Model-Based Multiple Imputation for Multilevel Analyses

• In Event: 45.075 - Multilevel Models, Multiple Imputation, and Latent Models: Unrelated but Equally Important Topics

Abstract

Some very common analysis models in the educational literature are known to cause problems for popular missing data handling approaches, leading to bias. Regression models with random coefficients, interactive, or polynomial effects are examples that are exceedingly common in the literature. In the context of single-level regression, an approach known as model-based imputation has shown great promise. The purpose of this paper is to extend model-based imputation to multilevel models with up to three levels, with functionality for categorical variables. Computer simulation results suggest that this new approach can be quite effective when applied to multilevel models with random coefficients and interaction effects, and the procedure is available in a free software package for macOS, Windows, and Linux.

Authors

• Craig K. Enders, University of California - Los Angeles

• Han Du, University of California - Los Angeles

• Brian Keller, University of California - Los Angeles