An Evaluation of the Performance of Multiple Imputation and Auxiliary Variables in a Unique Case Study

• In Event: 42.035 - Statistical Power and Missing Data

Abstract

Missing data is a pervasive problem in the collection of large-scale educational data of which the use of Multiple Imputation (MI) is one solution. While well established theoretically, the effectiveness of this method in practice can be more challenging to study. Using a unique dataset where missing data was both present and also collected, we were able to investigate the effectiveness of MI. Further, we compared the performance of different types of auxiliary variables to help inform researcher’s future data collection and variable inclusion efforts in the face of missing data. We found MI to be effective and the collection of school and classroom characteristics to be critical for the performance of MI.

Authors

• Christopher M. Swoboda, University of Cincinnati

• Gita Taasoobshirazi, Kennesaw State University

• Benjamin Kelcey, University of Cincinnati

• Keanen McKinley, University of Cincinnati