The Comparison of WLS and DWLS Estimation Methods in SEM to Construct Health Behavior Model

Abstract

It is unknown how reliable various point estimates, standard errors, and standard several test statistics are for standardized Structural Equation Modeling (SEM) parameters when categorical data used or misspecified models are present. This paper discusses the comparison between Weighted Least Square (WLS) and Diagonally WLS (DWLS) for examining hypothesized relations among ordinal variables. In SEM, the polychoric correlation is employed either in WLS or DWLS. This study constructs the Health behavior model as an endogenous latent variable in which exogenous latent variables are Perceived susceptibility and Health motivation. All indicators are in categorical types. Thus, data are not multivariate normal, or the model could be misspecified. This study compares the values of standard deviation and coefficient determination to determine a better model. The goodness of fit for the overall model is based on Tucker Lewis Index (TLI), Root Mean Square Error Approximation (RMSEA), and Confidence Fit Index (CFI). This present study found that the WLS estimator method tent to result in better values than DWLS’s. Health motivation has a higher effect on Health behavior than Perceived susceptibility, obtained based on both estimators.