Identifying the Incidence of "Grading on a Curve":
A Within-Student Across-Subject Approach

Theoretical work shows that grading on a curve, i.e., teachers assessing students relative to their classmates, can negatively affect students’ learning effort. However, little is known about its empirical incidence. To overcome bias from non-random sorting and omitted variables like teachers’ grading standards, we exploit within-student acrosssubject variation observing both teacher-assigned grades and test scores of German 4th-graders in reading and math. We find that having classmates with one standard deviation higher test scores lowers a student’s grade by about 10 percent of a standard deviation. Importantly, only female teachers grade on a curve and there is no association between students’ learning effort and relative grading.