General Seminar - Dakyung Seong

Spillover Effects in Fuzzy Regression Discontinuity Design 

This paper studies treatment spillovers in fuzzy regression discontinuity (RD) designs with socially interacting units, such as couples, friends, and neighbors. Because an individual's treatment take-up may respond to both their own and their peer's treatment assignments, standard fuzzy RD identification strategies are generally invalid in this setup. This paper establishes new identification results for direct, indirect, and social-interaction local average treatment effects under alternative restrictions on joint compliance behavior. We also develop local-polynomial estimators and bias-corrected inference procedures and examine their finite-sample performance via Monte Carlo simulations. The proposed identification and estimation approaches are applied to study the impact of retirement on health behaviors. The estimates suggest that spousal retirement significantly reduces drinks per drinking day among men whose spouses belong to certain compliance groups.