Peer Effects and Student Academic Achievement: An Application of Spatial Autoregressive Model with Group Unobservables

Peer effects have been of great relevance to policy implementation and evaluation. However, the empirical studies of peer effects are far from decisive, due to both conceptual and data problems (Manski (1993)). By using a recently available rich dataset from the National Longitudinal Study of Adolescent Health (Add Health), we are able to specify peers as actual friendship networks. Unlike the mean regression model in Manski (1993), the spatial autoregressive (SAR) models with group fixed effects have the potential to resolve most of the plaguing problems in social interaction estimation. In particular, the nonlinearity introduced in the peer measurement for individuals in the SAR model provides valuable information for identifying both the endogenous and contextual effects, thus the reflection problem resolved. Furthermore, we address the omitted variable bias problem and the selection bias problem by a school-grade fixed effect strategy. We find no evidence for endogenous social effects in student academic achievement after controlling for the school-grade fixed effect. However, some of the peer’s characteristics do matter. Specifically, associating with peers living with both parents helps improve a student’s GPA, while associating with peers whose mother receives welfare has negative effect. Therefore, there are peer effects among students, but the operating mechanism does not appear to be the behavior channel, which means that the social multiplier does not seem to exist among the students, at least in terms of academic attainment