Using Multilevel Dependencies to Improve Estimation With Missing Data

• In Event: Multilevel Modeling Methods and Applications

Abstract

Methods for estimation with missing data and multilevel models have been well studied individually, but estimation in multilevel settings with missing data provides complications and potential opportunities. Multiple imputation has great potential in this setting to use information gathered from the multilevel structure of the data to improve estimation. This paper illustrates multilevel imputation models and conditional empirical bayes estimates as improvements to the performance of the imputation model and thus, the results of a multilevel data analysis. A simulation study will evaluate the individual benefits of these additions using data generated from a realistic educational setting. Findings and recommendations for educational researchers will be provided.

Authors

• Christopher M. Swoboda, University of Cincinnati

• Jee-Seon Kim, University of Wisconsin - Madison