A Classroom Experiment on E ort Allocation under Relative Grading

Grading on the curve, or relative grading is one of the most commonly used grading schemes in education. Under this scheme, each student's grade is determined as a function of his or her percentile rank in the class. According to the law of large numbers, this mechanism is sensitive to changes in the size of the class. As a class size grows, the percentile ranks of students in a class draw closer to their respective percentile ranks in the student population, changing the incentives for students. I develop a model of this environment that incorporates strategic e ort in order to predict behavior for students with di erent abilities. I test this model in a large-scale classroom experiment that measures student e ort on 10 online quizzes. I randomly assign students to a grading cohort of either 10 or 100 students, where students receive high grades if their performance is in the top 70 percent of their cohort, and test the connection between the size of the grading cohort and the e ort students exert. Larger grading cohorts have less variance in the draw of students, and elicit greater average e ort. Smaller grading cohorts have greater variance in the draw of students, and elicit more e ort from the lowest ability students. Students in the bottom 30 percent exert greater e ort on average in the larger grading cohort, an allocation failure consistent with several possible behavioral biases. I discuss the welfare implications of the observed patterns of e ort and make policy prescriptions for di erent objectives of a classroom designer.