Discrete Choice with Generalized Social Interactions

This paper studies social interactions in discrete choice models in which individuals can respond differently to different network members, conforming to some while distinguishing from others. In this generalized framework, I show how to use within-network variation in individual choices to draw inference about differences in endogenous interaction effects, while accounting for unobserved contextual effects. I also show how to construct internal instruments to correct for measurement error bias, a key source of endogeneity in network models with incomplete information. I apply these results to data from two empirical settings: classroom peer effects in Tennessee primary schools and spillovers in deworming treatment uptake in Kenya. In both applications, I find that differences in social interaction effects, with individuals conforming more strongly to similar peers, play a key role in shaping people's choices, with implications for the targeting of policy interventions.