530 Prejudice, Racism, and Discrimination
attained lower accuracy (i.e., number of items correct relative
to the number attempted) than did either Black or White par-
ticipants in the other conditions. Steele and Aronson’s third
experiment demonstrated that the diagnostic ability manipu-
lation elicited among Black participants who were expecting
to take a difficult test (but did not do so) the racial stereotype
of Blacks held by Whites as well as an avoidance of self-
characterization in terms of this stereotype, and even an
avoidance of indicating one’s racial status on a demographic
postquestionnaire, relative to nondiagnostic and control con-
ditions. In Study 4, priming race by merely having partici-
pants indicate their race on a demographic questionnaire
before attempting a challenging intellectual test served to in-
hibit performance by Black participants and presumably to
elicit stereotype threat. Steele and Aronson proposed that the
mechanism underlying the impact of stereotype threat on the
test performance of their Black American participants was
probably an inefficiency of cognitive processing, not unlike
that produced by other evaluative pressures.
Croizet and Claire (1998) extended the applicability of the
stereotype threat concept to those of low socioeconomic sta-
tus (SES) outside the United States. Using a predominantly
female sample of French university students, these investiga-
tors likewise found that under stereotype threat, students of
low SES obtained fewer correct answers, attempted fewer
items, and had lower overall accuracy on verbal GRE items.
By contrast, much like Steele and Aronson (1995) had previ-
ously found in comparing Black and White American partic-
ipants, there was no difference in test performance between
participants of low and high SES when the same test was
described as nondiagnostic of one’s intellectual ability. Vary-
ing the salience of SES before the test by having participants
indicate their parents’ occupation and educational level, how-
ever, had no effect in this study.
Recent Studies
Spencer, Steele, and Quinn (1999) themselves applied stereo-
type threat theory to U.S. women’s math performance in three
studies including math-oriented students who had taken cal-
culus and had performed highly on the high-school mathemat-
ics section of the SAT. Their first experiment demonstrated
that a gender difference, with women underperforming men,
occurred only when the math GRE items used to assess math
performance were difficult rather than easy. Spencer and his
colleagues varied stereotype threat in the next two studies by
informing participants either that there was a gender differ-
ence previously obtained with the math GRE items they were
to solve (threat condition) or not (no stereotype threat). In
the no threat condition, women’s performance on the math
GRE test equalled that of men. By contrast, in the threat
condition, women underperformed men. Finally, their third
experiment demonstrated that the stereotype threat effect
was obtainable at a state university in the U.S. whose
academic standards were less rigorous and selective than
the elite university samples in prior studies and further ex-
plored possible mediating processes. The mediational tests
excluded evaluation apprehension and self-efficacy as a basis
for the impact of stereotype threat on women’s math perfor-
mance. Anxiety emerged as a weak mediator of stereotype
threat.
Finally, recent studies by different sets of investigators
show that stereotype threat can affect the performance of
White majority group members and does not require that one
be a member of a historically stigmatized group. Aronson
et al. (1999) conducted two experiments in which White
students of high math ability at an elite U.S. university were
presented information that Asian Americans outperform
Whites in math (stereotype threat condition) or not (no threat
condition). Additionally, in the second study they selected
math-oriented students who scored on the bottom and top
tertiles of rated importance of mathematics ability to their
self-concept as a means of assessing low versus high identifi-
cation with this domain. Their first study showed that White
students performed less well on a challenging math test when
threatened with a racial stereotype indicating their inferiority
relative to Asians. Their second study showed that this
stereotype effect occurred only when the White students were
math-identified and that evaluation apprehension was a
weak, potential mediator.
Stone, Lynch, Sjomeling, and Darley (1999) took advan-
tage of a golf test that they presented to Black and White
Princeton University participants as indicating their “natural
athletic ability” or their “sports intelligence.” Their first study
showed that performance by Black students on the golf test
suffered more when it constituted a stereotype threat (an in-
dication of sports intelligence—a negative stereotype for
Whites) than when it did not (an indication of natural athletic
ability—a positive stereotype for Blacks). By contrast, for
White participants for whom the opposite was true (i.e.,
sports intelligence is a positive stereotype, and natural ath-
letic ability is a negative or less positive stereotype than for
Blacks), the reverse pattern was obtained, as predicted from
stereotype threat theory. Their second study, focusing on
White participants only, showed that the detrimental effects
of stereotype threat on performance on the golf test occurred
only for “engaged” participants for whom performance in the
athletic domain was important to their self-worth and not for
those who were “disengaged.” In addition to showing the im-
portance of engagement for the stereotype threat effect, their