🤖 AI Summary
Missing data arising from unit and item nonresponse in complex survey designs pose significant challenges for valid statistical inference.
Method: This paper proposes a novel multiple imputation framework that integrates the original design weights—rather than synthetic or reweighted ones—with known marginal distributions of auxiliary variables. Specifically, it embeds the true design weights directly into an imputation model for nonignorable unit nonresponse, combining weighted generalized linear models with marginal constraint optimization to ensure calibration to population totals.
Contribution/Results: Simulation studies demonstrate that the proposed method strictly satisfies auxiliary marginal constraints while substantially reducing bias and mean squared error in target parameter estimates. It improves estimator consistency and overall representativeness relative to existing approaches relying on artificial weights, thereby enhancing the validity and efficiency of survey inference under complex nonresponse mechanisms.
📝 Abstract
Survey data typically have missing values due to unit and item nonresponse. Sometimes, survey organizations know the marginal distributions of certain categorical variables in the survey. As shown in previous work, survey organizations can leverage these distributions in multiple imputation for nonignorable unit nonresponse, generating imputations that result in plausible completed-data estimates for the variables with known margins. However, this prior work does not use the design weights for unit nonrespondents; rather, it relies on a set of fabricated weights for these units. We extend this previous work to utilize the design weights for all sampled units. We illustrate the approach using simulation studies.