Avon and Somerset Police closed a BME pass-rate gap by priming candidates before the test

Avon and Somerset Police's situational judgement test, part of its officer recruitment process, was producing a large pass-rate gap between Black and minority ethnic (BME) candidates and other candidates — roughly 30% versus roughly 65% by one account — even though the test's content had been reviewed and showed no overt bias in its questions or scoring. A test that looks fair on paper can still be shaped by psychological factors that have nothing to do with its content.

The Behavioural Insights Team, working with the force, hypothesized the gap was driven by stereotype threat — the well-documented effect where awareness of a negative stereotype about one's group can itself depress performance on a high-stakes evaluation, independent of actual ability. Standard fixes for test bias, like revising question wording or scoring rubrics, would do nothing to address a psychological effect happening before a candidate ever reads the first question.

Added self-affirming reflection questions to the pre-test email, priming candidates against stereotype threat, which closed the gap without changing the test.

Stereotype threat depresses performance before the test starts, so content fixes can't help.

Self-affirmation prompts buffer against anxiety by reaffirming candidates' competence and belonging.

The intervention targets the psychological state, not the test, so it doesn't lower standards.

It works only when the gap is psychological, not structural, so diagnosis is key.

When a fair-looking assessment still shows group gaps, look at the psychological state candidates bring in, not just the test's content; a small pre-assessment intervention can close the gap.

BME candidates who received the priming questions passed the situational judgement test at rates comparable to non-BME candidates, roughly closing a gap that had previously run from about 30% up to parity around 65%, without any change to the test's content, difficulty, or scoring.