Abstract

Despite the longstanding interest in spillover effects across different fields in economics and other social sciences, the empirical literature has not come to an agreement regarding identification and estimation in randomized experiments in the presence of externalities. This paper provides a causal framework to study spillover effects in a setting where units are clustered and their potential outcomes can depend on the treatment assignment of all the units within a cluster. Using this framework, I discuss parameters of interest and provide conditions under which spillover effects can be identified. I assess the performance of three commonly applied techniques, the difference in means, which ignores the presence of externalities, linear-in-means models and two-stage randomization designs. I then evaluate whether (or under what conditions) these three techniques identify the parameters of interest. Finally, I put forth easily implemented empirical strategies for estimating these spillover effects and I illustrate my proposed methodology using two empirical applications, one in education and one in political economy.