Another question arises: Do the effects of stereotype threat come entirely from the fear of being stereotyped, or do they come from something internal to black students—self-doubt, for example?

Beginning with George Herbert Mead's idea of the "looking-glass self," social psychology has assumed that one's self-image derives in large part from how one is viewed by others—family, school, and the broader society. When those views are negative, people may internalize them, resulting in lower self-esteem—or self-hatred, as it has been called. This theory was first applied to the experience of Jews, by Sigmund Freud and Bruno Bettelheim, but it was also soon applied to the experience of African-Americans, by Gordon Allport, Frantz Fanon, Kenneth Clark, and others. According to the theory, black students internalize negative stereotypes as performance anxiety and low expectations for achievement, which they then fulfill. The "self-fulfilling prophecy" has become a commonplace about these students. Stereotype threat, however, is something different, something external: the situational threat of being negatively stereotyped. Which of these two processes, then, caused the results of our experiments?

Joshua Aronson, Michael Lustina, Kelli Keough, Joseph Brown, Catherine Good, and I devised a way to find out. Suppose we told white male students who were strong in math that a difficult math test they were about to take was one on which Asians generally did better than whites. White males should not have a sense of group inferiority about math, since no societal stereotype alleges such an inferiority. Yet this comment would put them under a form of stereotype threat: any faltering on the test could cause them to be seen negatively from the standpoint of the positive stereotype about Asians and math ability. If stereotype threat alone—in the absence of any internalized self-doubt—was capable of disrupting test performance, then white males taking the test after this comment should perform less well than white males taking the test without hearing the comment. That is just what happened. Stereotype threat impaired intellectual functioning in a group unlikely to have any sense of group inferiority.

In science, as in the rest of life, few things are definitive. But these results are pretty good evidence that stereotype threat's impairment of standardized-test performance does not depend on cueing a pre-existing anxiety. Steven Spencer, Diane Quinn, and I have shown how stereotype threat depresses the performance of accomplished female math students on a difficult math test, and how that performance improves dramatically when the threat is lifted. Jean-Claude Croizet, working in France with a stereotype that links poor verbal skills with lower-class status, found analogous results: lower-class college students performed less well than upper-class college students under the threat of a stereotype-based judgment, but performed as well when the threat was removed.

Is everyone equally threatened and disrupted by a stereotype? One might expect, for example, that it would affect the weakest students most. But in all our research the most achievement-oriented students, who were also the most skilled, motivated, and confident, were the most impaired by stereotype threat. This fact had been under our noses all along—in our data and even in our theory. A person has to care about a domain in order to be disturbed by the prospect of being stereotyped in it. That is the whole idea of disidentification—protecting against stereotype threat by ceasing to care about the domain in which the stereotype applies. Our earlier experiments had selected black students who identified with verbal skills and women who identified with math. But when we tested participants who identified less with these domains, what had been under our noses hit us in the face. None of them showed any effect of stereotype threat whatsoever.

These weakly identified students did not perform well on the test: once they discovered its difficulty, they stopped trying very hard and got a low score. But their performance did not differ depending on whether they felt they were at risk of being judged stereotypically.