ABSTRACT:
I provide a non-parametric potential-outcomes framework to study causal 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 in a randomized experiment. In addition, I characterize and discuss the causal interpretation of the estimands that are recovered by three specifications that are widely popular in empirical work: a regression of an outcome on a treatment indicator (difference in means), a regression of an outcome on a treatment indicator and the proportion of treated neighbors (a reduced-form linear-in-means model) and a regression exploiting variability in treatment assignment probabilities in two-stage designs. Finally, I provide conditions for uniform consistency and asymptotic Normality of direct and spillover effects estimators with special focus on the effect of potential outcome modeling assumptions and treatment assignment mechanism on inference. I illustrate my findings with data from a randomized conditional cash transfer pilot in Colombia and with a simulation study.

BIO:
Gonzalo Vazquez-Bare is a Ph.D. candidate in the Department of Economics at the University of Michigan, where he is also completing the M.A. in Statistics. He works on econometrics and methodology, with focus on program evaluation and causal inference in experimental and non-experimental settings. He has worked with Matias Cattaneo and I in several papers on regression discontinuity designs, and has an exciting research agenda on estimation and inference of treatment effects in the presence of externalities and interference between units.