Inference on Peer Effects with Missing Peer Data: Evidence from Project

This paper studies peer effects on student achievement among first graders ran-domly assigned to classrooms in Tennessee’s Project STAR. The analysis uses previ-ously unexploited pre-assignment achievement measures available for sixty percent of students. Data are not missing at random, making identification challenging. The pa-per develops a new way, given random assignment of individuals to classes, to identify peer effects without other missing-data assumptions. Estimates suggest moderate, positive effects of mean peer lagged achievement. Allowing heterogeneous effects, evidence suggests lower-achieving students benefit more than higher-achieving stu-dents do from increases in peer mean. Further, the bias of a widely used but poorly understood peer-effects estimator is characterized