Stereotype threat

type: effect · origin: study · evidence: failed to replicate · themes: how we judge others

Claim assessed: Being reminded of a negative stereotype about your group lowers your test results.

Published 6 October 2026, updated 7 October 2026 · how we work

In short

Stereotype threat is the idea that people from a group with a negative stereotype do worse on a test when the stereotype is brought up, because they fear confirming it. It would mean that part of the score differences between groups comes not from ability but from the testing situation. The idea was very influential, made it into textbooks and was even cited in court cases about university admissions. Recent large, preregistered replications, however, no longer found the effect.

What it says

The classic studies. Steele and Aronson (1995) reported that Black students scored lower on a difficult test when it was presented as a measure of intellectual ability. Spencer et al. (1999) reported a similar effect in women, in maths. We haven’t read them; we describe them from the replication report below.

The replicated study. Johns, Schmader and Martens (2005) gave students a maths test. For some, the test was presented as a way to establish different norms for men and women; for others, as a problem-solving task, or they were told in advance that anxiety might come from the stereotype rather than a lack of ability. Women scored lower only in the first version. The study has been cited hundreds of times. We haven’t read it; we describe it from the replication report.

The meta-analysis on girls. Flore and Wicherts (2015) gathered 47 effects from studies of girls under 18, in maths, science and spatial skills. They found a small to moderate average effect (d = −0.22), but also several signs of selective publication that could seriously distort the literature. They proposed a large replication.

Johns et al. (2005) and the replication by Stoevenbelt et al. (2025), as reported in the replication report. Flore and Wicherts (2015), from the abstract. The measures aren't identical; compare the direction and order of magnitude.

The replication in nine labs. Stoevenbelt, Flore, Wicherts and their collaborators (2025) repeated Johns et al.’s study in nine labs in the Netherlands, Austria, Germany, Slovakia and the US, with 1,709 participants, following a plan registered in advance. Women scored somewhat lower than men on average, but that didn’t depend on how the test was presented. The threat effect in women was practically zero (g = 0.04), compared with d = −0.82 in the original study, which had 45 women. The effect didn’t appear in the American labs either. The authors note that the result fits other preregistered replications that didn’t find the effect.

A real-world test. Inglis and O’Hagan (2022) tested, at a UK maths competition, the idea that the gender question should come at the end of the test rather than the start, so as not to remind girls of the stereotype. With 1,169 students and the analysis registered in advance, the position of the question changed nothing.

Example

Before a maths test, a teacher says: “Let’s see whether the girls do as well as the boys.” The theory predicts that the girls will score lower because of this remark. Large replications suggest that the effect, if it exists, is much smaller than was thought. The example is an illustration, not a study cited here.

How to apply it

The steps below are a practical way to apply it that we propose, based on the studies cited.

Don’t explain group differences by stereotype threat. Current evidence doesn’t support it explaining an important part of these differences.
Still avoid remarks about stereotypes at tests. Even if the measured effect is close to zero, there is no good reason to make them.
Don’t base policies on simple fixes like “move the gender question”. Inglis and O’Hagan found no effect of this change.
When you hear of a spectacular effect, ask about preregistered replications. This case resembles priming: large effects in small studies, almost zero in large ones.

Limits and nuances

“Didn’t replicate” doesn’t mean “doesn’t exist”. The nine-lab replication covered only the US and Western and Central Europe, and a single kind of test. The authors note that the test may not have been difficult enough.
The gender gap remained. In the replication, women scored somewhat lower on average, in all conditions. Only the stereotype-threat explanation wasn’t supported.
The replication is still a preprint. The 2025 report is publicly available, but we didn’t find a journal version.
Many studies concern other groups and other domains. We followed mainly the case of women and maths, the best studied. For other groups, the evidence may differ.
What we read. The replication report by Stoevenbelt et al. (2025), from the Open Science Framework, in its May 2025 version (the abstract of the November 2025 version has the same numbers), and Inglis and O’Hagan (2022), from PLOS ONE, we read in full. For Flore and Wicherts (2015) we read the abstract. We haven’t read Steele and Aronson (1995), Spencer et al. (1999) or Johns et al. (2005); we describe them from the replication report. A large replication with Dutch students, by Flore et al. (2018), was blocked in the university repository and we didn’t use it directly.

Sources

Claude M. Steele, Joshua Aronson (1995). Stereotype threat and the intellectual test performance of African Americans
Michael Johns, Toni Schmader, Andy Martens (2005). Knowing is half the battle: Teaching stereotype threat as a means of improving women's math performance
Paulette C. Flore, Jelte M. Wicherts (2015). Does stereotype threat influence performance of girls in stereotyped domains? A meta-analysis
Matthew Inglis, Steven O'Hagan (2022). Stereotype threat, gender and mathematics attainment: A conceptual replication of Stricker & Ward
Andrea H. Stoevenbelt, Paulette C. Flore, Jelte M. Wicherts et al. (2025). Registered replication report: Johns, Schmader, & Martens (2005)