Integrating Auxiliary Information and Hot Deck Imputation for Addressing Nonignorable Missingness with Survey Weights

Abstract

Missing data is a common challenge in various data analysis tasks, especially for survey data. This thesis investigates imputation methods for addressing nonignorable missingness in surveys by combining auxiliary information and hot deck imputation. We present a novel approach that extends the model-based MD-AM (Missing Data with Auxiliary Margins) framework and can flexibly manage numerous variables, both categorical and continuous, subject to unit and item nonresponse. The simulations demonstrate that the proposed approach yields better accuracy and coverage rate than ICIN (Itemwise Conditionally In- dependent Nonresponse) specifications, which highlights its potential for practical applica- tions in survey analysis. This approach provides a viable alternative to strictly model-based imputation approaches with simplified imputation procedure and higher flexibility when handling complex survey designs.