An evaluation of estimation methods in confirmatory factor analytic models with ordered categorical data in LISREL
Robust estimation methods for SEM models that include ordered categorical data are currently available to researchers in LISREL software However; little research has evaluated the efficiency of these methods. In this research, robust ML and robust DWLS estimation methods were examined, as well as two other commonly applied SEM estimators, WLS and normal theory ML. The effects of sample size, item distributional conditions, severity of non-normality of the data, and model size on resulting parameter estimates, standard error estimates, and model test and fit statistics for each estimation method were evaluated. Results indicated that both robust ML and robust DWLS performed much better on each of the study outcome variables regardless of condition than the other two estimators. Results also revealed that although the robust methods led to consistently unbiased parameter estimates, generally, robust ML resulted in more accurate standard error estimates, less biased chi square statistics, and lower Type I error rates than robust DWLS.