Association for Public Policy Analysis & Management

A growing literature explores peer effects in education, but one frequent confounder for causal inference is within-environment sorting of students, most often towards others like themselves (i.e. homophily). While researchers believe certain ascriptive characteristics such as race and gender are important for homophily, there is little existing evidence to show the relative importance of these factors. This paper introduces methodology that builds upon the peer effects literature regarding partially overlapping networks. We approximate the social space and estimate a model capturing which peers influence each other under homophily. In our approximation of the social space, we also address network density, which we argue contributes to the wide range of peer effect estimates in the literature. The paper gives empirical evidence for the relative importance of sorting by gender, ethnicity, neighborhood, bus stop, bus route, language, and country of birth in the context of New York City elementary school classrooms. We find that the most important sorting avenues are shared bus, gender, and country of birth. Including several networks related to location and ethnicity (ex: ethnicity, language spoken at home, country of birth) allow us to better understand what aspects of these characteristics are most influential for within classroom sorting. For these location networks, we find evidence that residential proximity alone is not important, but opportunity for outside the classroom social interaction is.