Performance of Common Fit Indices Across Robust Estimation Methods for Multilevel Confirmatory Factor Analysis With Ordered Categorical Data

• In Event: Methods for Multilevel, Categorical, and Multidimensional Data

Abstract

The discrimination of commonly used fit indices between correctly or incorrectly specified models within a multilevel confirmatory factor analytic context was investigated using receiver operating characteristics (ROC) analyses. Combining the ROC analyses with traditional methods of investigating fit index performance resulted in converging evidence for the utility of the investigated fit indices. Optimal thresholds were identified and varied across different robust estimation method. Estimation method and sample size influenced the performance of fit indices to detect misspecification at level-1. All fit indices performed poorly for detecting misspecification of the level-2 model when level-2 sample size was below 100. We offer recommendations for using commonly reported fit indices and cautions about the use of general cut-off criteria for ML-CFA.

Authors

• R. Noah Padgett, Baylor University

• Grant B. Morgan, Baylor University