Pronin Steele Ross.pdf

Journal of

Experimental

Social Psychology

Journal of Experimental Social Psychology 40 (2004) 152–168

www.elsevier.com/locate/jesp

Identity bifurcation in response to stereotype threat:

Women and mathematics q

Emily Pronin, * Claude M. Steele, and Lee Ross

Department of Psychology, Stanford University, USA

Received 1 April 2002; revised 31 March 2003

Abstract

Three studies explored womens bifurcation of feminine identity as a response to threatening stereotypes in the domain of

mathematics. Study 1 demonstrated that women in a math class who previously had completed a large number of math courses

disavowed ‘‘feminine characteristics’’ strongly associated with stereotypes about womens potential for math success (e.g., ﬂirta-

tiousness, planning to have children) but not characteristics perceived to be weakly associated with these stereotypes (e.g., empathy,

nurturance), more than women who had completed fewer math courses. Studies 2 and 3 directly manipulated stereotype threat by

presenting a scientiﬁc article reporting stereotype-consistent sex diﬀerences in math aptitude. As predicted, women strongly iden-

tiﬁed with mathematics responded to this threatening article by disavowing feminine characteristics strongly associated, but not

those weakly associated, with the relevant negative stereotypes, while women not strongly identiﬁed with mathematics showed no

such diﬀerentiation in response. Theoretical and practical implications of these results are discussed.

Perseverance in the face of group-based stereotypes

about ones limitations poses a daunting challenge. Be-

yond enduring negative expectations and discourage-

ment from others, members of the stereotyped group

may respond to inevitable disappointments and di-

culties by questioning their own ﬁtness and acceptance

in the social environment. And they may be further

burdened by the knowledge that individual failures will

reinforce the negative views and assumptions held about

their group. The psychological threat associated with the

awareness that one may be viewed through the lens of a

negative stereotype has been termed stereotype threat

(Steele, 1997; Steele & Aronson, 1995).

Such threat can be pernicious in its eﬀects. In par-

ticular, there is evidence that it can impair performance

in a way that serves to validate the relevant stereotype.

Adverse eﬀects of stereotype threat on individual per-

formance have been demonstrated for female students

undergoing tests of mathematical ability (Brown & Jo-

sephs, 1999; Davies, Spencer, Quinn, & Gerhradstein,

2001; Inzlicht & Ben-Zeev, 2000; Shih, Pittinsky, &

Ambady, 2000; Spencer, Steele, & Quinn, 1999). Similar

eﬀects have been found among black students under-

going standardized academic testing (Steele & Aronson,

1995), students from low socioeconomic backgrounds

undergoing purported ‘‘intellectual ability’’ testing

(Croizet & Claire, 1998), and even white men undergo-

ing a purported test of ‘‘natural athletic ability’’ (Stone,

Lynch, Sjomeling, & Darley, 1999). Moreover, there is

disturbing evidence that the negative consequences in

question are apt to be strongest among the very people

who have invested their self-worth in, tied their future

prospects to, or otherwise become strongly identiﬁed

with the relevant domain of achievement (Aronson,

Lustina, Keough, Steele, & Brown, 1999; Steele, 1997).

Previous researchers have explored a number of ego-

protective responses to stereotype threat. One well-

documented response is ‘‘disengagement’’ (Crocker,

Major, & Steele, 1998; Major, Spencer, Schmader,

Wolfe, & Crocker, 1998) or ‘‘disidentiﬁcation’’ (Spencer

et al., 1999; Steele, 1997; Steele & Aronson, 1995) with

respect to the domain in question. That is, the individual

q Research reported in this article was supported by grants from

the National Institute of Mental Health and the Russell Sage

Foundation. We gratefully acknowledge Christina Fuhrman, Kather-

ine Jakle, Peggy Lo, Dina Pugh, and Diana Tellefson for their help in

conducting this research.

Corresponding author. Current address: Psychology Department,

Princeton University, Princeton, NJ 08544, USA.

E-mail address: epronin@princeton.edu (E. Pronin).

doi:10.1016/S0022-1031(03)00088-X

E. Pronin et al. / Journal of Experimental Social Psychology 40 (2004) 152–168

153

excludes performance in that domain as a basis for self-

evaluation, and may reject it as a basis of respect for

people in general (Crocker & Major, 1989). Indeed, the

individual may even foster an identity ‘‘oppositional’’ to

success in that domain (Ogbu, 1986).

Such ego-protective strategies are costly, both for the

individual and for his or her group. They are especially

costly when the domain in question is relevant to an

important avenue for professional or personal ad-

vancement, and when the individual has previously

demonstrated some aptitude or enjoyed some success in it.

Fortunately, disidentiﬁcation is apt to be an ego-protec-

tive strategy of last resort. Rather than forfeit ones prior

‘‘investment’’ (not only of ego, but of time, training and

other resources as well), the individual in question is likely

to continue to strive for achievement in the domain and

seek alternative ways to deﬂect the relevant threat.

One such alternative involves disidentiﬁcation not

from the domain of achievement in which the threat is

posed, but from the group targeted by the stereotype.

Again, this strategy can impose a heavy cost. Disiden-

tiﬁcation with ones in-group and assimilation into some

out-group may oblige one to abandon previously valued

aspects of identity and sources of self-esteem — including

aspects that are not ‘‘suspect’’ with respect to achieve-

ment in the relevant domain.

The present research focuses on a subtler and less

costly response to stereotype threat. Rather than dis-

identifying with a valued domain of achievement or

disidentifying globally with ones in-group, one can

disidentify selectively — that is, disidentify with the as-

pects of ones in-group that are linked to disparagement

in that domain, while continuing to identify with valued

in-group characteristics that are not seen as linked to

such disparagement. We term this subtler response,

which might serve to deﬂect the relevant threat, identity

bifurcation. It is likely to occur when the beneﬁts of such

domain identiﬁcation in terms of the individuals sense

of self outweigh the costs of disavowing or abandoning

particular in-group characteristics — especially while the

individual is present in the valued domain.

Our present contentions regarding identity bifurcation

derive in part from the literature on identity-oriented re-

sponses to group prejudice, primarily developed by social

identity theorists (Tajfel & Turner, 1986; also Ellemers,

1993; Hogg & Abrams, 1988). We combine two separate

principles described in this literature — ﬁrst, that individ-

uals aremotivated tomaintain group identities as a source

of positive self-evaluation, and second that individuals

are motivated to shed group identities that threaten po-

sitive self-evaluation. Our argument is essentially that

individuals tend to see their group identity not as an in-

divisible whole, but as a set of separable attributes (e.g.,

Biernat, Vescio, & Green, 1996; Reid & Deaux, 1996).

This ‘‘separability’’ allows the two motives we have noted

to be served simultaneously. That is, the individual re-

sponds to the threat of negative self-evaluation by dis-

identifying with ascribed group attributes that are viewed

as relevant to the negative stereotypes about the group (in

a particular domain), while maintaining or even

strengthening identiﬁcation with attributes seen as ‘‘ste-

reotype-irrelevant.’’

To focus our immediate concern with math identiﬁ-

cation and gender, consider the plight of a young woman

(whom well call Carol) who wins admission to an elite

college on the basis of the math and quantitative science

grades she achieved in high school, the recommendations

of her teachers, and her performance on standardized

tests. Once in college, she inevitably experiences the same

pressures faced by all prospective math majors who en-

counter more rigorous courses and stiﬀer competition

than they had faced in high school. However, Carol faces

additional burdens. She becomes increasingly aware that

characteristics commonly associated with her gender

(ranging from concern with hair and makeup to willing-

ness to consider future career interruptions in order to

have children) make her prospects as a successful math

major and futuremathematician suspect to her classmates

and to her professors. She soon recognizes that the in-

terest she evidences in her personal appearance and her

frankness in acknowledging the challenges of balancing

family and career concerns are seen as ‘‘counter-diag-

nostic’’ with respect to success in math-related ﬁelds.

When Carol walks into the math classroom, or into

her math professors oce hours, or chats with her math

study group, she feels heightened anxiety about the

prospect of negative judgment due to these aspects of

her female identity. To reduce this discomfort and

anxiety, she stops wearing makeup — at least on days

when she has to meet with her professor or is likely to be

scrutinized by her male peers. When ‘‘hanging out’’ and

chatting with friends who belong to her math study

group, she avoids the topic of having children. However,

even at these moments, she continues to value her

overall identity as a woman, and to see herself as the

possessor of valued but stereotypically feminine char-

acteristics (such as social sensitivity) that she neither sees

nor expects her math peers to see as counter-diagnostic

of math success. Moreover, outside of such settings,

when her math ability is less under scrutiny and she is

less subject to stereotype threat, she may identify freely

with all of the feminine traits that she values, and even

with ones that she does not particularly value but nev-

ertheless acknowledges as an aspect of herself.

Gender stereotypes and mathematics

The present research focuses on female students re-

sponse to experiences in which they are exposed to the

threatening stereotype that women have less mathe-

matical ability than men. (Fully 45% of students in an

154

E. Pronin et al. / Journal of Experimental Social Psychology 40 (2004) 152–168

introductory psychology class at Stanford, we should

note, reported holding this belief.) The dampening eﬀect

of this stereotype on womens career aspirations and

choices, and thus on the representation of women both

in advanced quantitative courses and in professional

endeavors that demand quantitative skills (see Eccles,

1987) is all too apparent. For example, a report on the

faculty composition of the School of Science at the elite

Massachusetts Institute of Technology (1999) indicated

that, as of 1994, the faculty included 22 women, and 252

men. The reasons for such under-representation are of

course a subject for much debate, but in terms of our

present concerns it is important to note that women who

‘‘drop out’’ of quantitative majors do not tend to have

lower scores on college entrance exams, or lower fresh-

man grade-point averages than their male peers (e.g.,

Felder, Felder, Mauney, & Dietz, 1994; Seymour, 1995).

Womens experiences and peer interactions as math

students, and their sense of academic and personal

identity, thus merit close attention.

It is among women who major in quantitative ﬁelds or

who are otherwise ‘‘highly identiﬁed’’ with mathematics

(and who thus stand to be adversely aﬀected by negative

evaluations) that we most expect to ﬁnd identity bifur-

cation. It should be emphasized that we do not predict

that such math-identiﬁed women will adopt more mas-

culine traits in these contexts. What the identity bifurca-

tion hypothesis suggests is that women experiencing

stereotype threat in math will selectively disidentify

with stereotypically feminine traits that are linked to

negative stereotypes about their math ability or belonging

in the domain. Thus, we expect women under stereotype

threat to disidentify with characteristics viewed as in-

congruent with or compromising of math achievement

and belonging (e.g., wearing makeup) but not with other

characteristics viewed as less problematic for such

achievement and belonging (e.g., being empathic).

In focusing on the challenges faced by women in

quantitative ﬁelds, we are not, of course, suggesting that

women do not experience stereotype threat in other areas

of academic or professional achievement. Indeed, it is

apparent that in a variety of highly competitive achieve-

ment domains, from surgery (American College of Sur-

geons, 1996) to investment banking (The New York

Times, 1999), women ﬁnd themselves in a distinct nu-

merical minority, and accordingly face negative stereo-

types and sources of stereotype threat that are likely to

evoke similar types of identity bifurcation. Our current

research is thus designed to oﬀer an illustration of a

phenomenon that we believe is widespread in education

and in the workplace.

relying on our own intuitions about the relevant ste-

reotypes, we began with a preliminary study to identify

stereotypically feminine characteristics that our research

participants would view as strongly associated versus

weakly (or not at all) associated with negative stereo-

types about womens ability and potential in math. In

Study 1 we tested whether these trait classiﬁcations were

reﬂected in the diﬀering self-descriptions, obtained spe-

ciﬁcally in the context of math class, of women who had

completed many versus few math courses. We then

conducted two laboratory studies to test the hypothesis

that experimentally induced stereotype threat would

increase identity bifurcation among women strong in

their identiﬁcation with math.

Identiﬁcation and validation of feminine characteristics

Our primary ethnographic source of information

about the relevant stereotypes for women in math was

Seymour and Hewitts (1997) extensive interviews with

women who were pursuing (or who had ‘‘dropped out’’

of) math-intensive academic majors. We also conducted

two 6-person focus groups at Stanford — one with wo-

men majoring in, and one with women doing graduate

work in, quantitative areas. Both sources revealed fem-

inine characteristics viewed as relevant to the stereotypes

about womens math ability and potential for success.

One female engineering major, interviewed by Sey-

mour & Hewitt (S&H), described eﬀorts to avoid a

stereotypically feminine appearance : ‘‘Theres no way Id

wear pantyhose, or mascara, or throw my hair up — like

when I go to my art class’’ (p. 249). A Ph.D. student

from our focus groups echoed this concern, saying, ‘‘I

gave up wearing skirts [amongst my colleagues].’’ An-

other woman (S&H, p. 251) pointed out the danger of

ﬂirting : ‘‘We get irritated with women who think theyre

gonna get by because they ﬂirt with the faculty.’’ In our

focus groups, a graduate student expressed a similar

sentiment, saying she ‘‘would never ﬂirt’’ in the context

of her quantitative classes. The perils of indulging in

gossip were also suggested by a student who said: ‘‘They

[male peers] are more intelligent and more serious,

whereas the women are more gossipy and chatty’’ (S&H,

p. 288). This concern was raised in our focus groups by

an undergraduate who reported preferring to do prob-

lem sets with men rather than women since ‘‘Guys dont

idle chit-chat.’’ Emotionality was seen as yet another

feminine trait that would raise suspicion — perhaps be-

cause of its semantic opposition to rationality, a pre-

sumed hallmark of math aptitude. One graduate student

in our focus groups commented that ‘‘being emotional’’

would jeopardize her career eﬀorts, noting speciﬁcally:

‘‘Tears do not earn you respect.’’ Perhaps most dis-

turbing was perceived conﬂict between the pursuit of

career success and expressed desire to have children (and

to spend time raising them). An engineering major (S&H,

Overview of studies

The present research sought to investigate identity

bifurcation in response to stereotype threat. Rather than

E. Pronin et al. / Journal of Experimental Social Psychology 40 (2004) 152–168

155

p. 291) recounted that her engineering professor ‘‘didnt

think that women could have a job and raise a family at

the same time’’ (see also Etzkowitz, Kemelgor, Neus-

chatz, & Uzzi, 1992). A female graduate student from

our focus groups repeated this concern, claiming that a

woman would be negatively judged in her ﬁeld for trying

to lead a ‘‘balanced life’’ involving work and family.

from work to raise children (M ¼ 4 : 62). The remaining

items, classiﬁed as low in stereotype relevance, were

sensitivity (M ¼ 3 : 75), being fashion conscious (M ¼

3 : 08), nurturance (M ¼ 2 : 94), and empathy (M ¼ 2 : 93).

Our empirical validation thus supported our a priori

analysis, and the resulting set of high stereotype-relevant

characteristics was, of course, seen as signiﬁcantly more

likely to put a women at risk for negative judgment in

quantitative domains (M ¼ 4 : 97) than the set of char-

acteristics categorized as less relevant (M ¼ 3 : 18),

t ð 41 Þ¼ 12 : 39, p <: 0001. There was also a marginal

tendency (p ¼ : 08) for our English-class participants to

believe that the various feminine characteristics were, on

the whole, somewhat more likely to lead a woman to be

negatively judged in mathematics ﬁelds (M ¼ 4 : 46) than

did our math-class participants (M ¼ 4 : 06), but there

was no interaction eﬀect between participant class (En-

glish vs. math) and characteristic type (high vs.

Preliminary study

To create the speciﬁc list of stereotypically feminine

characteristics that would be used in our studies, we

subjected a list of characteristics suggested by the Sey-

mour and Hewitt study and our focus group work to

empirical validation with an additional set of female

undergraduates.

Method

Female undergraduates completed a survey immedi-

ately after attending one of two intermediate level math

classes (n ¼ 22), or one of two English literature classes

(n ¼ 20). (They each received a large chocolate bar as a

reward for their eﬀorts.) The survey called for ratings of

several ‘‘feminine’’ characteristics that the Seymour and

Hewitt study and our focus group work suggested were

incongruent with or seen as likely to compromise wo-

mens prospects for success in math-related ﬁelds (e.g.,

ﬂirtatiousness, tendency to gossip), and also some char-

acteristics that those sources had not cited and were

presumably less relevant to or less incongruent with such

success (e.g., sensitivity, nurturance). Participants were

asked to ‘‘rate how much you think each of these

characteristics would put a woman at risk for being

negatively judged in quantitatively based ﬁelds and/or

careers’’ (1 ¼ Low risk for being negatively judged,

7 ¼ High risk for being negatively judged).

low

stereotype relevant), F < 1.

Our eﬀorts to identify and validate two sets of femi-

nine characteristics that diﬀered in the degree to which

they pose a threat to prospects for success and positive

regard in math-related ﬁelds were thus successful. We

cannot claim, and do not imagine, that our modest

undertaking exhausted the relevant categories. There are

no doubt other ‘‘feminine’’ traits that raters would deem

incongruent with success in math endeavors, and many

that they would deem irrelevant to such success. Our

present contention is simply that our two sets of char-

acteristics provided us with the means to test our bi-

furcation hypothesis in the studies that we now report.

It is worth emphasizing that none of the characteristics

on either our ‘‘high’’ or ‘‘low’’ stereotype-relevant lists

can reasonably be thought to attenuate math skills in any

direct fashion. Rather, the qualities on the former list, but

not those on the latter, seem incongruent with the image,

and thus the identity, of math majors. Thus, it is the

qualities on the former list from which we predict that

math-identiﬁed women will disidentify — regardless of

whether they (or others) believe that these characteristics

actually compromise ability or harm performance.

With this empirically derived dichotomy of feminine

characteristics in hand, we next sought to determine

whether women who had taken more math classes than

their peers would describe themselves in ways that con-

ﬁrmed the psychological signiﬁcance of this dichotomy.

Results and discussion

In establishing our two lists of feminine characteris-

tics for subsequent use, we used the midpoint of the

relevant 7-point rating scale as our criterion. Charac-

teristics that received mean ratings above the midpoint

(i.e., those assessed to have moderate or high probability

of causing a woman to be negatively stereotyped) were

categorized as high in stereotype relevance, and char-

acteristics that received mean ratings below the mid-

point (i.e., those assessed to have low or no probability

of producing those consequences) were categorized as

low in stereotype relevance.

In all cases, our raters assessments proved consistent

with the evidence from the Seymour and Hewitt study

and our own focus groups. The items thus classiﬁed as

high in stereotype relevance included the wearing of

makeup (M ¼ 5 : 01), ﬂirtatiousness (M ¼ 5 : 08), being

gossipy (M ¼ 5 : 51), emotionality (M ¼ 5 : 55), wanting

children (M ¼ 4 : 02), and planning to spend time away

Study 1

Our ﬁrst study simply sought to determine whether

the pattern of identity bifurcation we postulated would

be more apparent among women attending math class

who had spent a lot of time in quantitative courses, and

presumably been heavily exposed to the relevant sources

of stereotype threat, than among women attending the

156

E. Pronin et al. / Journal of Experimental Social Psychology 40 (2004) 152–168

same classes who had spent less time in such courses and

presumably been less heavily exposed to the relevant

stereotype threat. While the results of such a study

would be correlational, and thus could not prove that

the predicted pattern of identiﬁcation was caused by

stereotype threat, we anticipated that our results would

serve both to pave the way for our later experiments and

to oﬀer external validity for the division of high vs. low

stereotype-relevance characteristics that we would rely

on in those experiments.

In this study, math students rated their identiﬁcation

with various characteristics, including gender-stereo-

typic ones. We deliberately had them complete the sur-

vey in math class, since we deemed it especially likely

that the relevant pattern of selective disidentiﬁcation

would be present in that context.

The ‘‘feminine’’ characteristics classiﬁed as high in

stereotype relevance (in italics below) or low in stereo-

type relevance (underlined below) were interspersed

among various ‘‘masculine’’ and ‘‘gender-neutral’’ items

in a larger survey, titled ‘‘Survey of Stanford Students.’’

The survey consisted of ﬁve sections, each with its own

heading (and separate items): (1) Personality (competi-

tive, sensitive , adaptable, ﬂirtatious, adventurous, emo-

tional, friendly, nurturing , analytical), (2) Interpersonal

Style (extroverted, assertive, gossipy, empathic , aggres-

sive, talk about politics, talk about relationships,

reliable), (3) Appearance (physically ﬁt, fashionable ,

laid-back, style hair, wear makeup), (4) Activities

(hanging out with friends, playing sports, shopping,

reading, engaging in artistic activities, talking on the

telephone), and (5) Future Plans (graduate school, full-

time job, marriage, having children, leaving work to raise

children, having a satisfying career). 1

Gender and course exposure information were col-

lected on a page entitled ‘‘Background Information’’

appended to the survey. On this page, participants indi-

cated their year in school, gender, age, ethnicity, US or

foreign citizenship, and current or intended major. They

then listed the quantitative courses they had taken in high

school and college (including mathematics at the pre-

calculus level or beyond, as well as courses in the physical

sciences, computer science, engineering, and statistics).

Method

Participants

Participants were 51 female (and an additional 103

male) Stanford undergraduates in 7 math classes (92% of

those attending). Two of the classes were the ﬁnal course

in an introductory-level 3-course sequence, 3 were the ﬁ-

nal course in an intermediate-level 3-course sequence, and

2 were single term intermediate-level classes.

Procedure

At the end of his or her lecture, the professor an-

nounced that students were being asked to stay for a few

minutes after class to complete a survey. The experi-

menter then introduced herself as a graduate student in

the department of psychology, told students that their

help in completing the survey would be appreciated (and

compensated with a chocolate bar), oﬀered to answer

any questions about the survey once responses were

collected, and distributed the survey.

Results

The median number of quantitative courses taken by

the women in our sample was 10. Accordingly, women

were classiﬁed as high versus low in their exposure to

math depending on whether they reported having taken

more than 10 quantitative courses (n ¼ 24) or 10 or

fewer quantitative courses (n ¼ 27). The same criterion

was used for dividing the men in our sample into high

(n ¼ 72) and low (n ¼ 31) math exposure groups.

Our main prediction was that women who had

completed a relatively large number of math-intensive

classes would show the predicted pattern of bifurcated

gender identity to a greater degree than women who had

complete fewer courses, while we expected no such dif-

ferentiation among men. Thus, we predicted a 2 (math

class exposure: high vs. low) 2 (type of feminine

characteristic: high stereotype relevance vs. low stereo-

Measures

Participants ratings of their identiﬁcation with speciﬁc

‘‘feminine’’ characteristics constituted our dependent

measure. Items were modeled after the self-schema mea-

sure developed by Markus (1977). That is, participants

ﬁrst rated how much each characteristic applied to them,

and then how important having that characteristic was to

their sense of self. Consistent with methodology devel-

oped by Markus (see also Brown, 1998), these two re-

sponses were averaged together to obtain a single mean

representation of the extent to which a given character-

istic was part of the individuals ‘‘self-schema,’’ or iden-

tity. For example, they ﬁrst were asked, ‘‘How sensitive

are you?’’ (1 ¼ not at all sensitive,7 ¼ very sensitive), and

then, ‘‘How important is being sensitive to you?’’ (1 ¼ not

at all important,7 ¼ very important). These two types of

responses, as in earlier studies, showed a high degree of

internal consistency, Cronbachs a ¼ : 91.

1 This study was conducted before the study that served to validate

the composites of more vs. less ‘‘stereotype relevant’’ feminine

characteristics. At the time it was conducted, we anticipated that

talking about relationships, shopping, and talking on the telephone

would be viewed as non-relevant characteristics, and that styling one ’ s

hair would be viewed as relevant. Because our validation study did not

support our a priori classiﬁcation of these items, they are excluded

from the analyses reported here and from our subsequent studies.

(Including these items, it should be noted, would not change the

statistical signiﬁcance of the relevant ﬁndings.)

Plik z chomika:

Inne pliki z tego folderu:

Inne foldery tego chomika: