Tracing the origins of teacher gender bias

This paper examines whether teachers form gendered expectations that deviate from students’ actual academic performance. We exploit a unique institutional setting in England, where university admissions rely on teacher-predicted grades, and use rich administrative data linking student performance and teacher assessments across multiple STEM and non-STEM subjects. We find that, conditional on achieved grades, teachers systematically predict lower grades for boys relative to observationally similar girls within the same school across all subjects. In the second part of the analysis, we examine the extent to which observable student characteristics account for this pattern. We find that including measures of students’ cross-disciplinary skills reduces the gender gap in predicted grades by up to 70% in non-STEM subjects and fully explains or even reverses it in STEM. These results indicate that what appears to be a pro-girl bias is, at least in part, associated with students’ broader competencies rather than necessarily reflecting differential treatment by teachers. These findings have important implications for the fairness and effectiveness of teacher-based assessments in high-stakes contexts such as university admissions.