A quasi-experimental study of ethnic and gender bias in university grading.

Abstract

This paper contributes to the debate on race- and gender-based discrimination in grading. We apply a quasi-experimental research design exploiting a shift from open grading in 2018 (examinee's name clearly visible on written assignments), to blind grading in 2019 (only student ID number visible). The analysis thus informs name-based stereotyping and discrimination, where student ethnicity and gender are derived from their names on written assignments. The case is a quantitative methods exam at Roskilde University (Denmark). We rely on OLS regression models with interaction terms to analyze whether blind grading has any impact on the relative grading differences between the sexes (female vs. male examinees) and/or between the two core ethnic groups (ethnic minorities vs. ethnic majority examinees). The results show no evidence of gender or ethnic bias based on names in the grading process. The results were validated by several checks for robustness. We argue that the weaker evidence of ethnic discrimination in grading vis-à-vis discrimination in employment and housing suggests the relevance of gauging the stakes involved in potentially discriminatory activities.