 .006
L Fusiform Cortex -40 -50 -20 360 3.49 .04
L Middle Temporal Gyrus -52 -10 -18 337 3.31 .05

Black Approach > White Approach
L Occipital Pole -16 -104 -2 292 3.44 .04

Black Withdraw > White Withdraw
L Operculum / L Insula -46 12 0 465 3.91 .008

Black Withdraw > White Withdraw
(Reversal)

L Putamen / L Pallidum -16 0 12 675 3.27 <.0001
R Thalamus 16 -12 14 337 3.11 .02
L Lateral Occipital Cortex -44 -70 44 321 3.00 .03
L Occipital Pole -20 -106 0 311 3.56 .04
R Frontal Pole / R Superior Frontal Gyrus 4 60 32 294 3.10 .05
R Middle Frontal Gyrus 38 18 48 293 3.02 .05

Following from past research (for a review, see Eberhardt, 2005), we first examined
whether trials in which Black faces were present elicited greater activation (as inferred
based on BOLD method) in any areas of the brain, irrespective of other factors (Black
faces > White faces contrast). We found that black faces recruited significantly greater
activity in the right frontal pole, left occipital pole, right orbitofrontal cortex, left and
right lateral occipital cortex, left precuneus, left fusiform cortex, and left middle temporal
gyrus (see Figure 6). No significant BOLD activity was observed for the reverse contrast

38

(White faces > Black faces).
Figure 6. Race sensitive brain regions as indexed by significant BOLD activity
contrasting Black > White faces, collapsing across valence

In an attempt to understand underlying brain processes that are engaged during the
RRLT, we examined brain activity for approach versus withdrawal responses as a
function of race, separately for learning and reversal. More specifically, we took the
difference in BOLD activity for Black approach versus White approach and Black
withdraw versus White withdraw, along with corresponding reverse contrasts. It is
important to note that we are contrasting BOLD activity for correct trials only; thus,
approach and withdraw are analogous to valence (i.e., Good and Bad) as reported in the
behavioral findings. We observed significant activity in Black approach versus White
approach during learning in the left occipital pole (see Figure 7).

39

Figure 7. Significant BOLD activity in occipital pole contrasting Black approach to
White approach

We also observed significant activity for Black withdraw versus White withdraw during
learning in the left frontal operculum and left insula (see Figure 8).

40

Figure 8. Significant BOLD activity in left operculum / insula contrasting Black
withdraw to White withdraw during learning

The only contrast that elicited significant activity during reversal (where statistical power
was generally low) was Black withdraw versus White withdraw; to clarify, this is where
both Black and White faces that were previously good turned bad. For this contrast, we
found significant greater activity recruited for Black faces in several brain regions,
including the left putamen, left pallidum, right thalamus, left lateral occiptal cortex, right
frontal pole (see Figure 9), right superior frontal gyrus, and right middle frontal gyrus.

41

Figure 9. Significant BOLD activity in prefrontal cortex contrasting Black withdraw
to White withdraw during Reversal

For exploratory purposes, we also used both anatomical and functional region-of-
interest (ROI) analyses to help understand how brain activity correlates with individual
difference measures. The former defines ROIs a priori by using brain regions previously
found to be implicated in similar processes while the latter approach forms ROIs on the
basis of functional clusters that surpass the statistical threshold for significance for a
given contrast at the whole brain level (Poldrack, 2007). It is important that statistical
tests applied to functional ROI’s be constructed so as to be independent from the basis
used to derive the functional ROI, or the result is erroneously low p-values (see Poldrack
& Mumford, 2009). In both approaches, β values were extracted from 6mm spheres
centered at MNI coordinates reported in previous literature (for anatomical ROIs;

42

Ghahremani et al., 2009; Beer et al., 2008; Cunningham et al., 2007; for a review, see
Stanley et al., 2008) or MNI coordinates for peak activations in functional clusters
deemed significant in our whole brain analyses (for functional ROIs).
An anatomical ROI analysis was carried out in which we extracted signal for every
subject from four 6 mm spheres centered around MNI coordinates in regions that had
previously been reported to be preferentially activated by the reversal phase relative to
the acquisition phase of a reversal learning task that used abstract shapes as target stimuli.
These ROI’s included the right dorsolateral prefrontal cortex, the right inferior frontal
gyrus, the right frontal operculum, and the dorsal anterior cingulate cortex (Gharemani et
al). The signal extracted for each condition, both in acquisition and reversal, is presented
in Figure 10. Although the means were in the anticipated direction, we did not observe a
significant difference based on reversal condition (F(11,1) = 2.37, p = .15). No
significant effects of Race (F(11,1) = .19, p = .67) or Race X Valence interaction (F(11,1)
= .01, p = .97) were observed.

43

Figure 10. Signal change in anatomical regions of interest previously linked to
reversal learning

We were also interested in whether amygdala activity was elicited during the
RRLT. Presumably, if individuation is sufficient to eliminate amygdala activity (Wheeler
& Fiske, 2005), we should not see significant amygdala activity in our data. We
submitted β values from our anatomical ROIs for both the right and left amygdala to
repeated measures ANOVA. We found a significant main effect of valence in the left
amygdala, F (1, 13) = 6.69, p = .023, η
2
= .340, such that more activity in the left
amygdala was observed for good faces (M = 9.09 SE = 21.20) than bad faces (M = -
16.63, SE = 16.01). No other significant effects were found.
Table 4 reports correlations between individual difference measures and MR
signal in functional ROIs.

44

Table 4. Correlations (Spearman’s rho) between individual differences measures
and functional regions of interest

Inconsistent with prior findings that have demonstrated more visual processing for
ingroup faces relative to outgroup faces (e.g., Golby et al., 2001), we found more visual
cortical activity for outgroup faces compared to ingroup faces collapsing across valence
(i.e., Black > White). Furthermore, we observed more visual processing when Black
faces were a source of reward (Black approach > White approach). To determine whether
differential visual processing for outgroup relative to ingroup faces was driven primarily
by “rewarding” outgroup faces, we used the functional ROI (occipital cortex) for (Black
approach > White approach) to extract β values during the Black withdraw > White
withdraw contrast. We found that signal in the visual cortex when contrasting Black
withdraw > White withdraw was trending toward significance, t(13) = 2.02, p = .064,
suggesting that participants exhibited heightened visual processing of outgroup faces
relative to ingroup faces. Furthermore, activity in the visual cortex when contrasting

45

Black approach to White approach was significantly and negatively correlated with
behavioral inhibition, r
s
(14) = -.70, p < .01, reward activation, r
s
(14) = -.69, p < .01,
and internal motivation to control prejudice, r
s
(14) = -.58, p < .05. That is, more general
inhibition and internal motivation to control prejudice was associated with higher visual
cortical processing of outgroup faces when they were a source of reward. Although it is
reasonable to think that higher tendency toward reward-seeking would be associated with
more visual processing for “rewarding” outgroup faces, we found the opposite pattern of
findings. There was also significant negative correlation between behavioral inhibition
and striatal activity when contrasting Black withdraw to White withdraw in reversal r
s

(14) = -.62, p < .05. Finally, and of particular interest, we found a significant and positive
correlation between symbolic racism and activity in the operculum, r
s
(14) = .71, p =
.004, an area that has been demonstrated to be implicated in controlled processing. Thus,
it appears that endorsing beliefs such as, “Blacks get more than they deserve” and
“Blacks no longer face prejudice or discrimination,” is associated with increased
controlled processing when withdrawing from Black faces relative to White faces during
learning.

Discussion
As this was a novel attempt to develop a race reversal learning task, we deemed it
necessary to first examine behavioral performance on the task and how it related to
existing implicit and explicit race bias measures. Interestingly, we found that the pattern
of responses were comparable across the RRLT (during learning) and GNAT, where

46

response latencies differed as a function of race and valence. That is, across both tasks,
we found that for good faces, response times were significantly faster for ingroup faces
than outgroup faces; whereas for bad faces, the reverse was true. This might lead one to
believe that performance on the RRLT could be construed as an alternative proxy for
implicit race bias; however, the fact that we found no correlation between bias scores on
the RRLT and GNAT raises a puzzling issue. On the one hand, it has been shown that
existing implicit race bias measures have low correspondence with each other, thus it is
certainly possible that the RRLT is tapping into similar underlying constructs as other
implicit race bias tasks but is not being detected by Pearson’s correlation due to low
reliability (Cunningham et al., 2001). On the other hand, the RRLT differs vastly from
other tasks in that previous measures call for categorization and the RRLT requires
individuation of targets, and thus may be tapping into an entirely different network of
associations. Unsurprisingly, the RRLT did not correlate with explicit race bias measures.
Thus, it remains for future work to further investigate the reliability and construct validity
of the RRLT, and its correspondence to existing race bias measures.
We also believed that the RRLT would allow us to examine how individuation of
in- and out-group members influences 1) bias over time (at the behavioral level), and 2)
neural areas implicated in automatic (e.g., amygdala, anterior insula) versus controlled
(e.g., ACC, dlPFC) processing of race evaluation. Following from the IR model
(Cunningham et al., 2007), we might expect to see automatic processes (e.g., amygdala,
insula) to be active during initial trials, but more controlled processes (e.g., dlPFC, ACC)
recruited on later trials. Alternatively, we might expect to see automatic and controlled

47

processes activated simultaneously, as others have argued that situational factors can
influence goal activation through both controlled, rule-based thinking and automatic,
associative processes together (Kunda & Spencer, 2003). On initial trials of the RRLT,
individuals were presented with in- and out-group faces without any evaluative
information. Thus, we expected that category-level bias would be highest during initial
trials, but would become attenuated over time. Indeed, we found that bias scores in first
trials were higher relative to later trials, suggesting that individuation may have
attenuated activation of category-level race bias. However, the fact that we still see some
evidence of race bias with first trials excluded (p = .052) and that there was no interaction
suggesting attenuation beyond what occurred after the first trial warrants caution in
drawing any conclusions regarding the temporal unfolding of category-level bias in this
task. It is also worth noting that while the race-bias pattern was not present in the
reversal condition, the overall slowness of response to Black relative to White faces
indicates that category level processing did occur. It does not appear that after several
exposures to individual-level associations, that race became irrelevant to performance.
This speculation is further supported by the fact that we did not find any significant
activation when activity for initial Black minus initial White faces (and vice versa) was
contrasted, suggesting that category-level bias might not have been active on initial trials.
However, prior evidence has suggested that individuation goals are sufficient to remove
activity in automatic processing (e.g., Wheeler & Fiske, 2005), and given that the RRLT
task requires individuation, this may provide clues as to why significant activations on
initial encounters were not observed. Furthermore, neither the left nor right amygdala

48

showed significant activation, which also converges with prior findings from Wheeler &
Fiske (2005). Though this idea is certainly possible, the more likely explanation is a lack
of statistical power to detect brain activity on initial trials (i.e., fewer data on initial trials
compared to later trials combined). Indeed, given that we observed brain activity in
several regions when contrasting Black faces to White faces (collapsing across valence)
and found no activity for the reverse contrast (White > Black), it is unequivocal that our
task activated the superordinate category of race despite its demand for individuation. We
also found significant activity in regions implicated in controlled processing, executive
function, and memory, including the right frontal pole and right orbitofrontal cortex
(OFC). There was also significant activity in the left lateral occipital cortex, which is
primarily involved with visual processing. One possibility that may help explain these
findings is that we did not attempt to conceal the fact that the RRLT dealt with race, thus
participants may have been self-conscious (which may explain activity within the
precuneus) with respect to egalitarian norms and in turn may have become motivated to
control race biases. These motivational processes would likely recruit neural regions
involved in more controlled processing (right frontal pole, right OFC) over implicit racial
biases (e.g., Cunningham et al., 2007; Beer et al., 2008; for a review, see Stanley et al.,
2008) and participants may have made a greater attempt to memorize associations for
outgroup faces relative to ingroup faces so as to seem unbiased. This hypothesis is further
bolstered by the fact that we observed significant activity in the visual cortex for Black
approach versus White approach. However, self-reported internal motivation to control
prejudice is significantly and negatively correlated with visual processing for our Black

49

approach versus White approach contrast (see Table 4). Thus, participants who report
being more motivated to control prejudiced responses for internal reasons show less
visual processing of outgroup faces, which is contrary to this idea. Interestingly, though
inconsistent with our findings, Golby, Gabrieli, Chiao, and Eberhardt (2001) found that
White participants showed heightened visual cortical activity while viewing race ingroup
faces relative to outgroup faces. However, they also found cross-race recognition deficits
in their sample, where participants had better recognition for own-race faces than other-
race faces, and this was significantly correlated with activity in the visual cortex. There
were no differences in recognition for ingroup versus outgroup faces in our data; thus, the
heightened visual cortical activity that we see for outgroup faces compared to ingroup
faces seems to be independent of any cross-race recognition deficits, which may help to
explain the inconsistencies between our data and prior findings (Golby et al., 2001; Van
Bavel, Packer, & Cunningham, 2008). Another important distinction is the nature of the
tasks; whereas greater processing of in-group faces was observed in a passive viewing
context, our findings of the reverse was obtained in a context where participants had to
decide whether to approach or withdraw from the face, with a monetary incentive for
being correct. If the negative correlation between motivation to control prejudice and the
observed signal difference in the visual cortex is robust (and one should be cautious since
the relationship would not survive any correction for multiple comparison) then a
reasonable inference would be that there is a general tendency in this context to process
out-group members more intently, but that concern over race bias may have led
participants to avoid exhibiting this vigilance.

50

To help understand the neural underpinnings of approach versus withdrawal
responses as a function of race in the RRLT, we examined brain activity for approach
versus withdrawal responses crossed with race during both learning and reversal
separately, as well as contrasted with each other. The only significant activations that
came out of these contrasts was Black approach versus White approach (which we have
discussed above), and Black withdraw versus White withdraw in both learning and
reversal. More specifically, we observed activity in the frontal operculum and left insula
for the Black withdraw >White withdraw contrast during learning, and these brain
regions have been shown to be implicated in controlled processing and negative emotions
such pain and disgust (e.g., Wright, He, Shapira, Goodman, & Liu, 2004), respectively.
Likewise, Beer et al. (2008) also found significant activity in the insula that was
associated with implicit negative associations with outgroup members (as indexed by the
Quad model). However, they also found activity in the same neural area for positive
associations with ingroup members and detection of appropriate behavior when race
targets are paired with an incongruent valence, so the functional role of the insula during
race related tasks remains somewhat unclear.
Prior findings have demonstrated that “unlearning” negative associations
occurs more rapidly for such associations towards ingroup members than outgroup
members (Olsson et al., 2005). In our data, during reversal, we found that mean response
times were significantly slower for outgroup faces than ingroup faces, irrespective of
valence. It is important to note, however, that Olsson and colleagues did not require
participants to reverse associations, but rather measured conditioned responses (i.e. skin

51

conductance) after a period of extinction. Thus, one possible reason for this discrepancy
could be the difference between reversing versus extinguishing a conditioned response.
Olsson et al (2005) compared negative conditioned responses to neutral conditioned
responses for outgroup relative to ingroup, but did not examine positive conditioned
responses. Given the nature of the RRLT, we are in a position to examine how reversal
of both positive and negative pre-potent responses play out as a function of race. As such,
our data seem to suggest that reversing a positive outgroup association takes more effort
than reversing a negative outgroup association, which may be in part due to perceiver’s
self-interest to accumulate gains.
Finally, we observed significant activation contrasting withdrawal responses for
Black versus White faces during reversal learning. Several areas within the basal ganglia
(implicated in motor responding and learning) including the left putamen, left pallidum,
and right thalamus showed significant race differentiated activity. These areas have been
recently linked to behavioral switching from automatic to controlled processes (Hikosaka
& Isoda, 2010), and are functionally connected to areas involved in executive function. In
our data, it seems that participants who had previously encoded particular outgroup faces
as “good” required more effort in reversing the previously learned “Black-approach”
response. Given that we saw heightened activity in the visual cortex for Black approach
versus White approach, it seems quite plausible that participants may have exerted more
effort in visually processing outgroup faces relative to ingroup faces particularly if they
were a source of reward (i.e., one point gain) during learning. That is, it is possible that
during learning, participants treated Black good faces as sources of gains, where

52

evaluative information regarding the target took precedence over category-level
associations per the perceiver’s motivation to accumulate as many points as possible.
Moreover, it quite possible that this “Black-Good” association became relatively
automatic, which might explain why we observed more controlled processing during
reversal (when Black good faces turned to Black bad faces) to override the previously
learned approach response. To explore this issue further, we pulled signal from each of
the four race by valence contrasts (i.e., Black approach > White approach (and vice
versa) and Black withdraw > White withdraw (and vice versa) and tested whether signal
was significantly different from zero. We found evidence of more visual processing for
not only the Black approach > White approach condition (which reached statistical
significance in whole brain analyses), but also in Black withdraw > White withdraw,
suggesting that at least in our sample, individuals were visually processing outgroup
faces more than ingroup faces (but see Golby et al., 2001; Van Bavel et al., 2008). Thus,
while we know from our neuroimaging findings that the RRLT is sensitive to race,
further work is needed to bolster our preliminary findings and to make more nuanced
distinctions about the neural processes underlying the race bias effects we are seeing at
the behavioral level.
In an attempt to discover more fine-grained distinctions concerning the
malleability of implicit race bias, we examined brain activity in a priori anatomical
regions that are associated with updating pre-potent associations that is independent of
inhibition. We did not find any significant activation contrasting reversal to learning, both
collapsing across race and valence and contrasting reversal to learning separately as a

53

function of race and valence. The means were in the predicted direction, however, but
given our small sample size, we may lack sufficient statistical power to detect a
meaningful difference. Thus, future work might include more post-reversal trials (and
perhaps fewer acquisition trials) to increase statistical power.
Future Directions
While we have discussed our findings in terms of ingroup versus outgroup bias,
future research might include Black participants to truly assess whether this is an
ingroup/outgroup phenomenon, or if our findings are specific to race. Lieberman et al
(2005) found that both Black and White individuals showed heightened activation in the
amygdala in response to Black (relative to White) faces, suggesting that automatic
amygdala response to Black faces may reflect cultural knowledge rather than a race
response per se. It would be interesting to see whether Black participants show the
opposite pattern of results to what we found in our White sample. That is, if Black
participants demonstrate faster response latencies to Black good faces relative to White
good faces and vice versa for bad faces, then the RRLT may be truly assessing ingroup
versus outgroup bias. It would also be beneficial to know whether bias on the RRLT is
driven by ingroup favoritism or outgroup derogation; a lack of a control condition in the
present data prevents us from making any claims about this issue. To address this, future
versions of the RRLT should include a control condition (perhaps, “move the joystick
away from/toward you” without presentation of a face).
In our data, we hoped to be able to compare learning and reversal trials to see if
brain regions previously associated with updating (apart from inhibiting) were active

54

(Ghahremani et al., 2009), but we did not observe significant activation here. However, it
is possible that we did not have sufficient power to detect differences that might
otherwise be existent due to such a small amount of post-reversal data. Alternatively, it
might also be the case that we did not observe such differences in our data due to the
nature of the RRLT. More specifically, if category-level biases are active during learning,
it is quite possible that a “reversal” component exists during the learning phase, when
individuals are required to approach Black-good targets or avoid White-bad targets. Both
stimulus-response associations are stereotype-inconsistent and thus may require more
controlled processing. As such, it may be that this “reversal” component during learning
is mirroring true post-reversal activity, thus eliminating any differences between the two
phases of the RRLT. Indeed, comparing mean signal change in our a priori anatomical
ROIs following from Ghahremani et al. (2009) demonstrated no differences between in
these regions between learning and reversal. Still, future research is warranted to
disentangle whether malleability of implicit race bias is due to updating versus inhibiting
(or, a combination of both) previously learned associations.
Cunningham et al. (2004) demonstrated that the amygdala is sensitive to Black
(versus White) faces when faces were subliminally presented, but not when faces were
presented supraliminally. We found that the prefrontal cortex, visual cortex, and
precuneus were active during presentation of Black (relative to White) faces, and the lack
of any significant amygdala activity converges with Cunningham et al. (2004) given that
faces in the RRLT were presented supraliminally. It would be interesting, however, to
develop a variant of the RRLT in which Black and White faces are flashed subliminally,

55

shortly followed by evaluative association (e.g., green colored square or red colored
square) to which participants would need to approach or withdraw. Similar to evaluative
priming tasks, one might expect category level bias to be indexed such that participants
are faster to respond to red squares (“bad”) when preceded by a Black face versus a
White face and vice versa for green squares (“good”). Due to the fact that participants
would not consciously perceive Black and White faces, this may eliminate any self-
presentational concerns and thus would put us in a position to examine bottom-up (as
opposed to top-down) controlled processing and how it interacts with automatic
processes in attenuation of implicit race bias.

Conclusion
The novel findings presented in this study concern the development of a reversal
learning task in which individuals approach or withdraw from in- and outgroup faces
based on evaluative information about the target. While it was demonstrated that the
pattern of performance on the RRLT is similar to existing implicit race bias measures
(i.e., GNAT), it remains for future work to determine 1) the reliability and construct
validity of the RRLT and 2) its correspondence with other individual difference
measures. We were also interested in using the RRLT to assess how category level race
biases influence approach and withdraw responses to individuals paired with
individuating evaluative information. Our behavioral findings support the idea that
category-based race bias seems to decline as a function of time (Kunda, Davies, Adams,
& Spencer, 2002; Cunningham et al., 2007), but this is still left open for interpretation

56

given that we still see bias (faster for good White faces compared to good Black faces,
and slower for bad Black faces compared to bad White faces) overall (collapsing across
trials during learning). And, finally, our neuroimaging data provide clues as to what
underlying neural processes are involved during the RRLT. We observed evidence of
heightened vigilance to out-group members in the context of the task, as evidenced both
by activity in the visual cortex as well as in brain regions relevant to emotion and reward.
Follow-up work with modifications directed at increasing statistical power is necessary in
order to draw more definitive conclusions. Clearly, the RRLT is sensitive to race, and the
neuroimaging data did partially support the hypothesis that out-group individuals elicited
more controlled processing. However we did not observe strong evidence that this
controlled processing was either 1) related to valence (e.g., greater for bias-incongruent
pairings) or 2) related to bias during the RRLT. It remains for a larger neuroimaging
study to determine whether recruitment of activity in brain areas associated with
cognitive control facilitates performance on the RRLT, or whether in this highly
individuating task, it is irrelevant or perhaps even detrimental to associative learning
across trials.

57

References
Adolphs, R., Tranel, D., Damasio, H., & Damasio, A. (1994). Impaired recognition of
emotion in facial expressions following bilateral damage to the human amygdala.
Nature, 372, 669-672.

Allport, G.W. (1954). The nature of prejudice. New York: Addison-Wesley.

Amodio, D. M., Kubota, J. T., Harmon-Jones, E., & Devine, P. G. (2006). Alternative
mechanisms for regulating racial responses according to internal vs. external cues.
Social Cognitive and Affective Neuroscience, 1, 26-36.

Aron, A.R & Poldrack, R.A (2006).Cortical and subcortical contributions to
stop signal response inhibition: role of the subthalamic nucleus. Journal of
Neuroscience, 26 2424-2433.

Banaji, M.R. & Hardin, C.D. (1996). Automatic stereotyping. Psychological Science,
7(3), 136-141.

Bargh. J.A. (1997). The automaticity of everyday life. In R.S. Wyer (Ed.), Advances in
social cognition (Vol. 10, pp. 1-61). Mahwah, NJ: Lawrence Erlbaum
Associates, Inc.

Bargh, J. A. (1999). The cognitive monster: The case against controllability of automatic
stereotype effects. In S. Chaiken & Y. Trope (Eds.), Dual process theories in
social psychology. New York: Guilford.

Bargh, J.A. (1989). Conditional automaticity: Varieties of automatic influence in social
perception and cognition. In J.S. Uleman & J.A. Bargh (Eds.), Unintended
thought, New York: Guilford.

Bargh, J.A., & Chartrand, T. L. (1999). The unbearable automaticity of being. American
Psychologist, 54, 462-479.

Beer, J.S., Stallen, M., Lombardo, M.V., Gonsalkorale, K., Cunningham, W.A., &
Sherman, J.W. (2008). The quadruple process model approach to examining the
neural underpinnings of prejudice. NeuroImage, 43, 775-783.

Brewer, M. B. (1988). A dual process model of impression formation. In Srull, T.K. &
Wyer, R.S. (eds.), Advances in Social Cognition, Vol. 1, 1-36.

Brewer, M.B., Weber, J.G. & Carini, B. (1995). Person memory in intergroup contexts:
Categorization versus individuation. Journal of Personality and Social Psychology,
69(1), 29-40.

58

Bruner, J.S. (1957). On perceptual readiness. Psychological Review, 64, 123-152.

Blair, I.V. (2002). The malleability of automatic stereotypes and prejudice. Personality
and Social Psychology Review, 6(3), 242-261.

Butter, C. M. (1969).Perseveration and extinction in discrimination reversal tasks
following selective frontal ablations in Macaca mulatta. Physiology & Behavior,
4, 163–171

Carver, C. S., & White, T. L. (1994). Behavioral inhibition, behavioral activation, and
affective responses to impending reward and punishment: The BIS/BAS scales.
Journal of Personality and Social Psychology, 67, 23, 319-333.

Conrey, F. R., Sherman, J. W., Gawronski, B., Hugenberg, K., & Groom, C. (2005). Separating
multiple processes in implicit social cognition: The Quad model of implicit task
performance. Journal of Personality and Social Psychology, 89, 469–487.

Crosby, F. Bromley, S. & Saxe, L. (1980). Recent unobtrusive studies of Black and
White discrimination and prejudice: a literature review. Psychological Bulletin.
87, 546-63.

Cunningham, W. A., Preacher, K. J., & Banaji, M. R. (2001). Implicit attitude measures:
Consistency, stability, and convergent validity. Psychological Science, 12, 163-
170.

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(12), 806-813.

Cunningham, W., & Zelazo, P. D. (2007). Attitudes and evaluations: A social cognitive
neuroscience perspective. Trends in Cognitive Sciences.

Cunningham, W.A., Zalazo, P.D., Packer, D.J., & Van Bavel, J.J. (2007). The Iterative
Reprocessing Model: A multilevel framework for attitudes and evaluation. Social
Cognition, 25(5), 736-760.

Davidson, R. J., Ekman, P., Saron, C. D., & Senulis, J. A. (1990). Approach/withdrawal
and cerebral asymmetry: Emotional expression and brain physiology: I. Journal of
Personality & Social Psychology, 58, 330–341.

Devine, P.G. (1989). Stereotypes and prejudice: Their automatic and controlled
components. Journal of Personality and Social Psychology, 22, 22-37.

59

Devine, P. G., Plant, E. A., Amodio, D. M., Harmon-Jones, E. & Vance, S. L. (2002).
The regulation of explicit and implicit race bias: The role of motivations to
respond without prejudice. Journal of Personality and Social Psychology, 82,
835-848.

Dovidio, J.F. & Fazio. R.H. (1992). New technologies for the direct and indirect
assessment of attitudes. In J. Tanur (Ed.), Questions about questions: Inquiries
into the cognitive bases of surveys (pp. 204-237). New York: Russell Sage
Foundation.

Dunton, B.C. & Fazio, R.H. (1997). An individual difference measure of motivation to
control prejudiced reactions. Personality and Social Psychology Bulletin, 23(3),
316-326.

Eberhardt, J.L.(2005). Imaging Race. American Psychologist, 60(2), 181-190.

Fazio, R.H., Jackson, J.R., Dunton, B.C., & Williams, C.J. (1995). Variability in
automatic activation as an unobtrusive measure of racial attitudes: A bona fide
pipeline? Journal of Personality and Social Psychology, 69, 1013-1027.

Fazio, R.H. & Olson, M.A. (2003). Implicit measures in social cognition research: Their
meaning and use. Annual Review of Psychology, 54, 297-327.

Fazio, R.H. & Olson, M.A. (2006). Reducing automatically activated racial prejudice
through implicit evaluative conditioning. Personality and Social Psychology
Bulletin,32(4), 421-433
Fiske, S. & Taylor, S. E. (1991). Social cognition (2nd ed.). New York: McGraw-Hill.

Frey S. & Petrides, M. (2000). Orbitofrontal cortex: A key prefrontal region for encoding
information, Proceedings of the National Academy of Sciences, 97(15), 8723-8727.

Gawronski, B., Deutsch, R., LeBel, E. P., & Peters, K. R. (2008). Response interference
as a mechanism underlying implicit measures: Some traps and gaps in the
assessment of mental associations with experimental paradigms. European
Journal of Psychological Assessment, 24, 218-225.

Ghahremani, D. G., Monterosso, J., Jentsch, J.D., Bilder, R.M, & Poldrack, R.A. (2009).
Cerebral Cortex, doi:10.1093/cercor/bhp247.

Gilbert, D.T. & Hixon, J.G. (1991). The trouble of thinking: Activation and application of
stereotypic beliefs. Journal of Personality and Social Psychology, 60(4), 509-517.

Golby, A.J., Gabrieli, J.D.E, Chiao, J.Y., Eberhardt, J.L. (2001). Differential responses in the
fusiform region to same-race and other-race faces. Nature Neuroscience, 4, 845-850.

60

Greenwald, A. G., & Banaji, M. R. (1995). Implicit social cognition: Attitudes, self-
esteem, and stereotypes. Psychological Review, 102, 4-27.

Greenwald, A.G., McGhee, J.L. & Schwartz, J.L. (1998). Measuring individual
differences in social cognition: The implicit association test. Journal of
Personality and Social Psychology, 74, 1464-1480.

Henry, P. J., & Sears, D. O. (2002). The symbolic racism 2000 scale. Political
Psychology, 23, 253-283.

Hikosaka, O, Isoja, M. (2010). Switching from automatic to controlled behavior: cortico-
basal ganglia mechanisms. Trends in Cognitive Science, 14(4), 154-61.

Hornak, J., O’Doherty, J., Bramham, J., Rolls, E.T., Morris, R.G., Bullock, P.R. &
Polkey, C.E. (2004). Reward-related reversal learning after surgical excisions in
orbito-frontal or dorsolateral prefrontal cortex in humans. Journal of Cognitive
Neuroscience, 16, 463-478.

Jenkinson, M. & Smith, S. (2001). A global optimization method for robust affine
registration of brain images. Medical Image Analysis, 5(2), 143-156.

Jenkinson, M., Bannister, P.R., Brady, J.M., and Smith, S.M. (2002). Improved
optimisation for the robust and accurate linear registration and motion correction
of brain images. NeuroImage, 17(2), 825-841.

Kawakami, K., Phills, C. E., Steele, J. R., & Dovidio, J. F. (2007). (Close) Distance
makes the heart grow fonder: Improving implicit racial attitudes and interracial
interactions through approach behaviors. Journal of Personality and Social
Psychology, 92, 957-971.

Kawakami, K., Dovidio, J.F., Moll, J., Hermsen, S., & Russin, A. (2000). Just say no (to
stereotyping): Effects of training in the negation of stereotypic associations on
stereotype activation. Journal of Personality and Social Psychology. 78, 871-888.

Kunda, Z., Davies, P.G., Adams, B.D., Spencer, S.J. (2002). The dynamic time course of
stereotype activation: Activation, dissipation, and resurrection. Journal of
Personality and Social Psychology, 82(3), 283-299.

Kunda, Z., & Spencer, S. J. (2003). When do stereotypes come to mind and when do they
color judgment? A goal-based theoretical framework for stereotype activation and
application. Psychological Bulletin, 129, 522-544.

Kunda, Z., & Sherman-Williams, B. (1993). Stereotypes and the construal of
individuating information. Personality and Social Psychology Bulletin, 19, 90-99.

61

Kunda, Z., & Thagard, P. (1996). Forming impressions from stereotypes, traits, and
behaviors: A parallel-constraint-satisfaction theory. Psychological Review, 103,
284-308.

Lebrecht S, Pierce LJ, Tarr MJ, Tanaka JW (2009) Perceptual Other-Race Training
Reduces Implicit Racial Bias. PLoS ONE 4(1): e4215.
doi:10.1371/journal.pone.0004215

Lieberman, M.D., Hariri, A., Jarcho, J.M., Eisenberger, N.I., & Bookheimer, S.Y. (2005).
An fMRI investigation of race-related amygdala activity in African-American and
Caucasian-American individuals. Nature Neuroscience, 8(6), 720-722.

McConahay, J.B. (1986). Modern racism, ambivalence, and the modern racism scale. In
J.F. Dovidio and S.L. Gaertner (eds), Prejudice, Discrimination, and Racism,
New York: Academic Press.

Meissner CA, Brigham JC (2001) Thirty years of investigating the own-race bias in
memory for faces. Psychology, Public Policy, and Law 7: 3–35.

Miller, E., and Cohen, J. (2001). An integrative theory of prefrontal cortex function.
Annual Review of Neuroscience, 24, 167-202.

Miller, N., Brewer, M.B. and Edwards, K. (1985). Cooperative interaction in
desegregated settings: A laboratory analogue. Journal of Social Issues, 41, 63–
79.
Mitchell, J.P., Nosek, B.A., & Banaji, M.R. (2001). Contextual variations in implicit evaluation.
Journal of Experimental Psychology: General, 132(3) 455-469.

Mullen, B. (1991). Group composition, salience, and cognitive representations: The
phenomenology of being in a group. Journal of Experimental Social Psychology, 27,
297-323.

Nosek, B.A. & Banaji, M.R. (2001). The go/no go association task. Social Cognition, 19(6), 625-
664.

Olson, M.A. & Fazio, R.H. (2006). Reducing automatically activated racial prejudice through
implicit evaluative conditioning. Personality and Social Psychology Bulletin, 34(4), 421-
433.

Olsson, A., Ebert, J. P., Banaji, M. R., & Phelps, E. A. (2005). The Role of Social Groups
in the Persistence of Learned Fear. Science, 309:785-787.

62

Phelps, E.A., O’Connor, K.J., Cunningham, W.A., Funayama, E.S., Gatenby, J.C., Gore,
J.C., & Banaji, M.R. (2000). Performance on indirect measures of race evaluation
predicts amygdala activation. Journal of Cognitive Neuroscience, 12(5), 729-738.

Plant, E. A. & Devine, P. G. (1998). Internal and external motivation to respond without
prejudice. Journal of Personality and Social Psychology, 75, 811-832.

Poldrack. R.A. (2007). Region of interest analysis for fMRI. Social, Cognitive, and
Affective Neuroscience, 2, 67-70.

Poldrack, R.A. & Mumford, J.A. (2009). Independence in ROI analysis: where is the
voodoo? Social, Cognitive, and Affective Neuroscience,4(2), 208-213.

Richeson, J.A., Baird, A.A., Gordon, H.L., Heatherton, T.F., Wyland, C.L., Trawalter, S.,
& Shelton, J.N. (2003). An fMRI investigation of the impact of interracial
contact on executive function. Nature Neuroscience, 6, 1323-1328.

Stanley, D. Phelps, E. & Banaji, M. (2008). The neural basis of implicit attitudes. Current
Directions in Psychological Science, 17(2), 164-170.

Talairach, J. & Tournoux, P. (1988). Co-planar stereotaxic atlas of the human brain.
New York: Thieme.

Van Bavel, J.J., Packer, D.J., Cunningham, W.A. (2008). The neural substrates of
ingroup bias: An fMRI investigation. Psychological Science, 19, 1130-1138.

Watson, D., Clark, L. A., & Tellegen, A. (1988). Development and validation of brief
measures of positive and negative affect: The PANAS scales. Journal of
Personality and Social Psychology, 54, 1063-1070.
Wheeler, M.E., & Fiske, S.T. (2005). Controlling racial prejudice: Social-cognitive goals
affect amygdala and stereotype activation. Psychological Science, 16, 56-63.

Wittenbrink, B., & Schwarz, N. (Eds.). (2007). Implicit measures of attitudes. New York:
Guilford Press.

Wittenbrink, B., Judd, C.M., & Park, B. (1997). Evidence for racial prejudice at the implicit level
and its relationship with questionnaire measures. Journal of Personality and Social
Psychology, 72(2), 262-274.

Wright P, He G, Shapira NA, Goodman WK, Liu Y (2004). Disgust and the insula: fMRI
responses to pictures of mutilation and contamination. Neuroreport (15): 2347–
51.

63

APPENDIX: INSTRUCTIONS FOR RRLT TASK
In this game, you will be presented with faces on the screen of good and bad guys. For
each face, you will decide whether to approach or avoid the person. You should
try to approach the good guys and avoid the bad guys.
To approach someone, pull the joystick toward you. Think of it as bringing the
person to you. The face will get bigger.
• If you approach a good guy, you get a point
• If you approach a bad guy you lose a point.
To avoid someone, push the joystick away from you. Think of it as pushing the
person away from you. The face will get smaller. You cannot gain or lose a point when
you push someone away. But keep in mind that pushing away a good guy is a lost
opportunity to gain a point. Since 50% of the faces are good and 50% bad, you will
likely do the same whether you approach or avoid someone you have no idea about.
So there is no advantage to playing it safe.
There is no way to know whether a face is good or bad until you get experience
with the face. You will find out whether someone is a good or bad guy by the color of
the frame that appears around the face after you respond.
A green frame means the person is a good guy
A red frame means the person is a bad guy.
Please note that the color has nothing to do with whether you are right or wrong.
It simply tells you whether the person is good (green) or bad (red).
Please respond as quickly as you can on each trial; you have just 1 second to respond
on each trial – a message will let you know if you are too slow. If you are too slow on
several trials you can lose points.
You will be paid $.10 per point at the end of the scan. If you have less than 0 points
you will not lose money.
You will only have to remember 4 faces at a time. You'll notice that, some, but not
all of the faces will switch part of the way through from good to bad or from bad
to good. Try to adjust your responses accordingly.

Abstract (if available)

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

The angry brain: neural correlates of interpersonal provocation, directed rumination, trait direct aggression, and trait displaced aggression

PDF

Individual differences in heart rate response and expressive behavior during social emotions: effects of resting cardiac vagal tone and culture, and relation to the default mode network

PDF

Toward counteralgorithms: the contestation of interpretability in machine learning

Asset Metadata

Creator Ronquillo, Jaclyn Mae (author)

Core Title Neural correlates of learning and reversal of approach versus withdraw responses to in- and out-group individuals

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Psychology

Publication Date 06/09/2010

Defense Date 05/12/2010

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag OAI-PMH Harvest,Prejudice,Race,social neuroscience

Place Name USA (countries)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Monterosso, John R. (committee chair), Lickel, Brian (committee member), Miller, Norman (committee member), Overbeck, Jennifer R. (committee member), Read, Stephen J. (committee member)

Creator Email jaclyn.adachi@gmail.com,jronquil@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-m3119

Unique identifier UC1174421

Identifier etd-Ronquillo-3813 (filename),usctheses-m40 (legacy collection record id),usctheses-c127-334305 (legacy record id),usctheses-m3119 (legacy record id)

Legacy Identifier etd-Ronquillo-3813.pdf

Dmrecord 334305

Document Type Dissertation

Rights Ronquillo, Jaclyn Mae

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Repository Name Libraries, University of Southern California

Repository Location Los Angeles, California

Repository Email cisadmin@lib.usc.edu