Estimating Peer Influence in Multilayer Networks

Abstract

This paper proposes a model-based empirical method to identify influential individuals in risky behaviors. To determine the most influential individuals, we estimate peer influence using observational cross-sectional data from multiple social connections. Our empirical strategy employs the observed characteristics of distant individuals across multiple social networks as instruments to address the endogeneity arising from homophily. Using Add Health data, we find positive peer effects from friends and classmates on both cigarette smoking and marijuana use. Based on the estimated peer effects, we characterize the influencers in our sample.