Placebo zones in discontinuity‐based designs: Estimation, inference, and implementation

We propose a new model‐selection algorithm for regression discontinuity design and related estimators. The performance of candidate models is assessed within a “placebo zone” of the running variable. Candidate models can differ by bandwidth and other choice parameters. We outline (restrictive) sufficient conditions under which the approach is asymptotically optimal, and then show the approach also performs favorably under more general conditions in Monte Carlo simulations, including simulations calibrated to well‐known real‐world applications. We also propose a new randomization inference procedure which draws on the placebo estimates. Our Stata commands implement the procedure and compare its performance to other approaches.