Estimating average treatment effects with only discrete covariates

We propose a kernel smoothing method to estimate average treatment effects (ATE) when covariates are all discrete variables. This approach is based on Ouyang et al. (2009), who developed a Econometric Theory, 2009;25: 1–42 who developed least squares cross‐validation (CV) kernel smoothing method dealing with discrete covariates. This method automatically smoothes the irrelevant variables out of the regression model with a high probability, and avoids the problem of loss of efficiency related to the traditional nonparametric frequency‐based method. We derive the asymptotic distribution of our proposed ATE estimator. Monte Carlo simulations and empirical applications showcase the usefulness of the proposed method.