Identification of peer effects through social networks using variance restrictions

This paper presents a novel method to identify peer effects through exploiting fluctuations in the variances and covariances of outcomes over a social network. Endogenous and contextual peer effects can be disentangled by contrasting the covariances of the outcomes of peers with the covariances of the outcomes of peers of peers. Correlated effects are incorporated by allowing for unrestricted network level heterogeneity. The method is applied to the Project STAR data to study peer effects in academic attainment of first-year kindergarten students. Evidence is found of positive peer effects, which are statistically significant and consistent with the existing literature.