Abstract

We propose a mean squared error optimal estimator for regression discontinuity (RD) designs with a vector of running variables. For the estimator, we show the asymptotic normality of a multi-dimensional local polynomial estimator with dimension-specific bandwidths. In the Monte Carlo simulation, we demonstrate that our spatial RD estimator outperforms the conventional estimator with a shape restriction. We apply our procedure to a place-based policy in West Germany. This is a joint with Takuya Ishihara, Daisuke Kurisu, Yasumasa Matsuda.