Finite-sample coverage of MADID subsampling intervals

Determine the finite-sample coverage of confidence intervals for the model averaged difference-in-differences (MADID) estimator when subsampling uses only three observations per subsample, as in the Dobbs application.

Background

The paper develops subsampling-based confidence intervals for the MADID estimator because the estimator’s data-dependent weights can have a non-Gaussian limiting distribution. The asymptotic validity result requires the subsample size to diverge, whereas the Dobbs application uses only three observations per subsample. Consequently, the asymptotic coverage theorem does not provide a finite-sample guarantee in that application, leaving the actual finite-sample coverage unresolved.