When the Practitioner Saved the Theorist: Theory and Practice of Inference in Regression Discontinuity

Abstract

This paper analyzes the asymptotic distribution of local polynomial estimators in the context of regression discontinuity designs. The standard approach in the literature (Hahn Todd and van der Klaauw 2001, Porter 2003, Imbens and Lemieux 2008 and Lee and Lemieux 2009) is to assume the bandwidth, h, around the discontinuity shrinks towards zero as the sample size increases. However, in practice, the researcher has to choose an h > 0 to implement the estimator. This paper derives the …xed-h asymptotic distribution that allows for the bandwidth to be positive, providing re…ned approximations for the estimator’s behavior. When h > 0, the small-h asymptotic variance is equivalent to assuming that the density of the running variable and the conditional variance of the dependent variable are constant around the cuto¤. Simulations provide evidence that …xed-h asymptotic distributions better describe the behavior of both bias and variance of the estimator, leading to improved inference. Estimators for …xed-h standard errors are proposed and incorporate the theoretical gains of the improved approximations. The …xed-h variance estimators improve markedly over small-h estimators in the presence of some forms of heteroskedasticity. Interestingly, in the special case of homoskedastic errors using a local linear estimator, the variance estimators based on small-h asymptotics produce tests with similar size to the …xed-h variance estimators proposed in this paper.