Abstract
The impact of peers on student outcomes has important policy implications for how students are organized into classes and the overall impact of education interventions. But it is difficult to accurately measure peer effects because of the nonrandom sorting of students and teachers into classrooms and the endogeneity of peers' achievement. In this study, an education production function (EPF) is specified that includes student and peer characteristics as regressors. This model is estimated using student-level data from New York City public schools for 1995–2000. The richness of these data allows six sources of bias that arise in the EPF model to be addressed, including the above-mentioned nonrandom classroom assignment and the endogeneity of peers' achievement. This results in credible evidence of (small) peer group effects. Instrumenting for the mean of peers' achievement significantly reduces the associated peer effect. Nonlinear peer group effects are evident in the form of a small positive impact associated with the homogeneity of peers' achievement. Generally, peer characteristics do not appear to affect individual performance. Also included in this analysis is an application of a new methodology developed by Graham (2007) that identifies peer group effects through their impact on the variance in classroom mean test scores. The approach is less susceptible to the six biases that plague the EPF approach. The evidence from this exercise indicates that peer group effects are present and corroborates the results from the EPF approach.