Impact of Missing Data on Reliability for Multilevel Data

• In Event: AERA Roundtable Session 10
  In Roundtable Session: Missing Data in Measurement

Abstract

Reliability is an important statistic to measure the quality of a test in social and behavioral sciences. For single-level data, reliabilities such as Cronbach alpha, or coefficient omega have been proposed. For multi-level data, Geldhof, Preacher & Zyphur (2014) examined the level-specific reliabilities under two-level data structure. However, these reliability estimates did not consider the impact of missing data, which are very common in tests and questionnaires. This study aims to investigate the performance of various reliabilities for multi-level data with missing values. Results showed that when intra class correlations (ICC) is large, level-specific reliabilities perform better. When missing data proportions are large, or sample sizes are small, or list-wise deletion method is used, the reliability estimates did not perform well.

Authors

• Zhenqiu Laura Lu, University of Georgia

• Minju Hong, University of Georgia - Athens