Power of randomization-inference tests for jurisdiction-specific ATTs

Determine the power of randomization-inference tests for jurisdiction-specific average treatment effects on the treated in staggered-adoption difference-in-differences designs, particularly when only a few jurisdictions implement the policy.

Background

The paper studies inference for aggregate and sub-aggregate average treatment effects on the treated using UN-DID and DID-INT under staggered policy adoption. Randomization inference performs close to its nominal size in the paper’s placebo-law simulations for most aggregate, cohort-specific, and policy-specific estimands, but jurisdiction-specific tests can be conservative when only one jurisdiction implements the placebo policy at each treatment date.

The authors explicitly identify statistical power as unresolved for randomization-inference tests targeting jurisdiction-specific effects, especially in settings with few treated jurisdictions. Addressing this question would clarify the practical usefulness of jurisdiction-level treatment-effect inference when policy adoption is sparse.