Estimating peer-influence effects under homophily: Randomized treatments and insights

Abstract

When doing causal inference on networks, there is interference among the units. In a social network setting, such interference among individuals is known as peer-influence. Estimating the causal effect of peer-influence under the presence of homophily presents various challenges. In this paper, we present results quantifying the error incurred from ignoring homophily when estimating peer-influence on networks. We then present randomized treatment strategies on networks which can help disentangle homophily from the estimation of peer-influence.