Handling Spillover Effects in Empirical Research

Despite their importance, the discussion of spillover effects in empirical research misses the rigor dedicated to endogeneity concerns. We show that i) even with random treatment, spillovers lead to an intricate bias in estimating treatment effects, ii) there is a trade-off between endogeneity and spillover concerns, iii) the practice of using individual level regressions to identify direct effects and aggregate level regressions to learn about spillover effects can lead to misleading conclusions. We develop a simple guidance for empirical researchers, apply it to two empirical settings, and highlight differences in the results compared to current empirical practice.