This paper proposes a fully nonparametric kernel method to account for observed covariates in regression discontinuity designs (RDD), which may increase precision of treatment effect estimation. It is shown that conditioning on covariates reduces the asymptotic variance and allows estimating the...