Strategies for Imputing Missing Values in Hierarchical Data: Multilevel Multiple Imputation

• In Event: Poster Session 3
  In Poster Session: Division D Section 2 Quantitative Methods and Statistical Theory

Abstract

While statistical techniques for missing data treatments with single-level data have advanced greatly during the past several decades, there is a lack of methods for multilevel data. This is problematic as educational data frequently exhibit a multilevel structure and include missing observations. In this paper, it is shown that missingness would likely cause more severe problems at the second or higher levels than at the lowest level, and that single-level imputation methods may frequently result in biased estimates and incorrect standard errors. This paper provides a series of statistical strategies that are readily applicable for imputing missing values in multilevel data. Using real data examples and simulation, it is shown that these techniques could significantly improve estimation of multilevel models.

Authors

• Jee-Seon Kim, University of Wisconsin - Madison

• Christopher M. Swoboda, University of Cincinnati