Does chubby can get lower grades than skinny Sophie? Using an intersectional approach to uncover grading bias in German secondary schools

Abstract

We aim to uncover grading bias by gender, socio-economic status, ethnic/migration background as well as body weight in the German secondary school system. Following an intersectional approach, we test whether—controlling for ability—students receive different grades depending on (the specific combination of) ascriptive characteristics. Using data from the fourth starting cohort (SC4, 13.0.0, first survey in year 9 in 2010) of the National Educational Panel Study (NEPS) consisting of more than 14,000 ninth graders, we compute the predicted differences in grades for the different groups of students depending on whether they are a boy or a girl, whether they are obese/overweight or not, their socio-economic status (SES) and ethnic background. We rely on a grade equation approach, assuming that discrepancies between observed grades and achievement as measured in standardised tests are evidence of biased grading. We control for two different competence tests—the Domain General Cognitive Functions (DGCF) and a standardised domain-specific competence test—as objective measures of ability as well as secondary school track. Even after controlling for different personality and behavioural traits—the “big five”, the Strengths and Difficulties Questionnaire (SDQ), the Sick, Control, One, Fat and Food (SCOFF), health satisfaction and class retention—substantial differentials in grading across almost all factors and subjects remain. To account for the fact that many students may face bias on multiple grounds, we then compare the differences in predicted grades for groups with overlapping (dis)advantaging characteristics (e.g. low SES overweight Turkish boy vs a high SES non-overweight majority girl), while controlling for the objective ability measures. Significant differentials in grades are found in almost all cases, with the largest effect sizes for the subject German. We also compute models including all 2-way or 4-way interactions between the four axes of inequality and find the main effects largely unchanged. On the whole our findings are indicative of widespread additive intersectional effects of gender, social and ethnic origin as well as body weight on grading bias.