Sensitivity Analysis for Survey Weights

Abstract Survey weighting allows researchers to account for bias in survey samples, due unit nonresponse or convenience sampling, using measured demographic covariates. Unfortunately, practice, it is impossible know whether the estimated weights are sufficient alleviate concerns about unobserved confounders incorrect functional forms used weighting. In following paper, we propose two sensitivity analyses exclusion of important covariates: (1) a analysis partially observed (i.e., variables across sample, but not target population) and (2) fully either population). We provide graphical numerical summaries potential that arises from such confounders, introduce benchmarking approach quantitatively reason their results. demonstrate our proposed state-level 2020 U.S. Presidential Election polls.