Identification and Estimation of Marginal Treatment Effect Through Social Networks

Abstract

This thesis develops a method for examining the role peer effects play in treatment effect models. We focus on the scenario where peer effects are important in determining the treatment decision. Identification of both the treatment decision and the associated marginal treatment effect is explored. In particular, exogenous and endogenous peer effects are used as instruments to identify marginal treatment effects. A Bayesian estimation procedure is presented, utilising a network formation model to adjust for unobserved peer effects. The performance of the model and the estimation procedure is analysed through a Monte Carlo experiment. The proposed method is then applied to estimate the effect of high school peers on the decision to attend college, and the return to education associated with such a decision using the Add Health data-set. Both the Monte Carlo experiment and the empirical application underscore the importance of accounting for the presence of peer effects in treatment effects models, and allows us to consider policy implications of peer effects.