In observational causal inference, domain knowledge often leaves multiple covariate adjustments plausible, yet which of them satisfy ignorability is not testable. Different adjustments can yield conflicting conclusions about the average treatment effect, and standard remedies (e.g., adjusting for their union or intersection, or reporting the union or convex hull of the confidence intervals) can fail or produce confidence sets that do not shrink to a point with growing sample size. We propose a specification-robust procedure that returns a single point estimate and a confidence interval that is valid when at least one candidate adjustment satisfies ignorability, and has width shrinking at the parametric $1/\sqrt{n}$ rate. Our approach mirrors how trimming and overlap weighting handle overlap violations: We shift the target to a reweighted population, closest in KL divergence to the original population, for which credible, specification-robust inference is feasible. We also provide diagnostics to assess the population shift and an extension to protect any function of the covariates used for reweighting, similar to adding calipers in matching. Synthetic and real-data examples demonstrate that the proposed procedure provides substantially tighter confidence intervals than the convex hull while achieving nominal coverage for the new target population.

翻译：暂无翻译