Available online at www.sciencedirect.

com

Journal of Experimental Social Psychology 44 (2008) 818–825

www.elsevier.com/locate/jesp

q
Approaching math increases math = me and math = pleasant
a,*
Kerry Kawakami , Jennifer R. Steele a, Claudia Cifa a, Curtis E. Phills a, John F. Dovidio b

a
Department of Psychology, York University, 4700 Keele Street, Toronto, Canada M3J 1P3
b
Department of Psychology, Yale University, Box 208205, New Haven, CT, 06520-8205

Received 19 May 2006; revised 6 July 2007

Available online 8 August 2007

Abstract

In two studies, we examined the effect of extensive practice in approaching math on implicit identification with math, implicit math
attitudes, and behavior during a math test. The results from Study 1 demonstrated that women trained to approach math showed more
identification with and positive implicit attitudes toward math than women trained to avoid math. Notably, this latter pattern of findings
was only evident for women low in initial identification with this field. The results from Study 2 replicated these findings by showing that
women who were initially low in math identification and trained to approach math showed more implicit identification with math and
attempted more items on a math test than women trained to respond to math in a neutral way. The implications of these findings for
current theorizing on the gender gap in women’s representation in math related careers are discussed.
 2007 Elsevier Inc. All rights reserved.

Keywords: Implicit attitudes; Implicit prejudice; Math identification; Gender gap; Approach behaviors; Stereotype threat

So my best guess, to provoke you, of what’s behind all of 11% of the engineering workforce (Tietjen, 2004) and 8%
this is that the largest phenomenon, by far, is the general and 13% of tenured and tenure track positions in mathe-
clash between people’s legitimate family desires and matics (Ripley, 2005) and chemistry departments (Maras-
employers’ current desire for high power and high inten- co, 2005) at the top 50 research universities in the United
sity, that in the special case of science and engineering, States. This difference is unlikely to disappear in the near
there are issues of intrinsic aptitude, and particularly of future; men are currently over four times more likely to
variability of aptitude, and that those considerations are choose a major that is high in math content than women
reinforced by what are in fact lesser factors involving (Betz, 1997; Tietjen, 2004).
socialization and continuing discrimination. Remarks by In a recent conference focusing on the representation of
Lawrence H. Summers, President of Harvard University, women in math related fields within academia, then presi-
January 14, 2005. dent of Harvard University, Lawrence Summers (2005),
suggested two main causes for this gender gap. One reason
Gender disparity in the representation of men and was that women’s ambitions differed from men’s, with the
women in fields related to mathematics, engineering, the implication being that women were not as committed to
physical sciences, and information technology persists positions in science and mathematics. The second reason
(Cooper & Weaver, 2003; Crocker, Major, & Steele, was that men and women differed in their intrinsic aptitude
1998; Stangor & Sechrist, 1998). Women constitute only for science. Although President Summers acknowledged
the possibility that socialization and discrimination could
q
The research reported in this paper was supported by a Social Science play a role in creating this disparity, he argued that these
and Humanities Council of Canada (SSHRC) grant to the first author and factors were less influential than gender differences in apti-
by SSHRC post-doctoral and doctoral fellowships to the second and
fourth authors, respectively.
tude. In a subsequent critical review of the scientific data
*
Corresponding author. Fax: +1 416 736 5814. on sex differences in cognitive abilities, Spelke (2005) dis-
E-mail address: kawakami@yorku.ca (K. Kawakami). agreed with Summers, concluding that ‘‘research on the

0022-1031/$ - see front matter  2007 Elsevier Inc. All rights reserved.
doi:10.1016/j.jesp.2007.07.009
 K. Kawakami et al. / Journal of Experimental Social Psychology 44 (2008) 818–825 819

cognitive abilities of males and females from birth to matu- cation (Greenwald & Farnham, 2000; Markham & Brendl,
rity does not support the claim that men have a greater 2005; Nosek, Banaji, & Greenwald, 2002). Specifically, the
intrinsic aptitude for mathematics and science’’ (p. 956). present research employed a training procedure (Kawaka-
Rather than innate aptitude, Steele (1997) proposed that mi, Dovidio, Moll, Hermsen, & Russin, 2000; Kawakami,
an important factor accounting for the different gender rep- Dovidio, & Van Kamp, 2005, 2007) to examine whether
resentation in mathematics is identification with the practice in approaching a specific academic domain such
domain. Specifically, Steele suggested that many women as math can influence women’s implicit identification with
have difficulty identifying with fields associated with math- math (Studies 1 and 2), their implicit attitudes toward math
ematics because of inadequate support, few role models, (Study 1), and their behavior during a challenging math
and biased societal gender roles and stereotypes. He further test (Study 2).
argued that it is this disidentification with math that subse- Recent findings have demonstrated that individual dif-
quently impedes many women’s motivation, which leads to ferences in identification with mathematics can influence
their underperformance and, ultimately, to their underrep- the effectiveness of various types of interventions and expe-
resentation in math related fields (Davies, Spencer, Quinn, riences related to stereotype threat. These studies have
& Gerhardstein, 2002; Spencer, Steele, & Quinn, 1999; shown that women who are highly identified with math
Steele, 1992, 2003; Steele, James, & Barnett, 2002). By perform more poorly on a difficult math test when gender
focusing on a factor that can conceivably be changed, such stereotypes are made salient because of fear of confirming
as women’s identification with math rather than on a factor these associations in a domain that is relevant and impor-
that cannot be changed, such as innate ability, Steele and tant to them (Aronson et al., 1999; Cooper & Weaver,
his colleagues provide a useful strategy to narrow the gen- 2003; Spencer et al., 1999). In contrast, because math is
der gap—modifying women’s identification with math. not as important to their self-concept, low identified
Social cognitive research suggests that one possible way women feel less pressure to disconfirm stereotypes in math
to change women’s identification with math is to have them related situations and therefore are not influenced by ste-
‘‘approach’’ math (Seibt & Förster, 2004). At a basic level, reotype salience manipulations.
approach behaviors are often conceptualized as a frame of While recognizing the importance of studying stereo-
mind or motivation related to pulling something or some- type threat effects for women highly identified with
one toward one’s body (Förster, 2001). Research in this math, the primary focus of the present research was
area has provided consistent evidence that approach to examine the impact of an intervention aimed at
behaviors can influence attitudes toward mundane objects strengthening the identification and improving the atti-
and social categories in a predictable fashion. People gener- tudes of women who are initially low in identification
ally evaluate objects and categories more favorably follow- with math. Notwithstanding that the disidentification
ing approach, as opposed to avoidance, actions (Cacioppo, of this latter category may be due in part to stereotype
Priester, & Berntson, 1993; Förster & Strack, 1997; Priest- threat processes experienced early on in life, strategies
er, Cacioppo, & Petty, 1996). For example, recent research targeting this much larger group are critical to produc-
by Kawakami, Phills, Steele, and Dovidio (2007) found ing a society with a balanced representation of men and
that participants who were extensively trained to approach women in math.
Blacks by pulling a joystick toward themselves were less Based on previous research demonstrating that train-
prejudiced on an implicit attitude test than participants ing in responding in new ways to social categories
who were trained to avoid Blacks by pushing a joystick (e.g., nonstereotypic associations with skinheads,
away. approaching Blacks) positively impacts people’s orienta-
In the present research, we extended these earlier find- tion to these categories (Kawakami et al., 2000, 2005;
ings by examining the impact of approach behaviors Kawakami, Phills, et al., 2007), we predicted that
toward an academic domain. Whereas previous research approach training would positively impact women who
has focused on nonsocial objects and social categories, are low in math identification. Because this group has
we investigated whether approaching math concepts can less experience approaching math, lower initial levels of
influence attitudes toward that domain. Furthermore, the identification with math, and more negative attitudes
present studies also examined the novel idea that approach- towards this domain than women who are high in iden-
ing math can influence associations between math and the tification, we predicted that approach training in com-
self. Rather than limiting our examination of the effects of parison to control training conditions would increase
approach and avoidance to attitudes, two experiments the math identification of low identified women and
explored the possibility that approaching a specific concept improve their attitudes toward this domain.
can impact people’s identification with that concept. Unlike women low in math identification, high math
Because approach behaviors are related, both semantically identified women have presumably had extensive experi-
and behaviorally, with bringing stimuli closer to the self, ence in approaching math in high school and university
we predicted that training in approaching math by pulling math courses. We therefore expected that the training
this domain toward the self would increase association of would have less of an impact on their responses in this
this domain with the self, thereby increasing math identifi- domain (Kawakami et al., 2000, 2005; Kawakami, Phills,
820 K. Kawakami et al. / Journal of Experimental Social Psychology 44 (2008) 818–825

et al., 2007). Somewhat counter-intuitively, stereotype that women who were already highly identified with
threat research even suggests that interventions designed math would not be influenced by the training or would
to enhance the identification of group members who are show even lower implicit identification with and more
weakly identified can undermine group members already negative implicit attitudes toward math after training in
highly identified with the domain. Aronson and his col- approaching rather than avoiding math.
leagues (1999) propose that strong external pressure to
identify with a stereotype inconsistent domain can arouse Method
stereotype threat and undermine the intrinsic association
of people previously high in identification and thereby Participants and procedure
weaken the association of the self with the domain. Steele Forty-four female undergraduates participated in the
(1997) explains that ‘‘applying a strategy to school identi- experiment for course credit. Prior to the commencement
fied students (on the basis of their membership in a stereo- of the study, students were required to rate how much
type threatened group) that assumes weak identification, they liked math and how good they were at math on a
poor skills, and little confidence could backfire . . . [and] seven-point scale either via email or in person to deter-
increase stereotype threat and underperformance by signal- mine their identification with math.1 Twenty women
ing that their abilities are held under suspicion because of who circled four or higher on the scale to both questions
their group membership’’ (p. 624). (i.e., math identified) and 24 women who circled less
Based on this theorizing, we also examined the impact of than four on one of the questions (i.e., nonmath identi-
approaching math on women who were already highly fied) were invited to participate in the study (Ben-Zeev,
math identified. These women may be particularly vigilant Fein, & Inzlicht, 2005; Spencer et al., 1999). Upon their
to stereotypes, and so interventions that make stereotypes arrival, these women were led to individual cubicles and
salient can induce stereotype threat (Davies et al., 2002; informed that they would be involved in a series of sep-
Spencer et al., 1999). Because women in general are associ- arate studies. In reality, these students were randomly
ated with having difficulty in approaching and performing assigned to either an Approach or Avoid Math training
well in math, inconsistent behavior such as approaching task before being instructed to complete an attitude and
math will be threatening to women who are highly math identification IAT.
identified. Specifically, we expected that interventions that At the beginning of the approach training task, all partic-
encourage approaching rather than avoiding or responding ipants were informed that they would be presented on a
neutrally to math, would nullify or even impair any poten- computer screen with images related to math and the arts
tially positive effects of training on their identification with and that the experimenter was investigating their responses
mathematics. In summary, although we did not expect any to specific images to examine theories of cognitive pro-
positive effects of extensive practice in approaching math cesses. Participants in the Approach Math condition were
with highly math identified women, we hypothesized that instructed to approach math by pulling the joystick toward
this training would improve how women who are initially themselves (Chen & Bargh, 1999; Kawakami, Phills, et al.,
low in math identification evaluate math, associate the self 2007) when presented with math symbols and to avoid the
with math, and behave in math related activities. arts by pushing the joystick away from themselves when
presented with arts symbols. In contrast, participants in
Study 1 the Avoid Math condition were instructed to avoid math
by pushing the joystick away from themselves when pre-
In Study 1, participants were preselected on the basis sented with math symbols and to approach the arts by pull-
of their initial identification with math before being ran- ing the joystick toward themselves when presented with
domly assigned to an approach or avoidance math train- arts symbols.
ing condition. Following the training, all participants On each trial of the training task, the photograph
completed two Implicit Association Tests (IATs; Green- remained on the computer screen until the participant
wald & Farnham, 2000; Nosek et al., 2002) related to responded. If the response was correct, a blank screen
math attitudes and identification with math. We chose appeared for 1000 ms before the presentation of the next
to focus on implicit measures to avoid demand charac- photograph. If the response was incorrect, a blank screen
teristics and other momentary motivations related to appeared for 100 ms before a red ‘‘X’’ was presented for
the training procedure and the experimental context 800 ms followed by a blank screen for 100 ms before the
which may affect overt controllable responses (Dovidio next trial. In total, participants received 480 trials con-
& Fazio, 1992; Nosek et al., 2002). Our primary hypoth- sisting of 10 blocks of 48 trials. In each block, 24 images
esis was that women who were initially low in math iden- related to math (e.g., calculators, equations) and 24
tification would show higher implicit identification with images related to the arts (e.g., guitars, poetry) were pre-
and more positive implicit attitudes toward math after
training in approaching rather than avoiding math. 1
Although 46 students participated in Study 1, the data from one
Although the main target group in the present research student who did not follow instructions and one student whose mean
was women low in math identification, we also predicted responses were extreme outliers were excluded from the analyses.
 K. Kawakami et al. / Journal of Experimental Social Psychology 44 (2008) 818–825 821

sented in a random order. Participants were instructed to

complete each trial as quickly and as accurately as pos-
sible. Before beginning the actual trials, participants were
first presented with eight practice trials involving separate
stimuli and after each block participants were given a
short break.
The Attitude IAT was used to measure implicit atti-
tudes toward math (Nosek et al., 2002). In this task, par-
ticipants were instructed to categorize six words related
to math (e.g., algebra, formula), the arts (e.g., music,
arts), positive concepts (e.g., happy, rainbow), and nega-
tive concepts (e.g., vomit, hate). One set of critical trials
instructed participants to press the same key when cate- Fig. 1. Effects of approach training and initial math identification on
gorizing arts concepts and positive words and when cat- women’s implicit attitudes toward math.
egorizing math concepts and negative words. The other
set of critical trials instructed participants to press the (see Nosek et al., 2002). Higher scores reflected a more
same key when categorizing math concepts and positive positive attitude toward math relative to the arts.2
words and when categorizing arts concepts and negative To examine the effect of training on implicit attitudes
words. toward math, a 2 (Training: Approach Math vs. Avoid
The Identification IAT was used next to measure implicit Math) · 2 (Initial Math Identification: High vs. Low)
identification with math (Greenwald & Farnham, 2000; ANOVA was performed on attitude IAT scores. A signifi-
Nosek et al., 2002). The stimuli included in this task were cant two-way interaction was found, F(1, 40) = 4.78,
different from the attitude IAT. In this task, participants p = .04, see Fig. 1. As predicted, low math identified
were instructed to categorize four concepts related to math women were faster to associate positive concepts with math
(e.g., calculus, equations), the arts (e.g., dance, literature), after being trained to approach (M = 96 ms) than avoid
the self (e.g., me, myself), and others (e.g., they, other). (M = 268 ms) math, t(22) = 2.81, p = .01. In contrast,
One set of critical trials instructed participants to press there was no difference in the speed of associating positive
the same key when categorizing arts concepts and self concepts with math between high math identified women
words and when categorizing math concepts and others trained to approach (M = 144 ms) and to avoid
words. The other set of critical trials instructed participants (M = 128 ms) math, t(18) = .53, p = .60.
to press the same key when categorizing math concepts and
self words and when categorizing arts concepts and others Identification IAT
words. Each critical block in both the attitude and identifi- Before analyzing the data related to the attitude IAT,
cation IATs consisted of 72 trials and the order of the crit- response latencies in which participants gave incorrect
ical blocks for both IATs was counterbalanced across answers (5.0%) were excluded and latencies that were less
participants. than 300 ms and more than 2000 ms (1.2%) were recoded.
Difference scores, based on logarithmically transformed
data, between the two critical blocks were computed by
Results and discussion
2
Although analyses of the attitude and identification IAT data using an
Attitude IAT alternative scoring algorithm (Greenwald, Nosek, & Banaji, 2003; Nosek,
Before analyzing the data related to the attitude IAT, Greenwald, & Banaji, 2005), which was tested with data collected online
response latencies in which participants gave incorrect and in a less controlled environment and may be less sensitive to variations
answers (6.2%) were excluded and cutoff latencies (as deter- (including actual changes in implicit associations as a function of
experimental manipulations), showed minor differences in significance
mined by the percentage of trials) were used to reduce out- levels in several analyses, the overall pattern of findings was similar. The
lier effects (Ratcliff, 1993). Specifically, response latencies analyses with the alternative algorithm produced a marginally significant
that were less than 300 ms and more than 2000 ms (2.5%) two-way, Type of Training · Initial Math Identification interaction on the
were identified as outliers and recoded to 300 and 2000. attitude IAT, F(1, 40) = 2.56, p = .11. As expected, low identified women
Although all analyses were performed on logarithmically had more positive implicit attitudes toward math after being trained to
approach (D = .26) than avoid (D = .70) math, t(22) = 2.25, p = .03.
transformed data, for illustrative purposes the untrans- Attitudes for high math-identified women were the same after training in
formed means are presented. The IAT effects were then approaching and avoiding math (Ds = .36), t(18) = .001, p = 1.00. In
computed by creating difference scores between the two addition, the two-way interaction related to the identification IAT utilizing
critical blocks. Specifically, mean response latencies when this alternative algorithm was significant, F(1, 40) = 8.61, p = .006.
Whereas low math-identified women were faster to associate the self with
arts concepts were paired with negative words and math
math after being trained to approach (D = .06) than avoid (D = .21)
concepts were paired with positive words were subtracted math, t(22) = 2.03, p = .05, high math-identified women were slower to
from latencies when arts concepts were paired with positive associate the self with math after approaching (D = .17) than avoiding
words and math concepts were paired with negative words (D = .12) math, t(18) = 2.15, p = .05.
822 K. Kawakami et al. / Journal of Experimental Social Psychology 44 (2008) 818–825

only on women who were not highly identified with math.

A further goal of Study 2 was to extend the previous find-
ings by examining the influence of approach behavior on
math performance. In particular, after women were trained
in Study 2 to either approach math or respond to math in a
neutral way, they were presented with the identification
IAT and a challenging math quiz. Based on recent findings
by Davies et al. (2002) that measured women’s motivation
in math by the number of items they attempted, we
expected that approach training would not only increase
identification for low math identified women but would
also increase the number of questions they attempted on
Fig. 2. Effects of approach training and initial math identification on
a difficult math test.
women’s implicit identification with math.
Method
subtracting response latencies when arts concepts were Participants and procedure
paired with others and math concepts were paired with Fifty-six female undergraduates participated in the
the self from latencies when arts concepts were paired with experiment for course credit or a movie pass.3 Because
the self and math concepts were paired with others (see we wanted to focus in this study on the large majority of
Nosek et al., 2002). Higher scores reflected greater identifi- women, we excluded only those students who were very
cation with math relative to the arts. highly identified with math. Specifically, all participants
To examine the effect of training on implicit identifica- whose mean response was less than six on a nine-point
tion with math, a 2 (Training: Approach Math vs. Avoid scale designed to assess how pleasant they found five differ-
Math) · 2 (Initial Math Identification: High vs. Low) ent math related activities (e.g., doing an algebra problem
ANOVA was performed on identification IAT scores. As set, solving an equation, taking a calculus exam) were
depicted in Fig. 2, a significant two-way interaction was included in the study. Upon arrival, these participants were
found on the identification IAT, F(1, 40) = 8.09, p = .007. led to individual cubicles and were informed that they
As predicted, low math identified women tended to be fas- would be involved in a series of separate studies that were
ter in associating the self with math after training in unrelated. In reality, after being initially randomly assigned
approaching (M = 9 ms) than avoiding (M = 64 ms) to either an Approach Math or a Sideways Control train-
math, t(22) = 1.98, p = .06. In contrast, high math identi- ing task, these participants completed an identification
fied women were slower to associate the self with math IAT followed by a math quiz.
after approaching (M = 50) than avoiding (M = 29) Although the stimulus material and the procedure in the
math, t(18) = 2.07, p = .05. Approach Math training condition in Study 2 were the
As hypothesized, the results from Study 1 demonstrated same as in Study 1, the control training condition differed.
that training women who were initially low in math identi- Rather than having participants avoid math and approach
fication to approach math improved their negative atti- arts, participants in the control condition in Study 2 were
tudes toward and implicit identification with math. This presented with a more neutral task. Specifically, while half
intervention was therefore successful in changing the asso- of the participants in the Sideways Control condition were
ciations of math with the self and with positive concepts for instructed to push the joystick to the right when presented
our main target group, low math identified women. with math symbols and to the left when presented with arts
Approach training, however, did not influence attitudes symbols, the other half of the control participants were told
and adversely impacted identification for women who were to move the joystick to the left and right, respectively, when
initially highly math identified. These latter findings sup- presented with math and arts symbols.
port theorizing related to stereotype threat processes which After completing the training, all participants were pre-
suggests that people already highly identified with a stereo- sented with the same Identification IAT used in Study 1.
type inconsistent domain may be vigilant to stereotype Each participant was subsequently presented with a diffi-
diagnostic situations and further reinforces the importance cult math quiz which consisted of 12 multiple choice math
of ‘‘rendering onto the right students the right interven- questions [e.g., If f(v) = 4v2 2v 1, then f(1/2) equals
tion’’ (Steele, 1997, p. 624). (a) 1 (b) 1 (c) 0 (3) ½ or (3) 2]. Participants were
instructed to answer as many questions correctly as possi-
Study 2 ble within a 20 minute time frame (Shih, Pittinsky, & Amb-

In accordance with the philosophy that strategies aimed 3

Although 58 students participated in Study 2, the data from two
at improving women’s identification with math should be students who did not follow instructions were excluded from the analyses.
specific to the target group (Steele, 1997), Study 2 focused An additional two participants did not complete the math quiz.
 K. Kawakami et al. / Journal of Experimental Social Psychology 44 (2008) 818–825 823

ady, 1999) and were allowed the use of paper for this task General discussion
but not calculators.
Many theorists have noted that the self-concept is a
Results and discussion social entity—it changes according to the people and stim-
uli in its environment (Anderson & Chen, 2002; Kawakami
Identification IAT et al., submitted for publication; Markus & Kunda, 1986).
Before analyzing the data related to the identification Recent research has demonstrated that the self can be mea-
IAT, response latencies in which participants gave incor- sured at both an explicit and implicit level with implicit
rect answers (4.7%) were excluded and outlier latencies that self-concepts reflecting the various types of information
were less than 300 ms or more than 2000 ms (2.0%) were that we associate with the self in memory (Greenwald &
recoded. Difference scores of the logarithmically trans- Farnham, 2000). The primary goal of the present research
formed data from the two critical blocks were computed was to examine whether extensive training in approaching
with higher scores reflecting greater identification with math could influence associations between the self and
math relative to the arts.4 math.
To examine the effect of training in approaching math The results from two studies demonstrate that training
on implicit identification with math, a Training (Approach women who are initially low in math identification to
Math vs. Sideways Control) t-test was performed on the approach math can increase their implicit identification
identification IAT scores. As predicted, women who were with this domain. Furthermore, the findings also show that
trained to approach math by pulling a joystick toward this type of training can improve implicit attitudes toward
themselves were faster to associate the self with math math (Study 1) and increase the number of questions
(M = 57 ms) than women in a neutral control condition attempted on a difficult math test (Study 2). Not surpris-
who were trained to push a joystick sideways to math sym- ingly, approach training did not immediately enhance over-
bols (M = 22 ms), t(54) = 2.01, p = .05. all math performance. Although this prospect needs to be
examined in future research, it is conceivable that with sus-
Math quiz tained effort over an extended period, this strategy could
To examine the effect of training in approaching math on ultimately lead to improved mathematical ability and moti-
participants’ responses on the math quiz, a Training vation (Davies et al., 2002; Steele, 1997).
(Approach Math vs. Sideways Control) t-test was performed Together these findings have significant implications for
on the number of questions that the participants answered decreasing the gender gap. If women’s lack of participation
correctly and the number of questions that participants in math related careers and majors at university is primar-
attempted (both correct and incorrect answers), separately. ily due to factors such as identification with and attitudes
As expected, although approaching math did not increase toward this domain (Crocker et al., 1998; Halpern, Wai,
the total number of correct answers, it did impact the number & Saw, 2005; Steele, 1997), rather than intrinsic ability
of questions attempted. Specifically, low math identified (Summers, 2005), the present research suggests that one
women in both the Approach Math (M = 3.72) and the Side- way to address this issue may be to change the general
ways Control (M = 3.24) conditions answered a similarly avoidance orientation that many women have toward
low number of questions correctly, t(52) = .89, p = .38. math. Our results suggest a direct way of modifying low
However, also as predicted, women who were trained to identification with an academic domain is to provide
approach math attempted more questions (M = 10.31) than women with extensive training in associating the self with
women who were trained to respond in a neutral way to math math. These findings, however also suggest that other indi-
(M = 8.72), t(52) = 2.12, p = .04. rect ways of inducing women to ‘‘approach math’’ such as
In summary, the results from Study 2 replicate the find- more female friendly teaching formats, positive female role
ings in Study 1 by demonstrating that training women who models, segregated math classes for women, and the reduc-
are low in initial math identification to systematically tion of negative stereotypes related to women and math in
approach math can increase their implicit identification the media and in our culture in general (Cooper & Weaver,
with this domain. Furthermore, the results also show that 2003; Steele, 1997; Steele, Reisz, Williams, & Kawakami,
whereas training in approaching math did not influence 2007) may also be effective in changing math identification.
these women’s performance on a math test, it did increase The present findings indicate that for the women in our
the number of challenging math questions attempted. study who were not strongly identified with math, a ‘‘wise
strategy’’ to increase positive attitudes, positive identifica-
tion, and more active participation is to approach math
(Steele, 1997). However, this type of intervention appears
4
Analyses of the identification IAT using the alternative scoring to have an adverse effect for women who were already iden-
algorithm (Greenwald et al., 2003) also demonstrated that women who
tified with math, presumably because the introduction of
were low in initial identification with math were faster to associate the self
with math after being trained to approach math (D = .17) than after being
an intervention designed to make women ‘‘approach’’
trained to push a joystick sideways to math (D = .06), t(54) = 2.09, math made the gender stereotype salient. Whereas in earlier
p = .04. research it seemed necessary to make skills assessments
824 K. Kawakami et al. / Journal of Experimental Social Psychology 44 (2008) 818–825

seem reliable and diagnostic and to make the participants’ Dovidio, J. F., & Fazio, R. H. (1992). New technologies for the direct and
group identity salient to induce stereotype threat (Aronson indirect assessment of attitudes. In J. Tanur (Ed.), Questions about
survey questions: Meaning, memory, attitudes, and social interaction
et al., 1999; Steele, 1997), more recent research has demon- (pp. 204–237). New York: Russell Sage Foundation.
strated that simply making the stereotype salient is suffi- Förster, J. (2001). Success/failure feedback, expectancies, and approach/
cient (Davies et al., 2002). Moreover, our research avoidance motivation: How regulatory focus moderates classic rela-
suggests that the specific way a stereotype is made salient tions. Journal of Experimental Social Psychology, 37, 253–260.
is particularly critical for highly math identified women. Förster, J., & Strack, F. (1997). Motor actions in retrieval of valenced
information: A motor congruence effect. Perceptual and Motor Skills,
Our research shows that for these women approaching 85, 1419–1427.
math had adverse effects but avoiding math did not, even Greenwald, A. G., & Farnham, S. D. (2000). Using the Implicit
though both of these actions could conceivably make ste- Association Test to measure self esteem and self-concept. Journal of
reotypes salient. Future research might therefore produc- Personality and Social Psychology, 79, 1022–1038.
tively examine the differential effects of various Greenwald, A. G., Nosek, B. A., & Banaji, M. R. (2003). Understanding
and using the Implicit Association Test: I. An improved scoring
interventions that highlight behaviors that are inconsistent algorithm. Journal of Personality and Social Psychology, 85, 197–216.
in comparison to consistent with the group stereotype. Halpern, D., Wai, J., & Saw, A. (2005). A psychobiosocial model: Why
Whereas stereotype inconsistent behavior makes salient females are sometimes greater than and sometimes less than males in
the group’s inability to perform well in a domain and math achievement. In A. M. Gallagher & J. C. Kaufman (Eds.),
induces stereotype threat, stereotype consistent behavior Gender differences in mathematics (pp. 48–72). New York: Cambridge
University Press.
mainly underlines the negative behaviors related to the Kawakami, K., Dovidio, J. F., Moll, J., Hermsen, S., & Russin, A. (2000).
group. It may not be the concern of confirming the stereo- Just say no (to stereotyping): Effects of training in negation of
type that leads to stereotype threat but the pressure of dis- stereotypic associations on stereotype activation. Journal of Personal-
confirming the stereotype. ity and Social Psychology, 78, 871–888.
Overall, our results suggest the exciting prospect that Kawakami, K., Dovidio, J. F., & Van Kamp, S. (2005). Kicking the habit:
Effects of nonstereotypic association training and correction processes
shifting women’s orientation may be one key strategy to on hiring decisions. Journal of Experimental Social Psychology, 41,
decrease the destructive consequences of the current gender 68–75.
gap in mathematics. As compared with a generation ago Kawakami, K., Dovidio, J. F., & Van Kamp, S. (2007). The impact of
when many more men than women majored in biology, naı̈ve theories related to strategies to reduce biases and correction
medicine, or economics, or a century ago when colleges processes on the application of stereotypes. Group Processes and
Intergroup Relations, 10, 141–158.
were almost exclusively filled with men (Spelke, 2005), Kawakami, K., Phills, C., Greenwald, A. G., Freed, G., Simard, D., &
the findings from these two studies suggest the real possibil- Mills, J. (submitted for publication). The real chameleon: Assimilating
ity that the present disparity in math can also be the self to the social environment.
eliminated. 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.
References Journal of Personality and Social Psychology, 92, 957–971.
Marasco, C. A. (2005). Women faculty makes little progress. Chemical &
Anderson, S. M., & Chen, S. (2002). The relational self: An interpersonal Engineering News, 83, 38.
social-cognitive theory. Psychological Review, 109, 619–645. Markham, A. B., & Brendl, M. (2005). Constraining theories of embodied
Aronson, J., Lustina, M. J., Good, C., Keough, K., Steele, C. M., & cognition. Psychological Science, 16, 6–10.
Brown, J. (1999). When White men can’t do math: Necessary and Markus, H. R., & Kunda, Z. (1986). Stability and malleability of the
sufficient factors in stereotype threat. Journal of Experimental Social self-concept. Journal of Personality and Social Psychology, 51,
Psychology, 35, 29–46. 858–866.
Ben-Zeev, T., Fein, S., & Inzlicht, M. (2005). Arousal and stereotype Nosek, B. A., Banaji, M. R., & Greenwald, A. G. (2002). Math = male,
threat. Journal of Experimental Social Psychology, 41, 174–181. me = female, therefore math „ me. Journal of Personality and Social
Betz, N. (1997). What stops women and minorities from choosing and Psychology, 83, 44–59.
completing majors in science and engineering? In D. Johnson (Ed.), Nosek, B. A., Greenwald, A. G., & Banaji, M. R. (2005). Understanding
Minorities and girls in school (pp. 105–131). Thousand Oaks, CA: Sage. and using the Implicit Association Test: G. Method variables and
Cacioppo, J. T., Priester, J. R., & Berntson, G. G. (1993). Rudimentary construct validity. Personality and Social Psychology Bulletin, 31,
determinants of attitudes: II. Arm flexion and extension have 166–180.
differential effects on attitudes. Journal of Personality and Social Priester, J. R., Cacioppo, J. T., & Petty, R. E. (1996). The influence of
Psychology, 65, 5–17. motor processes on attitudes toward novel versus familiar semantic
Chen, M., & Bargh, J. A. (1999). Consequences of automatic evaluation: stimuli. Personality and Social Psychology Bulletin, 22, 442–447.
Immediate behavioral predispositions to approach or avoid the Ratcliff, R. (1993). Methods for dealing with response time outliers.
stimulus. Personality and Social Psychology Bulletin, 25, 215–224. Psychological Bulletin, 114, 510–532.
Cooper, J., & Weaver, K. D. (2003). Gender and Computers: Understand- Ripley, A. (2005). Who says a woman can’t be Einstein? Time, 35–44.
ing the digital divide. Mahwah, N.J.: Erlbaum. Seibt, B., & Förster, J. (2004). Stereotype threat and performance: How
Crocker, J., Major, B., & Steele, C. M. (1998). Social stigma (4th ed.. In D. self-stereotypes influence processing by inducing regulatory foci.
T. Gilbert, S. T. Fiske, & G. Lindzey (Eds.). The handbook of social Journal of Personality and Social Psychology, 87, 38–56.
psychology (Vol. 2, pp. 504–553). Boston, MA: McGraw-Hill. Shih, M., Pittinsky, T. L., & Ambady, N. (1999). Stereotype susceptibility:
Davies, P. G., Spencer, S. J., Quinn, D. M., & Gerhardstein, R. (2002). Identity salience and shifts in quantitative performance. Psychological
Consuming images: How television commercials that elicit stereotype Science, 10, 81–84.
threat can restrain women academically and professionally. Personality Spelke, E. S. (2005). Sex differences in intrinsic aptitude for mathematics
and Social Psychology Bulletin, 28, 1615–1628. and science? A critical review. American Psychologist, 60, 950–958.
 K. Kawakami et al. / Journal of Experimental Social Psychology 44 (2008) 818–825 825

Spencer, S. J., Steele, C. M., & Quinn, D. (1999). Stereotype threat and Steele, J., James, J., & Barnett, R. (2002). Learning in a man’s world:
women’s math performance. Journal of Experimental Social Psychol- Examining the perceptions of undergraduate women in male-
ogy, 35, 4–28. dominated academic areas. Psychology of Women Quarterly, 26,
Stangor, C., & Sechrist, G. B. (1998). Conceptualizing the determinants of 46–50.
academic choice and task performance across social groups. In J. K. Steele, J. R., Reisz, L., Williams, A., & Kawakami, K. (2007). Women in
Swim & C. Stangor (Eds.), Prejudice: The target’s perspective mathematics: Examining the hidden barriers that gender stereotypes
(pp. 105–124). San Diego, CA: Academic Press. can impose. In R. J. Burke & M. C. Mattis (Eds.), Minorities in science,
Steele, C. M. (1992). Race and the schooling of Black Americans. The technology, engineering and mathematics: Upping the numbers
Atlantic Monthly, 68–78. (pp. 159–183). Northhampton, MA: Edward Elgar.
Steele, C. M. (1997). A threat in the air: How stereotypes shape intellectual Summers, L. (2005). Remarks at NBER conference on diversifying
identity and performance. American Psychologist, 52, 613–629. the science and engineering workforce. Retrieved March 25,
Steele, J. (2003). Children’s gender stereotypes about math: The role of 2006 from http://www.president.harvard.edu/speeches/2005/
stereotype stratification. Journal of Applied Social Psychology, 33, nber.html.
2587–2606. Tietjen, J. S. (2004). Why so few women, still? IEEE Spectrum, 41, 57–58.