Website CV Paper
Fields
Econometrics
Job market paper
Sensitivity Analysis for Difference-in-Discontinuities Designs
Difference-in-discontinuities (DiDC) designs identify the treatment effect of a policy introduced at a cutoff that also determines exposure to a pre-existing confounding policy by comparing the post- and pre-period regression discontinuities. The canonical design assumes that the confounding discontinuity remains constant over time, which may be difficult to justify in practice. This paper develops an identification and inference framework when this assumption may be violated. Following the sensitivity framework developed in the difference-in-differences literature, we leverage observed discontinuities from multiple pre periods to inform restrictions on the evolution of the counterfactual discontinuities after the introduction of the policy. The resulting identified sets nest the conventional point-identified estimand as a special case. We use an existing joint asymptotic normality result for local-polynomial RD estimators and the consistency of the full cross-period covariance estimator to justify confidence sets based on moment inequality tests in this nonparametric DiDC setting. These sets provide pointwise asymptotic coverage under explicit regularity, bias, bandwidth, and sensitivity class conditions. Simulations show that, when confounding varies over time, the proposed procedure has better coverage probability relative to conventional DiDC inference.