Investigating Reliabilities Under Multilevel Structure With Missing Values

• In Event: Advances in Estimating Reliability: Innovative Approaches and Applications

Abstract

This study investigated the performance of various reliability indexes under the multilevel structure with different factors including missing data mechanisms, missing data proportions, sample sizes, intra class correlations, and level-1 initial reliability values. Three multilevel reliability indexes, multilevel Cronbach alpha, multilevel composite omega, and multilevel maximal H, were evaluated through absolute percent bias and convergence rates. Result from the pilot study indicated that with smaller missing data proportion, larger sample size, larger intra class correlation, and higher level-1 initial reliability, reliability indexes have less absolute percent biases and higher convergence rates.

Authors

• Minju Hong, University of Georgia - Athens

• Zhenqiu Laura Lu, University of Georgia