A Comparison of Bayesian, MLM (Maximum Likelihood With Mean Adjustment), and WLSMV (Weighted Least Squares With Mean and Variance Adjustment) in Confirmatory Factor Analysis Under a Range of Non-Ideal Situations

• In Event: Roundtable Session 117
  In Roundtable Session: Discussions in Item Response Theory

Abstract

This simulation study is an ongoing study that compares the performance of four CFA estimators: Bayesian approach with informative and non-informative priors, robust maximum likelihood (MLM) and robust weighted least square techniques (WLSMV) in dealing with ordinal data under non-ideal situations. The manipulated factors include three underlying distribution of the ordinal indicator and eight sample sizes. Population model and item categories will be set at constant. Sample models will be correctly specified. The four estimation methods will be compared in terms of percentage of non-convergence, model goodness of fit, bias and efficiency of parameter estimates, and bias of parameter estimates’ standard error estimates. This is the first study to do a direct comparison of all four estimation techniques.

Authors

• Zijia Li, University of Kentucky

• Michael Toland, University of Kentucky

• Yuchen Yang, University of Kentucky