Confirmatory factor analysis with ordinal data : effects of model misspecification and indicator nonnormality on two weighted least squares estimators
Year: 2009
Type: article
Abstract: Full weighted least squares (full WLS) and robust weighted least squares (robust \nWLS) are currently the two primary estimation methods designed for structural equation \nmodeling with ordinal observed variables. These methods assume that continuous latent \nvariables were coarsely categorized by the measurement process to yield the observed \nordinal variables, and that the model proposed by the researcher pertains to these latent \nvariables rather than to their ordinal manifestations. \nPrevious research has strongly suggested that robust WLS is superior to full WLS \nwhen models are correctly specified. Given the realities of applied research, it was \ncritical to examine these methods with misspecified models. This Monte Carlo simulation \nstudy examined the performance of full and robust WLS for two-factor, eight-indicator confirmatory factor analytic models that were either correctly specified, overspecified, or \nmisspecified in one of two ways. Seven conditions of five-category indicator distribution \nshape at four sample sizes were simulated. These design factors were completely crossed \nfor a total of 224 cells. \nPreviously findings of the relative superiority of robust WLS with correctly \nspecified models were replicated, and robust WLS was also found to perform better than \nfull WLS given overspecification or misspecification. Robust WLS parameter estimates \nwere usually more accurate for correct and overspecified models, especially at the \nsmaller sample sizes. In the face of misspecification, full WLS better approximated the \ncorrect loading values whereas robust estimates better approximated the correct factor \ncorrelation. Robust WLS chi-square values discriminated between correct and \nmisspecified models much better than full WLS values at the two smaller sample sizes. \nFor all four model specifications, robust parameter estimates usually showed lower \nvariability and robust standard errors usually showed lower bias. \nThese findings suggest that robust WLS should likely remain the estimator of \nchoice for applied researchers. Additionally, highly leptokurtic distributions should be \navoided when possible. It should also be noted that robust WLS performance was \narguably adequate at the sample size of 100 when the indicators were not highly \nleptokurtic.
Author: Phillip Wingate Vaughan
Language: English
Field: Decision Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W1901908397
