A1863
Title: Generic covariate adjustment for regression discontinuity designs
Authors: Zhengfei Yu - University of Osaka (Japan) [presenting]
Jun Ma - Renmin University of China (China)
Yuya Sasaki - Vanderbilt University (United States)
Abstract: An empirical entropy balancing estimator is introduced for regression discontinuity (RD) designs that incorporates covariate balance into estimation. The proposed method reweights the standard local polynomial RD estimator using entropy balancing weights, which are obtained by minimizing a divergence from uniform weights subject to covariate balance conditions. The estimator admits an empirical likelihood representation that efficiently exploits covariate balance conditions as correctly specified over-identifying moment restrictions, yielding an asymptotic variance no greater than that of the standard estimator that ignores covariates. Relative to linear regression adjustment methods, the proposed approach offers two key advantages: it guarantees estimation consistency for nonlinear causal estimands, such as quantile treatment effects in RD designs, and it improves estimation efficiency for causal parameters involving derivatives, such as those arising in regression kink designs, by flexibly incorporating balance conditions on both levels and derivatives of the covariates.
