Converging Evidence That Stereotype Threat Reduces Working Memory Capacity

Abstract

Although research has shown that priming negative stereotypes leads to lower performance among stigmatized individuals, little is understood about the cognitive mechanism that accounts for these effects. Three experiments tested the hypothesis that stereotype threat interferes with test performance because it reduces individuals' working memory capacity. Results show that priming self-relevant negative stereotypes reduces women's (Experiment 1) and Latinos' (Experiment 2) working memory capacity. The final study revealed that a reduction in working memory capacity mediates the effect of stereotype threat on women's math performance (Experiment 3). Implications for future research on stereotype threat and working memory are discussed.

Full text

Converging Evidence That Stereotype Threat Reduces
Working Memory Capacity
Toni Schmader and Michael Johns
University of Arizona
Although research has shown that priming negative stereotypes leads to lower performance among
stigmatized individuals, little is understood about the cognitive mechanism that accounts for these effects.
Three experiments tested the hypothesis that stereotype threat interferes with test performance because
it reduces individuals’ working memory capacity. Results show that priming self-relevant negative
stereotypes reduces women’s (Experiment 1) and Latinos’ (Experiment 2) working memory capacity.
The final study revealed that a reduction in working memory capacity mediates the effect of stereotype
threat on women’s math performance (Experiment 3). Implications for future research on stereotype
threat and working memory are discussed.
One of our brightest and most motivated undergraduate research
assistants was recently lamenting her upcoming date with the
Graduate Record Examination (GRE). In describing her past strug-
gles with these sorts of standardized tests, she stated that as soon
as she sees the math problems, she becomes intensely fascinated
by the physical details of her pencil or any other proximal stimu-
lus, so long as it is not the actual math problem she is meant to
solve. Although her experience could represent nothing more than
active avoidance of an aversive stimulus, we wondered if it might
also illustrate one of the ways that stereotype threat interferes with
performance on complex tests of cognitive abilities. Cognitive
psychology has identified working memory capacity as the ability
to focus one’s attention on a given task while keeping task-
irrelevant thoughts at bay (Engle, 2001). Thus, one explanation for
our research assistant’s frustration with standardized math tests is
that she experiences lower levels of working memory capacity in
these testing situations, perhaps because gender stereotypes place
an extra burden on her cognitive resources. In the research pre-
sented here, we set out to test the hypothesis that stereotype threat
interferes with performance on complex cognitive tasks by reduc-
ing individuals’ working memory capacity. In this article, we
integrate existing research on stereotype threat with what is known
about working memory capacity. We then present the results of
three experiments that examine whether manipulations of stereo-
type threat reduce working memory capacity.
Stereotype Threat and Performance
Stereotype threat refers to the phenomenon whereby individuals
perform more poorly on a task when a relevant stereotype or
stigmatized social identity is made salient in the performance
situation. Steele and his colleagues (Steele, 1997; Steele & Aron-
son, 1995; Steele, Spencer, & Aronson, 2002) maintained that this
reduced performance results from an added pressure or concern
that a poor performance could be seen as confirming a negative
social stereotype about their ingroup. Thus, in sharp contrast to
socialization theories, inherent ability theories, or even educational
resource theories for why men outperform women on math tests or
why European Americans outperform African Americans on stan-
dardized tests, stereotype threat offers a uniquely situational ex-
planation for these group-performance differences (see Steele,
1997, for review).
In support of the stereotype threat explanation, research shows
that group-performance differences can be eliminated when the
same test is given in a stereotype-free context (see Steele et al.,
2002, for a review). For example, African Americans show in-
creased stereotype activation and perform worse than their White
peers when the task they are performing is described as diagnostic
of intellectual ability (Steele & Aronson, 1995). However, when
the same task is framed as unrelated to intelligence, levels of
stereotype activation are much lower and African American stu-
dents perform equally to White students. Similarly, women per-
form worse than men on a math test when they are told that the test
has revealed gender differences in the past, but they perform
equally to men when they are told that the test is “gender fair”
(Spencer, Steele, & Quinn, 1999). The ease with which stereotype
threat can be created in testing situations is demonstrated further
by evidence that White men, a group that is not traditionally
thought of as being negatively stereotyped as being poor at math,
perform more poorly on a math test when they believe they will be
compared with Asian men (Aronson et al., 1999).
Toni Schmader and Michael Johns, Department of Psychology, Univer-
sity of Arizona.
Portions of this research were presented at the 14th Annual Meeting of
the American Psychological Society, New Orleans, Louisiana, June 2002,
and at the Fourth Annual Meeting of the Society for Personality and Social
Psychology, Los Angeles, February 2003. This research was supported in
part by National Science Foundation Grant BCS-0112427 and by a faculty
small grant awarded to Toni Schmader from the University of Arizona
Foundation. We are indebted to the countless hours put in by the research
assistants who ran these studies—Kristen Hooks, Amy Hooks, Jodie
Deutsch, Ryan Hendrix, Mike Stanfill, David Trotter, Yassiin Nasser, and
Martin Kaasa—and also to Amy Baesler for her help preparing materials.
We also thank Marchelle Barquissau and Brian Lickel for their valuable
help and advice in planning this research.
Correspondence concerning this article should be addressed to Toni
Schmader, Department of Psychology, University of Arizona, Tucson,
Arizona 85721. E-mail: schmader@u.arizona.edu
Journal of Personality and Social Psychology Copyright 2003 by the American Psychological Association, Inc.
2003, Vol. 85, No. 3, 440–452 0022-3514/03/$12.00 DOI: 10.1037/0022-3514.85.3.440
440

Taken together, these findings suggest that activating negative
stereotypes about a social identity that one possesses can create an
extra situational burden that interferes with the ability to perform
as well at a mental task as might otherwise be possible. Although
the body of research establishing the existence of these effects is
ever expanding, the processes by which performance is reduced
have remained elusive in most studies. The assertion that reduced
performance results from an added pressure or concern (Steele,
1997; Steele & Aronson, 1995) has led many researchers to ex-
amine the role of several affective mechanisms, such as anxiety,
evaluation apprehension, or physiological arousal, in producing
stereotype threat. For example, several studies have demonstrated
that stereotype threat conditions lead to higher levels of anxiety
that parallel performance decrements on complex tasks (Aronson
et al., 1999; Spencer et al., 1999; Stone, Lynch, Sjomeling, &
Darley, 1999). In addition, Spencer et al. (1999) found evidence
that self-reported anxiety partially mediated the effects of a ste-
reotype threat manipulation on women’s math performance. How-
ever, many other stereotype threat studies have reported no differ-
ences in self-reported anxiety between stereotype threat conditions
(Gonzales, Blanton, & Williams, 2002; Schmader, 2002;
Schmader & Johns, 2003; Steele & Aronson, 1995).
Although self-reported measures of anxiety show mixed results,
there is some evidence to suggest that stereotype threat involves
heightened levels of anxiety and arousal. For example, Blascovich,
Spencer, Quinn, and Steele (2001) reported that African Ameri-
cans show increases in blood pressure under conditions of stereo-
type threat, and in our previous research (Schmader & Johns,
2003), women who thought that their test performance would be
used as an indicator of women’s math ability, in general, felt that
it was more important that they do well on the test. In addition,
manipulations designed to reduce anxiety or arousal, such as
engaging in self-affirmations or expressive writing (Martens,
Johns, Greenberg, & Schimel, 2002), or providing a cue to mis-
attribute arousal (Stone et al., 1999), seem to diffuse the effects of
stereotype threat on performance.
The above findings provide converging evidence that stereotype
threat involves affective experiences associated with increased
anxiety and apprehension. However, these past studies also high-
light the extent to which researchers interested in the underlying
mechanism involved in stereotype threat have focused almost
exclusively (with one notable exception, Quinn & Spencer, 2001,
discussed below) on the affective side of the threat equation. In
contrast, our goal in the present research was to examine how the
fear of confirming a negative stereotype about a salient group
identity disrupts cognitive processing. Because most stereotype
threat studies involve evidence of reduced performance on a com-
plex cognitive task, we were interested in understanding the nature
of the cognitive disruption that stereotype threat causes. Specifi-
cally, we were interested in the effects of stereotype threat on
working memory capacity.
Working Memory Capacity
The contemporary conceptualization of working memory capac-
ity has its roots in theory and research on short-term memory, or
the type of memory used to retain and manipulate information for
immediate or near-immediate use. Baddeley and Hitch (1974)
expanded earlier theory on short-term memory by proposing a
model of working memory that distinguished between three dis-
tinct but interactive cognitive functions: the phonological loop, the
visuospatial sketchpad, and the central executive processor. The
current concept of working memory capacity is an articulated
version of the central executive processor (Engle, 2001), and refers
to that type of memory that is used to focus attention on tempo-
rarily activated information of interest while inhibiting other in-
formation that is irrelevant to the task at hand. Thus, working
memory capacity includes both the temporary storage of informa-
tion as well as an attentional capability (Engle, Tuholski, Laughlin,
& Conway, 1999). People with higher working memory capacity
are better able to suppress task-irrelevant information (Rosen &
Engle, 1998) as evidenced by their lower susceptibility to the
cocktail party effect (Conway, Cowan, & Bunting, 2001).
Research on working memory has focused almost exclusively
on assessing individual differences, the basic assumption being
that interpersonal variation in working memory capacity is predic-
tive, if not indicative, of variation in fluid intelligence (Engle et al.,
1999). Thus, much of the research that has been conducted on
working memory capacity has been aimed at developing construct
valid measures that can predict performance on complex cognitive
tasks (Engle et al., 1999; Klein & Fiss, 1999; Oberauer, Su¨
␤
,
Schulze, Wilhelm, & Wittman, 2000). This research indicates that
individuals with higher levels of working memory capacity per-
form better on a variety of cognitive ability tests (e.g., La Pointe &
Engle, 1990; Turner & Engle, 1989). Similarly, the measure of
working memory capacity used in the present research has been
shown to correlate significantly with both verbal (r ⫽ .34) and
quantitative (r ⫽ .33) Scholastic Aptitude Test (SAT) scores
(Turner & Engle, 1989).
The test of working memory capacity that we used in our
research is the operation-span task developed by Turner and Engle
(1989). The test consists of two separate tasks that are performed
concurrently. One task is a processing task in which participants
are presented with a mathematical equation [e.g., (2 ⫻ 3) ⫺ 5 ⫽
1] and must decide whether the answer given is correct or incor-
rect. The second task is a memory span task in which participants
are given a word to recall at a later point. Each word is presented
after an equation. Thus, participants might be presented with five
equation and word pairings before being cued to recall the five
words. Participants’ ability to correctly recall the words provides
an index of working memory capacity in that it reflects the ease
with which they can both process the math problems while simul-
taneously holding the words in their mind.
Although working memory capacity research has not tradition-
ally examined how situational manipulations can reduce working
memory capacity temporarily, there is evidence to suggest that
chronic levels of stress and anxiety might be associated with lower
levels of working memory capacity (Eysenck & Calvo, 1992). For
example, individuals who score high in trait anxiety or report
experiencing more life stress perform worse than their less-
stressed counterparts on measures of working memory capacity
(e.g., Derakshan & Eysenck, 1998; Klein & Boals, 2001). Klein
and Boals (2001) argued that life stress reduces working memory
capacity because people under stress dedicate some of their mental
resources to suppressing unwanted negative thoughts and feelings
that intrude during other tasks. In addition to the direct relationship
between general stress and working memory capacity, other re-
search suggests that stressful performance situations reduce the
441
STEREOTYPE THREAT AND WORKING MEMORY

working memory capacity of those who are high in trait anxiety
(Sorg & Whitney, 1992) or math anxiety (Ashcraft & Kirk, 2001).
This collection of findings is consistent with the supposition that
working memory capacity might be reduced in stressful testing
situations because stress-related thoughts consume valuable cog-
nitive resources. We conceptualize stereotype threat as a stressor in
that a negative social stereotype that is primed in a performance
situation poses a threat to one’s social identity (Schmader, 2002).
Furthermore, given evidence that stereotype threat involves in-
creased levels of stereotype activation (Davies, Spencer, Quinn, &
Gerhardstein, 2002; Steele & Aronson, 1995), one might imagine
that cognitive resources are being expended processing this infor-
mation, leaving fewer resources for the task at hand (Spencer,
2003). Quite simply, the present research was designed to test the
idea that a manipulation of stereotype threat would result in lower
working memory capacity among individuals targeted by that
stereotype.
Although working memory capacity, per se, has not been pre-
viously examined in stereotype threat research, work by Quinn and
Spencer (2001) does highlight some of the cognitive effects of
stereotype threat. They found that although women performed
worse than men on a set of complex word problems, they per-
formed equally to men when solving the same problems with the
key equations extracted from the text of the word problem. A
second study also demonstrated that women under stereotype
threat had greater difficulty generating strategies they could use to
solve the word problems. Quinn and Spencer interpreted these
results as evidence that stereotype threat interferes with problem-
solving ability, perhaps because it results in diminished cognitive
resources. Our research was designed to test this possibility
directly.
Overview of Research
The primary goal of this research was to investigate the impact
of stereotype threat manipulations on working memory capacity.
In two of the three studies presented here, we focused on women
who are targeted by the stereotype that women are inferior to men
in mathematical ability. Of interest, the role of working memory
capacity in solving mathematical equations has been studied (e.g.,
Ashcraft & Kirk, 2001), and there is evidence that individuals who
have higher trait math anxiety also score lower on an individual-
difference measure of working memory capacity (Ashcraft, 2002).
However, working memory capacity has not been examined as an
explanation for gender differences in math performance, let alone
differences that are produced by experimentally manipulated con-
ditions of stereotype threat.
Experiment 1 tested the hypothesis that women (but not men)
would show reduced working memory capacity when a testing
situation was framed as measuring math ability. Experiment 2 was
a conceptual replication of Experiment 1, comparing the working
memory capacity of Caucasian and Latino participants when the
task was said to be related to general intelligence. In the final
study, we used a different manipulation of stereotype threat and a
measure of working memory that does not involve math to directly
test the hypothesis that a reduction in working memory mediates
the effects of stereotype threat on women’s performance on a
standardized math exam.
Experiment 1
In our first study, women and men completed a measure of
working memory capacity under conditions of stereotype threat or
not. We hypothesized that if stereotype threat interferes with
working memory, then women who complete a working memory
test described as a measure of mathematical ability—a stereotype-
relevant ability domain—would show lower working memory
scores compared with men and compared with women in a control
condition. This pattern of results would provide initial evidence
that making the stereotype about women’s math abilities applica-
ble to test performance impairs working memory capacity—a
psychological capacity critical to performing well on complex
intellectual tasks (e.g., Engle et al., 1999).
Method
Participants and Design
The participants in this study were 40 male and 35 female undergraduate
psychology students who participated for credit toward a course require-
ment or for $10. Following past stereotype threat research on women and
math (Schmader, 2002; Spencer et al., 1999), we selected participants who
indicated in a previous mass survey session that they had scored 500 or
higher on the quantitative section of the SAT (or equivalent converted
American College Test score). In the same mass survey, we assessed
stereotype knowledge with the question, “Regardless of what you person-
ally believe, do you think there is a stereotype about women having less
mathematical ability than men?” (rated on a 7-point scale from 1 ⫽ not at
all to 7 ⫽ definitely). Only those who responded at or above the scale
midpoint of 4 were recruited to participate in the study. Participants were
randomly assigned to one of two test description conditions in a 2 (male or
female) ⫻ 2 (stereotype threat or control) factorial design. Data from 9 men
and 7 women in the working memory test description condition were
excluded from analyses because of lost data, leaving a final sample of 31
men and 28 women.
1
Materials
Working memory test. To measure working memory capacity, we
adapted a dual-processing test called the operation-span task that has been
developed and used extensively by Engle and his colleagues to assess
working memory (e.g., La Pointe & Engle, 1990; Turner & Engle, 1989).
In this task, participants evaluate mathematical equations while memoriz-
ing words for later recall. The equations begin with the multiplication or
division of two positive integers [e.g., 9 ⫻ 6]. The product of this operation
is then added to, or subtracted from, another positive integer. The answer
for the entire operation is included in the expression, and the participant is
asked to evaluate whether the equation is correct or incorrect [e.g., Is (9 ⫻
6) – 4 ⫽ 50?]. A word is presented after each mathematical equation and
at the end of a series of equation/word combination trials (i.e., a set)
participants are asked to recall as many of the words from the preceding
series as possible. The math equations are merely meant to engage partic-
ipants in a certain amount of cognitive processing; working memory
capacity is indexed as the number of words that participants recall correctly
from each equation/word set.
1
The lost data were caused by a computer programming error when we
initially began the study. The computer failed to record the data from one
block of trials in the control condition, resulting in missing data for this
subset of participants. Including these 16 participants in the primary
analyses does not change the results.
442
SCHMADER AND JOHNS

The mathematical equations used in the present study were presented in
the same format as just described and were designed to be moderately
challenging. For equations requiring multiplication, the integers used in
this expression ranged from 5 to 12. For equations requiring division, the
integers used were multiples of the integers ranging from 5 to 16 [e.g.,
128/16]. The integer added to, or subtracted from, the solution to the first
expression could range from 5 to 14. The answer presented with each
equation was either correct or incorrect by plus or minus one and could be
positive or negative. A pool of 72 equations was generated using these
criteria (36 correct and 36 incorrect). The 72 words used in the test were
randomly selected from a pool of one-syllable words used by La Pointe and
Engle (1990).
The test we created included 18 sets of equation/word trials that con-
tained three to five equation/word combinations (i.e., there were six blocks
of each set size for a total of 72 equation/word trials). The sets were
presented in random order so participants did not know how many equa-
tions and words they would be required to evaluate and recall at the
beginning of each set. The equations and words were presented in random
order within each set, but equations and words were assigned to the same
set for each test. The test was administered on a computer and the
participant controlled the presentation of the stimuli with their responses.
Participants were instructed to evaluate the correctness of the equations
quickly and accurately while also remembering the words for later recall.
Each set began with the presentation of an equation to be evaluated for
correctness. After recording their evaluation using the keyboard (1 ⫽
correct, 2 ⫽ incorrect), participants were presented a to-be-remembered
word for 2 s. A blank screen lasting 1 s separated the presentation of each
equation and word. After presentation of all equation/word combinations in
a set, participants were prompted to recall all of the words in that set. Each
set was separated by the prompt “next set,” which was displayed for 3 s.
The computer recorded the words recalled, participants’ correct and incor-
rect responses to the equations, and the time spent on each equation.
Procedure
Two female research assistants ran the experimental sessions in same-
gender groups ranging from 2 to 4 participants. Upon entering the lab,
participants were seated in individual rooms and asked to listen to, and read
along with, a prerecorded description of the study broadcast on an intercom
system and displayed on the computer monitor. This study description
served as the manipulation of stereotype threat and was delivered by a male
who identified himself as a researcher in the psychology department. In the
control condition, he described the test as a reliable measure of working
memory capacity. He went on to describe working memory capacity as the
ability to hold different pieces of information simultaneously while trying
to process one specific piece of information. In the stereotype threat
condition, the male researcher described the test as a reliable measure of
“quantitative capacity.” Quantitative capacity was described as the ability
to solve complex mathematical equations while trying to process multiple
pieces of information related to the problem-solving task. To prime the
stereotype related to women’s math ability, participants were also told that
gender differences in math performance might stem from underlying
gender differences in quantitative capacity. To increase the personal rele-
vance of the performance situation, participants in both conditions were
informed that they would receive feedback about their performance.
After the stereotype threat manipulation, participants were presented
with a general overview of the testing procedure. They were told that their
performance on the test would be based on both how accurately they
evaluated the equations and the number of words they recalled correctly.
After the instructions, participants were allowed to complete a practice set
of three equation/word trials. Before beginning the test, participants in both
conditions were asked to indicate whether they were right-handed or
left-handed, and participants in the stereotype threat condition were also
asked to indicate their gender and their last name.
Following the test, participants completed a brief test experience ques-
tionnaire that contained an anxiety scale and items related to perceptions of
the test and testing situation. Anxiety during the tests was measured using
questions adapted from the Spielberger State Anxiety Scale (Spielberger,
Gorsuch, & Lushene, 1970). Participants rated on a 7-point scale (ranging
from 1 ⫽ not at all to 7 ⫽ very much) how much they felt anxious,
comfortable, jittery, worried, at ease, nervous, relaxed, and calm while
taking the test. The items on the scale were averaged (after reverse scoring
comfortable, at ease, relaxed, and calm) to form an index of state anxiety,
where higher numbers indicated more anxiety (
␣
⫽ .90). Participants were
also asked to rate the difficulty of the working memory test on a 7-point
scale ranging from 1 (extremely easy)to7(extremely difficult).
To determine if participants perceived the test description in a manner
consistent with stereotype threat, we asked them to rate how much they
agreed with the following statements (tailored to participant gender): “Iam
concerned that the researcher will judge [women/men], as a whole, based
on my performance on this test” and “The researcher will think that
[women/men], as a whole, have less math ability if I did not do well on this
test.” Responses were recorded using a 7-point scale ranging from 1
(strongly disagree)to7(strongly agree) and were averaged to create a
reliable index of gender identity threat (
␣
⫽ .90). After completing this
questionnaire the participants were debriefed and thanked.
Results
An initial Gender ⫻ Stereotype Threat analysis of variance
(ANOVA) on participants’ quantitative SAT scores revealed that
men’s SAT scores (M ⫽ 620) were higher than women’s(M ⫽
579), F(1, 55) ⫽ 4.60, p ⬍ .05. Because previous research has
found that performance on the operation-span task is significantly
correlated with SAT scores (Turner & Engle, 1989), we controlled
for these group differences in SAT scores by conducting 2 (gen-
der) ⫻ 2 (stereotype threat) analyses of covariance (ANCOVAs)
controlling for SAT when SAT was found to be a significant
covariate.
Working Memory Capacity
Absolute span score. In past research, working memory ca-
pacity is assessed either as the total number of words recalled or
the number of words recalled taking into account only sets recalled
perfectly (called the absolute span score). The absolute span score
is derived by summing the total number of words from only those
sets of words where all the words in the set were recalled correctly.
So, if a participant only recalled three words from a four-word set,
then these three words would not count toward the total score. If,
however, all four words were recalled correctly, all four words
would count toward the final score. In this way, the absolute span
score is thought to provide a more sensitive measure of working
memory capacity (La Pointe & Engle, 1990). In all of the studies
reported here, we used the absolute span score as the assessment of
working memory capacity; however, in all three studies, analyses
using the total words recalled yielded the same conclusions.
An ANCOVA on the absolute span score yielded significant
main effects of gender, F(1, 54) ⫽ 4.81, p ⬍ .05 (women
M ⫽ 44.23 and men M ⫽ 49.80), and stereotype threat, F(1,
54) ⫽ 23.84, p ⬍ .001 (stereotype threat M ⫽ 41.05 and control
M ⫽ 52.98), and the predicted two-way interaction, F(1,
54) ⫽ 15.69, p ⬍ .001. SAT was a significant covariate, F(1,
54) ⫽ 4.98, p ⬍ .05. Simple main effects tests revealed that
women in the stereotype threat condition (M ⫽ 33.44) recalled
443
STEREOTYPE THREAT AND WORKING MEMORY

fewer words than men in the stereotype threat condition
(M ⫽ 48.66), F(1, 54) ⫽ 19.38, p ⬍ .001, d ⫽ 1.19, and than
women in the control condition (M ⫽ 55.03), F(1, 54) ⫽ 37.49,
p ⬍ .001, d ⫽ 1.66. The span score for men in the control
condition (M ⫽ 50.94) was not significantly different from the
score of men in the stereotype threat condition and women in the
control condition (Fs ⬍ 1.5). The means are displayed in Figure 1.
This pattern of results supports our prediction that stereotype threat
reduces women’s working memory capacity.
Equation evaluation. Although working memory is assessed
as a function of performance on the word recall, we also analyzed
performance on the equations to assess whether there were any
significant differences due to the stereotype threat manipulation.
Prior research indicates that manipulations of stereotype threat
only reduce women’s performance on very complex tests of their
math ability (Spencer et al., 1999), typically those that require
extracting an equation from a word problem (Quinn & Spencer,
2001). Thus, we did not expect the stereotype threat manipulation
to affect women’s ability to evaluate the equations correctly. In
fact, Quinn and Spencer (2001) found that stereotype threat does
not reduce women’s ability to solve basic equations like those used
in our task. In line with these past results, there were no significant
effects of gender or stereotype threat on percentage of equations
solved correctly (Grand mean (GM) ⫽ 78%; Fs ⬍ 1.5). Analysis
of the average amount of time (in seconds) spent on each equation
revealed a marginal main effect of stereotype threat, F(1,
54) ⫽ 3.44, p ⬍ .10. SAT was a significant covariate, F(1,
54) ⫽ 4.41, p ⬍ .05, but no other effects were significant
(Fs ⬍ 1.5). Regardless of gender, participants in the stereotype
threat condition tended to spend more time evaluating each equa-
tion (M ⫽ 8.53) than did participants in the control condition
(M ⫽ 7.42).
Test Experience Questionnaire
Anxiety. Analysis of the anxiety measure did not yield any
significant main effects or interactions (Fs ⬍ 2).
Perceived difficulty. Analysis of the difficulty ratings revealed
a significant two-way interaction, F(1, 54) ⫽ 4.71, p ⬍ .05. SAT
was a significant covariate, F(1, 54) ⫽ 4.04, p ⬍ .05; neither main
effect was significant (Fs ⬍ 1.5). Simple effects analyses showed
that within the stereotype threat condition, women rated the test as
more difficult (M ⫽ 4.63) than did men (M ⫽ 3.54), F(1,
54) ⫽ 5.70, p ⬍ .05. The difficulty ratings of women (M ⫽ 3.96)
and men (M ⫽ 4.26) in the control condition were not significantly
different (F ⬍ 1).
Gender identity threat. Analysis of the gender identity threat
composite yielded a main effect for the stereotype threat manipu-
lation, F(1, 55) ⫽ 8.29, p ⬍ .01. Both women and men in the
stereotype threat condition expressed greater concern (M ⫽ 2.47)
that the researcher would evaluate their performance in terms of
their gender identity compared with participants in the control
condition (M ⫽ 1.61). No other effects were significant (Fs ⬍ 2).
This mean pattern reveals that the manipulation did produce some
conscious awareness that the researcher might use their gender as
a factor to evaluate their performance, but this concern was not
unique to women.
Discussion
The results of this first study provide initial support for the
hypothesis that a stereotype threat manipulation—framing a task in
terms of a stereotyped group deficiency—can lead to a measurable
decrease in cognitive resources. As predicted, women completing
a working memory test described as a test related to mathematical
ability showed reduced cognitive capacity, as measured by the
number of words they were able to recall within the task. Men, and
women in a nonthreat control condition, did not exhibit this work-
ing memory decrement and were able to recall an equal number of
words. This is the exact pattern of performance that would be
predicted by stereotype threat. Measures designed to assess the
psychological experience created by the testing situation indicated
that women experienced the test as more difficult than men in the
stereotype threat condition despite the finding that both men and
women indicated that they were concerned their performance on
the test might be evaluated as a function of their gender identity.
This would seem to indicate that although gender was salient for
both women and men in the stereotype threat condition, it was only
women, who are specifically targeted by the stereotype, that were
affected by the fear of confirming the researchers’ expectations. As
in some other stereotype threat research (e.g., Schmader, 2002;
Steele & Aronson, 1995), however, participants did not report
differential levels of anxiety as a function of the test description.
Experiment 2
The results of Experiment 1 represent an important step in
understanding exactly how stereotype threat works to undermine
the performance of negatively stereotyped group members on
challenging academic tasks. In the stereotype threat condition,
both women and men expressed apprehension that the researcher
would evaluate their performance in relation to their gender iden-
tity, but only the performance of women suffered under this
concern. This pattern of results is consistent with our hypothesis
that having to contend with negative group stereotypes places an
additional cognitive burden on people that can interfere with their
ability to perform up to their potential on complex cognitive tasks.
One further question is whether the same effect can be found with
a different stereotyped group in a different domain. Thus, the
second study was designed to replicate the negative effects of
Figure 1. Experiment 1: The effects of gender and stereotype threat on
participants’ working memory capacity.
444
SCHMADER AND JOHNS

stereotype threat on the working memory capacity of Latinos who
are negatively stereotyped on intellectual tasks (e.g., Gonzales et
al., 2002).
An additional goal of our second experiment was to address an
alternative explanation for the results of Study 1. The stereotype
threat manipulation we used in Experiment 1 involved two com-
ponents: We described the test as a measure of math ability, and
we also explicitly mentioned the idea that there are gender differ-
ences in math performance. One could argue that this direct
priming of gender stereotypes might have created an experimental
demand that produced the results we observed rather than actual
effects on cognitive capacity. The fact that we did not observe
effects on women’s performance when evaluating the math equa-
tions—both accuracy and time spent evaluating the equations—
speaks against the idea that participants’ behavior was primarily
affected by expectancies; however, we wanted to address this
possibility by using a more subtle manipulation of stereotype
threat that did not explicitly mention group differences in
performance.
For this study we compared the performance of Latino students
to the performance of White students under conditions of stereo-
type threat. On the basis of the cultural stereotype that Latinos are
less intelligent than Whites (Gonzales et al., 2002), we created
stereotype threat by framing the working memory test as a task that
was highly predictive of general intelligence. Rather than explic-
itly mentioning ethnic differences in performance, we primed
ethnic identity in the stereotype threat condition by asking partic-
ipants to identify their ethnicity on a demographic questionnaire
prior to completing the working memory measure (Steele & Aron-
son, 1995). We expected that Latino students would show evi-
dence of reduced working memory capacity compared with Whites
when the test was described as a measure related to intelligence.
We did not expect any performance differences when the test was
described as a measure of working memory.
Method
Participants and Design
We recruited 33 Latino (20 women, 13 men) and 40 White (27
women, 13 men) psychology students only on the basis of their self-
reported ethnicity in an earlier mass survey. All participants received credit
for a research requirement. Participants were randomly assigned to one of
two conditions in a 2 (Latino or White) ⫻ 2 (stereotype threat or control)
factorial design. The data of 1 White participant were lost because of a
computer error, leaving a final sample of 72.
Materials and Procedure
Two White female experimenters conducted the sessions in mixed ethnic
groups ranging from 2 to 4 participants.
2
The materials and procedures
were the same as Study 1 except for the stereotype threat manipulation. The
stereotype threat manipulation was again presented in a prerecorded de-
scription delivered by a male “researcher,” but unlike the previous study,
the researcher described the test as a reliable measure of working memory
capacity in both the control condition and the stereotype threat condition.
In the stereotype threat condition, the researcher went on to say that
research had shown that performance on the working memory test is
“highly predictive” of performance on intelligence tests and that their
performance on the test would be used to “help establish norms for
different groups.” The test description in both conditions ended by inform-
ing participants that they would receive feedback about their performance.
As in Study 1, this was done to increase the personal relevance of the
testing situation. In addition, to further prime ethnic identity in the stereo-
type threat condition, participants in this condition were also asked to
indicate their ethnicity before beginning the test.
After the task, all participants completed a test experience questionnaire
containing the same measure of anxiety (
␣
⫽ .87) and difficulty used in the
previous study. The questionnaire also included the two identity threat
items from the previous study modified to assess whether participants
thought that the researcher would evaluate them on the basis of their
ethnicity. Responses were averaged to create a reliable index of ethnic
identity threat (
␣
⫽ .90). We also had data on participants’ self-reported
verbal and quantitative SAT scores from a mass survey that they had
completed at the beginning of the semester. Participants were debriefed and
thanked after completing the final questionnaire.
Results
Because of the quasi-experimental design of this study, we
conducted initial 2 (ethnicity) ⫻ 2 (gender) ⫻ 2 (stereotype threat)
ANOVAs on participants’ quantitative and verbal SAT scores to
discern whether there were any group differences in these vari-
ables that should be controlled in our analyses. There were no
significant differences in quantitative SAT (all Fs ⬍ 2), but there
was a marginal ethnic difference in verbal SAT scores, F(1,
59) ⫽ 2.60, p ⫽ .10. No other effects were significant (Fs ⬍ 1.5).
Because prior research reveals a significant relationship between
verbal SAT and working memory capacity (Turner & Engle,
1989), we controlled for this variable in all analyses where it was
a significant covariate. Degrees of freedom for these analyses are
lower because controlling for verbal SAT resulted in the loss of 5
additional participants who did not report their SAT scores.
Working Memory Capacity
Absolute span score. Analysis of the absolute span score pro-
duced the predicted Ethnicity ⫻ Stereotype Threat interaction,
F(1, 58) ⫽ 4.48, p ⬍ .05 (see Figure 2). Verbal SAT was a
significant covariate, F(1, 58) ⫽ 11.83, p ⬍ .001. Simple effects
testing revealed that Latinos in the stereotype threat condition
(M ⫽ 41.29) recalled significantly fewer words than did Whites in
the stereotype threat condition (M ⫽ 51.93), F(1, 58) ⫽ 6.45, p ⬍
.05, d ⫽ .66, and Latinos in the control condition (M ⫽ 50.65),
F(1, 58) ⫽ 4.19, p ⬍ .05, d ⫽ .55. Recall by Whites in the control
condition (M ⫽ 48.10) was equivalent to that of Latinos in the
control condition and to the recall of Whites in the stereotype
threat condition (Fs ⬍ 1). The analysis also yielded an unexpected
main effect of gender, F(1, 58) ⫽ 4.10, p ⬍ .05 (men recalled more
words than women), but no other effects were significant (Fs ⬍ 2).
Equation evaluation. Accuracy of responses to the equations
did not vary as a function of ethnicity, stereotype threat, or the
interaction of the two (Fs ⬍ 1). There was, however, a significant
gender difference in equation accuracy, F(1, 58) ⫽ 4.97, p ⬍ .05
(men were more accurate than women), that was qualified by an
unexpected Gender ⫻ Stereotype Threat interaction, F(1,
2
We ran mixed ethnic groups in this study in contrast to homogeneous
groups that we had used for Experiment 1 because we were concerned that
it would be seen as unusual to have only four Latinos in a session and that
this alone might prime ethnicity.
445
STEREOTYPE THREAT AND WORKING MEMORY

58) ⫽ 5.21, p ⬍ .05. No other effects were significant (Fs ⬍ 1).
Verbal SAT was a marginally significant covariate, F(1,
58) ⫽ 3.34, p ⬍ .08. Simple effects tests of this interaction showed
that although men and women were equally accurate in their
evaluation of the problems in the control condition (both Ms ⫽
81%; F ⬍ 1), women (M ⫽ 76%) were less accurate than men
(M ⫽ 86%) in their evaluations of the problems when the test was
described as predictive of intelligence, F(1, 58) ⫽ 11.21, p ⬍ .001.
There were no significant differences in the average time (in
seconds) participants spent evaluating each equation (GM ⫽ 9.02;
Fs ⬍ 2.5).
Test Experience Questionnaire
Anxiety. Analysis of the anxiety composite yielded a marginal
Gender ⫻ Ethnicity interaction, F(1, 64) ⫽ 3.63, p ⬍ .10, and a
significant Ethnicity ⫻ Stereotype Threat interaction, F(1,
64) ⫽ 5.07, p ⬍ .05. Simple effects analysis indicated that Latinos
in the stereotype threat condition reported significantly more
anxiety (M ⫽ 4.05) compared with Latinos in the control condi-
tion (M ⫽ 3.14), F(1, 64) ⫽ 4.84, p ⫽ .05, whereas the stereo-
type threat manipulation had no effect on self-reported anxiety
of Whites (M ⫽ 3.41 in stereotype threat; M ⫽ 3.78 in control;
F ⬍ 1).
Test difficulty. Analysis of the perceived difficulty of the test
yielded a gender main effect, F(1, 64) ⫽ 6.82, p ⬍ .01 (women
perceived the test to be more difficult, M ⫽ 4.37, than did men,
M ⫽ 3.65), and a marginal Ethnicity ⫻ Stereotype Threat inter-
action, F(1, 64) ⫽ 3.66, p ⬍ .06. No other effects were significant
(Fs ⬍ 1.5). Simple effects tests revealed that Whites (M ⫽ 4.19)
and Latinos (M ⫽ 3.99) saw the test as equally difficult in the
control condition (F ⬍ 1), whereas Latinos (M ⫽ 4.44) rated the
test to be more difficult than did Whites (M ⫽ 3.54) under
stereotype threat, F(1, 64) ⫽ 4.96, p ⬍ .05.
Ethnic identity threat. Analysis of the ethnic identity threat
measure yielded only a marginal Ethnicity ⫻ Stereotype Threat
interaction, F(1, 64) ⫽ 2.69, p ⫽ .11. Latinos reported slightly
more ethnic identity threat in the stereotype threat condition
(M ⫽ 1.91) than in the control condition (M ⫽ 1.47), whereas
Whites reported slightly less ethnic identity threat in the stereotype
threat condition (M ⫽ 1.47) than in the control condition
(M ⫽ 1.97), although none of the simple effects were significant.
No other effects were significant (Fs ⬍ 1).
Discussion
The results of the second study provide further evidence for the
negative effects of stereotype threat on working memory capacity.
When the working memory test was described as a measure related
to intelligence, Latinos recalled fewer words compared with
Whites and compared with Latinos in the nonthreat control group.
These results demonstrate that the working memory task we chose
is not uniquely relevant to stereotypes about math performance but
can be used to assess the capacity deficits experienced by other
stereotyped groups during a cognitively taxing test. This study also
provided a stronger test of our hypothesis because, although par-
ticipants were told that their performance on the test would be used
to establish norms for different groups, the issue of group differ-
ences in the stereotyped performance domain was never mentioned
explicitly.
The less direct nature of the manipulation might explain why
there were no significant differences in the self-reported ratings of
ethnic identity threat across test description and ethnicity. Both
Latinos and Whites indicated on a 7-point scale that they were
fairly unconcerned (Ms ⬍ 2) that the researcher would evaluate
their performance in terms of their ethnicity. However, unlike our
previous studies focusing on gender stereotypes, this experiment
did show anxiety effects that paralleled the effects on working
memory capacity. When the working memory test was described
as being related to intelligence, Latinos reported feeling more
anxious than Whites did. However, further analysis shows that
participants’ self-reports of anxiety were uncorrelated with their
working memory score (r ⫽ .01), highlighting a disconnection
between the processing effects of stereotype threat and the con-
scious experience of stereotype threat.
Experiment 3
The studies presented so far provide consistent support for the
proposition that stereotype threat impairs performance on chal-
lenging cognitive tests by reducing the cognitive resources avail-
able to the test taker. However, these studies have only shown that
a manipulation of stereotype threat reduces working memory ca-
pacity. We know from other research that manipulations of ste-
reotype threat also reduce performance on standardized tests (e.g.,
Spencer et al., 1999) and that there is a significant relationship
between working memory capacity and performance on standard-
ized tests (Turner & Engle, 1989). Thus, the next step was to
conduct an experiment designed to test whether working memory
capacity mediates the effect of stereotype threat on academic test
performance. To test for mediation, we had women complete both
the working memory measure and a standardized math test under
stereotype threat or nonthreat conditions. We expected that if
working memory capacity mediates stereotype threat effects, then
scores on the working memory test should account for perfor-
mance decrements on a standardized math test.
To test our hypothesis more precisely, two other changes were
made in this final study. First, we modified our measure of work-
ing memory capacity to rule out an alternative explanation for our
Figure 2. Experiment 2: The effects of ethnicity and stereotype threat on
participants’ working memory capacity.
446
SCHMADER AND JOHNS

results. In the previous studies, the processing component of the
working memory task required solving mathematical equations.
This task is obviously related to the stereotyped domain in Exper-
iment 1 and also could be perceived as related to the stereotyped
domain in Experiment 2 (i.e., intelligence). This situation raises
the possibility that when the stereotype was made salient in the
performance domain, participants might have assumed that the
processing component of the working memory measure was more
important for their performance on the test. As a consequence, they
might have shifted more of their attentional focus to evaluating the
equations rather than remembering and recalling the words. If this
were the case, then lowered word recall would occur as a function
of shifting focus and not a general reduction in working memory.
To address this artifactual explanation, we used a measure of
working memory capacity that required a similar amount of pro-
cessing time but did not involve mathematical equations. In this
study, the processing component of the task involved counting the
number of vowels in a sentence. We reasoned that if the effect of
stereotype threat on word recall is caused by a disproportionate
focus on the processing task, then making the task unrelated to the
stereotyped domain should reduce any influence of this potential
artifact.
In addition to modifying the measure of working memory, we
also adopted a different manipulation of stereotype threat. In the
previous two experiments, we manipulated stereotype threat by
telling participants that the working memory task was in fact
measuring, or was highly related to, the relevant stereotyped
ability (i.e., math or intelligence). We did this in an effort to assess
the effects of stereotype threat online, that is, while participants
actually believed they were in the performance situation. However,
connecting the stereotype threat manipulation to the working
memory test directly allows for the possibility that reduced work-
ing memory capacity simply represents another stereotype threat
effect and not necessarily a mediating mechanism. Our final ex-
periment was designed to address this conceptual ambiguity. Spe-
cifically, we sought to demonstrate that stereotype threat reduces
working memory even when the threat is not directly connected to
the measure of working memory. To test this possibility, we used
a manipulation that would create a general atmosphere of stereo-
type threat during the session. Specifically, women assigned to the
stereotype threat condition participated in a session as the solo
woman in a session of other male participants (Inzlicht and Ben-
Zeev, 2000; Sekaquaptewa & Thompson, 2003) and were told that
they would be taking a math exam later in the session (i.e., after
they completed the measure of working memory capacity).
Women in the control condition were run in a session of only other
women and were told that they would be completing a problem-
solving task later in the session. Previous research has shown that
priming stereotype threat can reduce performance expectancies
(Stangor, Carr, & Kiang, 1998) and can also lead to self-
handicapping before the critical performance (Stone, 2002). These
results suggest that stereotype threat can influence behavior even
in anticipation of the performance situation. Given these past
findings, we expected that women would exhibit a reduction in
working memory capacity after stereotype threat was primed but
before beginning the stereotype-relevant task and that this reduc-
tion would mediate the effect of stereotype threat on performance.
Method
Participants and Design
The participants were 31 female undergraduates who were randomly
assigned to complete two tasks (the working memory test and a standard-
ized math test) under stereotype threat or control conditions. The partici-
pants were selected on the basis of the same criteria and method used in
Experiment 1 (math SAT ⱖ 500 and self-reported knowledge of the
stereotype) and completed the study in exchange for course credit or $10.
Data from 3 participants in the control condition were lost, 1 because of a
computer malfunction and 2 because of a procedural error, leaving a final
sample of 28.
Materials
For this study, we created a version of the working memory test that did
not involve mathematical equations. Instead, the processing component of
the test required participants to count the number of vowels contained
within a given sentence. The sentences were 7–12 words long and con-
tained an average of approximately 10 vowels. After being presented with
each sentence, participants were given a word to recall. Furthermore, to
minimize fatigue we reduced the total number of equation/word combina-
tion trials from 72 to 60. The set sizes for this test were increased to range
from four to six equation/word combination trials (four blocks of each set
size) to account for any loss of sensitivity caused by reducing the total
number of words.
The standardized math test we used consisted of 30 multiple-choice
word problems taken from the quantitative section of practice GREs.
Because these problems had been included in actual GRE tests, we had data
on performance norms as indexed by the proportion of test takers who
answered each problem correctly in previous administrations. To ensure
that the test was difficult but appropriate for the skill level of our sample,
we selected problems that ranged from 44% to 80% accuracy, with a mean
accuracy of 64%. The final test was administered to participants in paper-
and-pencil format.
Procedure
Stereotype threat manipulation. To prime and maintain the salience of
stereotype threat during the completion of the working memory test, we
adapted procedures used by Inzlicht and Ben-Zeev (2000; see also Seka-
quaptewa & Thompson, 2003) to test the effects of solo gender status on
women’s math performance. Participants in both conditions completed the
study in groups of 3 people. Women in the stereotype threat condition
completed a session conducted by a male experimenter with 2 other male
confederates acting as participants. Women in the control condition com-
pleted the working memory test in a session conducted by a female
experimenter along with 2 other female participants.
Upon entering the lab, the participants assembled in a large room where
they received a brief overview of the experimental session and signed the
consent form. Participants in both conditions were told that they would be
completing two completely unrelated tasks. In the stereotype threat con-
dition, the experimenter explained that the main goal of the study was to
administer a test of mathematical aptitude to collect normative data on men
and women. In the control condition, participants were told that the
primary purpose of the study was to administer a problem-solving exercise
to collect normative data on college students. After this introduction, the
experimenter explained that because of a room scheduling conflict they
would first need to complete a brief computer task in another part of the
lab. The participants were then escorted to individual rooms equipped with
computers where they completed the vowel-counting version of the work-
ing memory test. There was no mention of math or math ability in
connection with this test.
447
STEREOTYPE THREAT AND WORKING MEMORY

Working memory questionnaire. Once all the participants completed
the working memory test, they were given a brief questionnaire designed
to get a sense of how they perceived the test. To assess whether participants
might have seen the test as a measure linked to math ability, we asked them
to rate the extent to which they thought performance on the test would be
related to mathematical ability and memory ability on a 7-point scale
ranging from 1 (not at all related)to7(highly related). To address the
possibility that participants in the stereotype threat condition might have
shifted more of their attention to the processing task because it seemed
most related to math ability, we asked participants to rate on an 11-point
scale how important they thought each part of the task (counting vowels
and remembering words) was for determining their overall performance
(1 ⫽ counting vowels is most important, 6 ⫽ words and vowels are equally
important, 11 ⫽ remembering the words is most important). Participants
also indicated how they allocated their attention while completing the task
using an 11-point scale ranging from 1 (counting the vowels)to6(focused
equally on counting the vowels and remembering the words) to 11 (re-
membering the words). Finally, to assess the possibility that stereotype
threat reduces women’s expectations for their test performance, partici-
pants were asked to rate how well they expected to perform on the
upcoming math test/problem-solving exercise using a 9-point scale ranging
from 1 (poor)to9(excellent).
Math test. Participants were then escorted back to the large room
where they were seated individually at small tables and introduced to the
math test. In the control condition, the test was described as a pilot test of
materials to be used in future research on problem-solving processes. No
mention of math was made. In the stereotype threat condition, participants
were told the test was a reliable measure of mathematical aptitude. To
avoid demand characteristics, there was no explicit mention of gender
differences in performance or ability. Participants in both conditions were
told that the test would be difficult but within their range of ability. They
were asked to give a genuine effort and to avoid random guessing and were
told that they would receive feedback about their performance afterwards.
Before beginning, women in the stereotype threat condition were asked to
indicate their gender and provide their last name on the cover of the test
booklet. Participants were given 20 min to work on the test and then
completed a brief posttest questionnaire that included the measures of
anxiety, perceived difficulty (of the math test), and gender identity threat
(
␣
⫽ .84) used in Experiments 1 and 2. To further assess expectancies, we
also asked participants to rate how they thought the researcher expected
men and women to do relative to each other on the math test, using a
7-point scale ranging from 1 (men will score better than women)to4(men
and women will score the same)to7(women will score better than men).
After completing the questionnaire, participants were probed for suspicion,
thoroughly debriefed, and thanked for their participation.
Results and Discussion
Working Memory Capacity
Absolute span score. As predicted, women in the stereotype
threat condition (M ⫽ 26.78) recalled fewer words on the working
memory test than did women in the control condition (M ⫽ 38.86),
t(26) ⫽ 3.13, p ⬍ .01, d ⫽ 1.19.
Vowel counting. The accuracy of the number of vowels
counted correctly did not vary as a function stereotype threat
(GM ⫽ 87%; t ⬍ 1), and there were no significant differences in
the average time (in seconds) participants spent counting the
vowels (GM ⫽ 7.60; t ⬍ 1). It is also important to point out that
the average amount of time participants spent counting the vowels
was not substantially different than the average amount of time
participants spent evaluating the difficult equations in Experi-
ments 1 and 2 (M ⫽ 8.32).
Perceptions of the Working Memory Task
Women’s perceptions of what performance on the working
memory task would be related to (mathematical ability and mem-
ory ability) were analyzed with a 2 (stereotype threat) ⫻ 2 (ability
type) mixed-factors ANOVA. This analysis revealed only a main
effect of ability type, F(1, 26) ⫽ 58.76, p ⬍ .001. Participants in
both conditions indicated that they perceived performance on the
working memory test to be more related to memory ability
(M ⫽ 6.14) than mathematical ability (M ⫽ 3.71).
Additional analyses revealed that participants in the control
condition (M ⫽ 8.20) and the stereotype threat condition
(M ⫽ 7.46) did not differ significantly in their perceptions of
which part of the working memory task was more important
(vowel counting or word recall; p ⬎ .20); and participants in both
conditions generally viewed word recall to be the more important
portion of the test (compared with the midpoint ⫽ 6), t(26) ⫽ 6.31,
p ⬍ .001. There were also no significant differences between
conditions in what part of the test participants reported focusing on
(stereotype threat M ⫽ 8.00; control M ⫽ 7.31; p ⬎ .15), and
participants focused more on remembering the words than count-
ing the vowels (compared with the midpoint ⫽ 6), t(26) ⫽ 6.52,
p ⬍ .001. Together these results reduce the plausibility of the
alternative explanation that lowered word recall in the stereotype
threat condition resulted from participants simply shifting their
focus to the processing task and away from memorizing and
recalling the words.
Math Test Performance
Because very few participants were able to answer all 30 ques-
tions in the time allotted for the test, performance on the math test
was analyzed as women’s accuracy on the test, that is, the number
of problems answered correctly divided by the number of prob-
lems attempted (Inzlicht & Ben-Zeev, 2000; Marx & Roman,
2002). Replicating previous stereotype threat research, women in
the stereotype threat condition (M ⫽ 0.49) were less accurate on
the math test than women in the control condition (M ⫽ 0.66),
t(26) ⫽ 2.38, p ⬍ .05, d ⫽ .88. There were no significant
differences in the number of math problems attempted by women
in the stereotype threat condition (M ⫽ 16.15) and women in the
control condition (M ⫽ 15.47; t ⬍ .50), suggesting that women in
both conditions expended comparable effort on the test.
Test Experience Questionnaire
Anxiety, perceived difficulty, and gender identity threat. As in
Experiment 1, there were no significant differences in self-reported
anxiety between the two conditions (t ⬍ 1). The average anxiety
rating for both conditions was at the midpoint of the scale
(GM ⫽ 4.11). However, whereas stereotype threat had a signifi-
cant effect on women’s difficulty ratings and perceptions of gender
identity threat in Experiment 1, the stereotype threat manipulation
in Experiment 3 did not affect either their difficulty ratings for the
math test (GM ⫽ 4.93) or their ratings of gender identity threat
(GM ⫽ 2.75; both ts ⬍ 1). This inconsistency in the conscious
experience of stereotype threat across the two studies with women
probably reflects the fact that our manipulation of threat was more
448
SCHMADER AND JOHNS

explicit in the first experiment in which we explicitly told women
that there were gender differences on the test.
Performance expectancies on the math test/problem-solving ex-
ercise. After the working memory task, we asked participants to
rate how well they expected to do on the upcoming task (described
as either a math test in the stereotype threat condition or as a
problem-solving exercise in the control condition). There were no
significant differences in the expectancies of participants in the
stereotype threat condition (M ⫽ 5.62) and participants in the
control condition (M ⫽ 6.00; t ⬍ 1). After completing the math
test, participants also rated how they thought the researcher ex-
pected men and women to do relative to each other on the task.
Women in the stereotype threat condition (M ⫽ 3.00) were no
more likely to believe that the researcher expected gender differ-
ences than were women in the control condition (M ⫽ 3.47; p ⫽
.19). Overall, women in both conditions (GM ⫽ 3.25) tended to
believe that the researcher expected men to outperform women
(compared with the midpoint ⫽ 4), t(26) ⫽⫺4.27, p ⬍ .001.
Mediational Analyses
To test whether the reductions of working memory mediated the
effect of stereotype threat on test performance, we computed a
series of regression equations as prescribed by Baron and Kenny
(1986). According to this approach, three relationships between
the target variables must be demonstrated to establish a basis for
testing mediation. The independent variable must predict both the
dependent and the mediator variable and the mediator must predict
the dependent variable. Once these conditions are established, the
dependent variable is regressed onto the independent variable and
mediator in a final regression analysis. Support for mediation is
obtained by demonstrating that the effect of the independent vari-
able (stereotype threat) on the dependent variable (math test per-
formance) is significantly reduced when accounting for the effect
of the hypothesized mediator (working memory capacity). The
results of these analyses are shown in Figure 3.
As reported above, stereotype threat had a significant negative
effect on women’s working memory capacity (the mediator),
␤
⫽
⫺.52, t(26) ⫽⫺3.13, p ⬍ .01, and math test performance (the
dependent variable),
␤
⫽⫺.42, t(26) ⫽⫺2.38, p ⬍ .03. A third
regression analysis established that working memory capacity was
a significant predictor of accuracy on the math test,
␤
⫽ .64,
t(26) ⫽ 4.28, p ⬍ .001. Finally, when performance on the math test
was regressed onto both stereotype threat and working memory
capacity, stereotype threat was no longer a significant predictor of
math test performance,
␤
⫽⫺.12, t(26) ⫽⫺.66, p ⬎ .50, whereas
working memory capacity remained significant in the equation,
␤
⫽ .58, t(26) ⫽ 3.26, p ⬍ .01. A Sobel test of the reduction in the
direct stereotype threat effect was significant (Z ⫽ 2.26, p ⬍ .02),
providing support for our hypothesis that stereotype threat inter-
feres with women’s math test performance by reducing their
working memory capacity.
To rule out other explanations for these relationships (e.g.,
model misspecification), we also conducted a reverse mediation
analysis with math test performance serving as the mediator and
working memory serving as the dependent variable. In contrast to
the primary mediation analysis, stereotype threat remained a mar-
ginally significant predictor of working memory when controlling
for math test performance,
␤
⫽⫺.31, t(26) ⫽⫺1.94, p ⫽ .06. A
Sobel test confirmed that the reduction in the stereotype threat
effect when controlling for performance on the math test was not
significant (Z ⫽ 1.51, p ⬎ .10). In sum, these analyses provide
greater support for the hypothesis that working memory mediates
effects on math test performance than for an alternative model in
which math test performance mediates stereotype threat effects on
working memory capacity.
General Discussion
Previous stereotype threat research has predominantly exam-
ined affective variables that are thought to account for the
performance decrements observed by individuals who face the
possibility of confirming a negative stereotype (e.g., Spencer et
al., 1999). Relatively less work has examined the effects that
stereotype threat manipulations might have on cognitive pro-
cessing per se (see Quinn & Spencer, 2001, for an exception).
The experiments reported here were designed to test the hy-
pothesis that stereotype threat reduces an individual’s perfor-
mance on a complex cognitive test because it reduces the
individual’s working memory capacity. Results of three exper-
iments provide support for this hypothesis. Experiments 1 and 2
demonstrated that manipulations of stereotype threat led to
lower working memory scores among individuals who are tar-
geted by the stereotype (women and Latinos) but had no effect
on those who are not targeted by the stereotype (men and
Whites). Results of the third experiment reveal that the reduc-
tions in working memory capacity observed under stereotype
threat mediate the reductions in performance on a standardized
test. Taken together, these findings suggest that members of
stigmatized groups perform poorly on cognitive tests when
negative stereotypes have been primed because this added in-
formation interferes with their attentional resources.
Although the pattern of results on the primary variable of
interest (i.e., working memory capacity) was quite consistent
across three studies using different groups, different manipula-
tions, and different measures, the patterns of data on participants’
self-reported experiences were more variable. For example, al-
Figure 3. Experiment 3: Tests of working memory capacity as a mediator
of stereotype threat effects on women’s math test performance. *p ⬍ .05.
**p ⬍ .01. ***p ⬍ .001.
449
STEREOTYPE THREAT AND WORKING MEMORY

though Latinos in Experiment 2 reported feeling more anxious
under conditions of stereotype threat, the same pattern was not
observed among women in Experiments 1 or 3. Even though
Latinos’ reports of anxiety paralleled their mean differences for
working memory capacity, there was no correlation between these
two measures. Similarly, women reported more gender identity
threat in the stereotype threat condition in Experiment 1 but not in
Experiment 3. The inconsistency of self-reported experiences
paired with consistent performance decrements suggests that a
conscious awareness of stereotype threat may not be a necessary
component of the phenomenon. In practical terms, the apparent
dissociation between processing effects and the conscious experi-
ence of anxiety or identity threat implies that most individuals who
suffer the deleterious effects of stereotype threat are not necessar-
ily aware of their predicament. Although they might not experi-
ence a test as more difficult or become more anxious than other
nonstigmatized students, the salience of negative stereotypes might
nevertheless consume critical cognitive resources and lead to a
poor performance that is unrepresentative of their true abilities.
Without awareness that a problem exists, individuals under stereo-
type threat are unable to do anything to circumvent these effects.
Of course, this concern highlights a need for future research to
investigate the strategies that an individual might use to combat the
effects of stereotype threat.
We want to state explicitly that our emphasis on the cognitive
deficits associated with stereotype threat is not meant to imply
that negative affect does not contribute to the effect of stereo-
type threat on performance. As Steele et al. (2002) have noted
recently, stereotype threat appears to be a multifaceted experi-
ence that does not lend itself to one phenomenological expla-
nation. Furthermore, it would be difficult to argue in light of
research from both the stereotype threat literature and the
working memory literature that anxiety does not play a role in
interfering with performance on cognitive tasks. However, we
would suggest that future research use more novel approaches
to measuring affective processes that do not rely on self-report
measures exclusively. Our findings highlight the need to ap-
proach the “threat” from a more implicit angle given that
stereotype threat does not appear to produce consciously re-
portable affective states on a consistent basis or even lowered
expectations of performance.
The findings from these experiments not only advance our
understanding of stereotype threat, but represent an advance for
research on working memory capacity as well. As stated earlier,
working memory capacity has been examined exclusively as an
individual-difference measure that has been assumed to be rela-
tively stable over time. Our demonstration that manipulations of
stereotype threat can produce situational reductions in working
memory underscores the need for researchers to use caution when
assuming that such measures index a stable and unchanging abil-
ity. Furthermore, these studies offer an important new method-
ological tool for measuring the effects of situational manipulations
on working memory capacity. Presumably the working memory
measures that we used in our studies could be used in other social
psychological research areas (e.g., research on persuasion and
stereotype application) that posit a role for cognitive capacity in
mediating judgments.
Processes That Might Deplete Cognitive Resources Under
Stereotype Threat
Although these studies constitute a direct examination of the
hypothesis that stereotype threat reduces cognitive capacity, future
research will be needed to determine exactly where attentional
capacity is diverted when stereotype threat occurs. There are
several possible processes that might be examined. One possibility
is that cognitive capacity is being consumed by the attempt to
suppress negative group stereotypes. Current conceptualizations of
working memory capacity assume that this cognitive system is
intimately involved in directing attention to the task at hand while
suppressing intrusive or irrelevant information (Engle, 2001;
Rosen & Engle, 1998). For example, Klein and Boals (2001)
posited that individuals high in life stress have lower levels of
working memory capacity because they are chronically trying to
suppress unwanted negative thoughts and feelings that they have.
Thus, one can imagine that once group stereotypes have been
activated by the task description in our experiments, some amount
of working memory capacity is being devoted to screen out this
information during the task.
Consistent with these ideas, recent evidence confirms that con-
structs associated with female stereotypes are activated among
women during stereotype threat and that the accessibility of these
constructs partially mediates their lowered performance on a math
test (Davies et al., 2002). Furthermore, Spencer (2003) has con-
ducted research concurrent with our own that suggests that women
might try to suppress these stereotypic constructs. For example, he
has found that adding further cognitive load to women who are
already experiencing stereotype threat leads to a heightened acti-
vation of stereotype related constructs, supporting the notion that
the load interfered with their attempts to suppress this information.
He also showed that giving women instruction to replace stereo-
typic thoughts that they might have during the test with less
threatening thoughts eliminates the negative effects of stereotype
threat on performance. What is remarkable is that these replace-
ment thoughts can be something self-affirming, such as an impor-
tant social identity, or something quite irrelevant, such as a red
Volkswagen.
Another mechanism to explore in future research is the role of
heightened arousal in impairing working memory. Given research
showing that participants under stereotype threat show heightened
physiological arousal (Blascovich et al., 2001) as well as the
finding that stress impairs working memory (e.g., Klein & Boals,
2001), it seems quite reasonable to assume that physiological
changes occurring during stressful situations, in general, and under
stereotype threat, more specifically, impair cognitive functioning.
Still, more work is needed to understand the specific processes by
which this happens. In one promising study on this topic, Croizet,
Despre`s, Gauzins, Huguet, and Leyens (2003) have examined
heart-rate variability as a physiological variable that is predictive
of cognitive resources (i.e., higher variability indicates higher
resources). Their research shows that both stigmatized and non-
stigmatized individuals show decreased heart-rate variability when
told that a test is a diagnostic measure of their intelligence.
However, these changes in heart-rate variability predict and me-
diate test performance only among stigmatized individuals. Al-
though these results represent an important step forward in linking
physiology and cognition, still more research is needed that inte-
450
SCHMADER AND JOHNS

grates the affective, cognitive, and physiological processes under-
lying the experience of stereotype threat.
Conclusions
The primary purpose of the research reported here is to advance
our knowledge of the ways in which negative stereotypes exert
their influence on the individuals they target. Following the work
of Steele et al., we approached this issue from the perspective that
negative social stereotypes can create an added psychological
burden in situations where one’s behavior might be interpreted as
evidence for the validity of such belief systems. Across three
studies, we provide converging evidence that performing under the
specter of a negative stereotype can deplete the cognitive resources
of stigmatized group members and impair performance on chal-
lenging academic tasks. These findings represent an important
advance in understanding exactly how stereotypes work to under-
mine the talents and abilities of individuals who might otherwise
meet their full potential in high-stakes performance situations,
such as college entrance exams. Beyond simply advancing our
theoretical knowledge, however, we hope these findings also meet
their full potential and contribute to broader efforts aimed at
developing strategies to combat the deleterious effects of stereo-
type threat on performance.
References
Aronson, J., Lustina, M. J., Good, C., Keough, K., Steele, C. M., & Brown,
J. (1999). When White men can’t do math: Necessary and sufficient
factors in stereotype threat. Journal of Experimental Social Psychol-
ogy, 35, 29–46.
Ashcraft, M. H. (2002, October). Math anxiety: Personal, educational, and
cognitive consequences. Current Directions in Psychological Sci-
ence, 11, 181–185.
Ashcraft, M. H., & Kirk, E. P. (2001). The relationships among working
memory, math anxiety, and performance. Journal of Experimental Psy-
chology: General, 130, 224–237.
Baddeley, A. D., & Hitch, G. (1974). Working memory. In G. A. Bower
(Ed.), The psychology of learning and motivation (Vol. 8, pp. 47–89).
New York: Academic Press.
Baron, R. M., & Kenny, D. A. (1986). The moderator–mediator variable
distinction in social psychological research: Conceptual, strategic, and
statistical considerations. Journal of Personality and Social Psychol-
ogy, 51, 1173–1182.
Blascovich, J., Spencer, S. J., Quinn, D., & Steele, C. (2001). African
Americans and high blood pressure: The role of stereotype threat.
Psychological Science, 12, 225–229.
Conway, A. R. A., Cowan, N., & Bunting, M. F. (2001). The cocktail party
phenomenon revisited: The importance of working memory capacity.
Psychonomic Bulletin & Review, 8, 331–335.
Croizet, J. C., Despre`s, G., Gauzins, M., Huguet, P., & Leyens, J. (2003).
Stereotype threat undermines intellectual performance by triggering a
disruptive mental load. Unpublished manuscript, Universite´ Blaise Pas-
cal, Clermont-Ferrand, France.
Davies, P. G., Spencer, S. J., Quinn, D. M., & Gerhardstein, R. (2002).
Consuming images: How television commercials that elicit stereotype
threat can restrain women academically and professionally. Personality
and Social Psychology Bulletin, 28, 1615–1628.
Derakshan, N., & Eysenck, M. W. (1998). Working memory capacity in
high trait-anxious and repressor groups. Cognition and Emotion, 12,
697–713.
Engle, R. W. (2001). What is working memory capacity? In H. L. Roediger
III & J. S. Nairne (Eds.), The nature of remembering: Essays in honor
of Robert G. Crowder (pp. 297–314). Washington, DC: American Psy-
chological Association.
Engle, R. W., Tuholski, S. W., Laughlin, J. E., & Conway, A. R. A. (1999).
Working memory, short-term memory, and general fluid intelligence: A
latent variable approach. Journal of Experimental Psychology: General,
128, 309–331.
Eysenck, M. W., & Calvo, M. G. (1992). Anxiety and performance: The
processing efficiency theory. Cognition and Emotion, 6, 409–434.
Gonzales, P. M., Blanton, H., & Williams, K. J. (2002). The effects of
stereotype threat and double-minority status on the test performance of
Latino women. Personality and Social Psychology Bulletin, 28, 659–
670.
Inzlicht, M., & Ben-Zeev, T. (2000). A threatening intellectual environ-
ment: Why women are susceptible to experience problem-solving defi-
cits in the presence of men. Psychological Science, 11, 365–371.
Klein, K., & Boals, A. (2001). The relationship of life event stress and
working memory capacity. Applied Cognitive Psychology, 15, 565–579.
Klein, K., & Fiss, W. H. (1999). The reliability and stability of the Turner
and Engle working memory task. Behavior Research Methods, Instru-
ments, and Computers, 31, 429–432.
La Pointe, L. B., & Engle, R. W. (1990). Simple and complex word spans
as measures of working memory capacity. Journal of Experimental
Psychology: Learning, Memory, and Cognition, 16, 1118–1133.
Marx, D. M., & Roman, J. S. (2002). Female role models: Protecting
women’s math test performance. Personality and Social Psychology
Bulletin, 28, 1183–1193.
Martens, A., Johns, M., Greenberg, J., & Schimel, J. (2002). Combating
stereotype threat: The effects of self-affirmation and emotional expres-
sion on women’s math performance. Unpublished manuscript, Univer-
sity of Arizona.
Oberauer, K., Su¨
␤
, H. M., Schulze, R., Wilhelm, O., & Wittman, W. W.
(2000). Working memory capacity – Facets of a cognitive ability con-
struct. Personality and Individual Differences, 29, 1017–1045.
Quinn, D. M., & Spencer, S. J. (2001). The interference of stereotype threat
with women’s generation of mathematical problem-solving strategies.
Journal of Social Issues, 57, 55–71.
Rosen, V. M., & Engle, R. W. (1998). Working memory capacity and
suppression. Journal of Memory and Language, 39, 418–436.
Schmader, T. (2002). Gender identification moderates stereotype threat
effects on women’s math performance. Journal of Experimental Social
Psychology, 38, 194–201.
Schmader, T., & Johns, M. (2003). Suspicious minds: Do beliefs about
gender stereotypes moderate stereotype threat effects? Unpublished
manuscript, University of Arizona.
Sekaquaptewa, D., & Thompson, M. (2003). Solo status, stereotype threat
and performance expectancies: Their effects on women’s performance.
Journal of Experimental Social Psychology, 39, 68–74.
Sorg, B. A., & Whitney, P. (1992). The effect of trait anxiety and situa-
tional stress on working memory capacity. Journal of Research in
Personality, 26, 235–241.
Spencer, S. (2003, February). Media images and stereotype threat: How
activation of cultural stereotypes can undermine women’s math perfor-
mance. Paper presented at the annual meeting of the Society for Per-
sonality and Social Psychology, Universal City, CA.
Spencer, S. J., Steele, C. M., & Quinn, D. M. (1999). Stereotype threat and
women’s math performance. Journal of Experimental Social Psychol-
ogy, 35, 4–28.
Spielberger, C. D., Gorsuch, R. R., & Lushene, R. (1970). The State-Trait
Anxiety Inventory (STAI) test manual. Palo Alto, CA: Consulting Psy-
chologists Press.
Stangor, C., Carr, C., & Kiang, L. (1998). Activating stereotypes under-
mines task performance expectations. Journal of Personality and Social
Psychology, 75, 1191–1197.
451
STEREOTYPE THREAT AND WORKING MEMORY

Steele, C. M. (1997). A threat in the air: How stereotypes shape the
intellectual identities and performance. American Psychologist, 52, 613–
629.
Steele, C. M., & Aronson, J. (1995). Stereotype threat and the intellectual
test performance of African Americans. Journal of Personality and
Social Psychology, 69, 797–811.
Steele, C. M., Spencer, S. J., & Aronson, J. (2002). Contending with group
image: The psychology of stereotype and social identity threat. In M.
Zanna (Ed.), Advances in experimental social psychology (Vol. 34, pp.
379–440). New York: Academic Press.
Stone, J. (2002). Battling doubt by avoiding practice: The effects of
stereotype threat on self-handicapping in white athletes. Personality and
Social Psychology Bulletin, 28, 1667–1678.
Stone, J., Lynch, C. I., Sjomeling, M., & Darley, J. M. (1999). Stereotype
threat effects on Black and White athletic performance. Journal of
Personality and Social Psychology, 77, 1213–1227.
Turner, M. L., & Engle, R. W. (1989). Is working memory capacity task
dependent? Journal of Memory and Language, 28, 127–154.
Received May 21, 2002
Revision received May 13, 2003
Accepted May 14, 2003 䡲
452
SCHMADER AND JOHNS