Estimation of Cross-Classified Random Effects Models With Missing Data Using Multiple Imputation

• In Event: Multilevel Modeling: Nonnested Multilevel Data, Measurement Invariance, and Nonlinear Models

Abstract

Cross-Classified Random Effects Modeling (CCREM) can be used to model non-hierarchical nesting structures in data (Beretvas, 2010). When data are missing in hierarchical settings, recent advances have allowed the use of Multiple Imputation (MI) to obtain unbiased parameter estimates (van Buuren, 2012), however the application of MI in CCREM has been understudied. Ignoring cross-classification when using MI is hypothesized to create bias throughout the CCREM data analysis. The purpose of this paper is to propose the use of a CCREM imputation model for MI and evaluate its performance against alternatives in a simulation study. These results will be evaluated with relative bias and RMSE, and suggestions and code for implementation for researchers will be provided.

Authors

• Christopher M. Swoboda, University of Cincinnati

• Benjamin Kelcey, University of Cincinnati

• Jiaqi Zhang, SKT Education Group