Academic peer effects with different group assignment policies : residential tracking versus random assignment

I study the relative academic performance of students tracked or randomly assigned to South African university dormitories. Tracking reduces low-scoring students' GPAs but has little effect on high-scoring students. This lowers mean GPA and raises GPA dispersion. I also directly estimate peer effects using random variation in peer groups across dormitories. Living with higher-scoring peers raises students' GPAs and this effect is larger for low-scoring students. Peer effects operate largely within race groups but operate both within and across programs of study. This suggests that spatial proximity alone does not generate peer effects. Interaction of some sort is required, but direct academic collaboration is not the relevant form of interaction. I integrate the results from variation in group assignment policies and variation in group composition by drawing on the matching and sorting literatures. Both sets of results imply that own and peer academic performance are substitutes in GPA production and that GPA may be a concave function of peer group performance. The cross-dormitory results correctly predict a negative effect of tracking on low-scoring students but understate the magnitude of the observed effect. I show that this understatement reflects both policy-sensitive parameter estimates and problems with extrapolation outside the support of the data observed under random assignment. This underlines the value of using both cross-policy and cross-group variation to study peer effects