Stereotype threat—the argument that the reason blacks or other Groups of Concern score worse on tests is because of stereotypes that they score worse on tests—has been a wildly popular concept since the 1990s. It has proven fairly easy in psych lab experiments to get college student volunteers of the chosen varieties to not score well on zero-stakes tests by hinting to them that they are expected not to do well. To fulfill their professors' desires for a publishable result, all they have to do is slack off instead of work hard on the meaningless test, and college students are good at slacking off.
On the other hand, it would be unethical to try to drive down the scores of members of Groups of Concern on high-stakes tests, so there is very little experimental data on whether stereotype threat actually exists on high-stakes tests, which is what everybody cares about. So, Walters, Lee, and Trapani did a study in for ETS in 2004 looking at various factors that had been alleged in experimental studies to cause Stereotype Threat using real data from the high-stakes GRE. This is about as close as anybody can come ethically to studying Stereotype Threat on a high-stakes test.
The study investigated the applicability of previous experimental research on stereotype threat to operational Graduate Record Examinations® (GRE®) General Test testing centers. The goal was to document any relationships between features of the testing environment that might cue stereotype threat as well as any impact on GRE test scores among African American, Hispanic, Asian American, and female test-takers. Among such features were the gender and ethnicity of test proctors and more general factors, such as the size, activity level, and social atmosphere of test centers. Our analyses revealed several relationships among environmental factors and several variations in test performance for all groups. However, we found no direct support for stereotype threat and, in fact, found some effects for proctor ethnicity that ran counter to a stereotype-threat explanation.