Most of studies in educational domains study used structural equation modeling(SEM) having the classical maximum likelihood estimation (MLE) which has been known as its consistency, efficiency, and easy computation. Despite its practical properties, a ...
Most of studies in educational domains study used structural equation modeling(SEM) having the classical maximum likelihood estimation (MLE) which has been known as its consistency, efficiency, and easy computation. Despite its practical properties, assumptions such as strict normality assumption and a large sample cause problems and ask an alternative method. The aim of this study was to explore positive and negative aspects of the classical SEM estimation(bootstrap, MLMV, and ML). Furthermore, performances of the classical SEM estimations and bayesian estimation approaches under the various conditions(normal and non-normal distribution, sample sizes, factor loading). This study would describe how bayesian estimation method behaviored under the conditions of model fit and heywood case. For those research purposes, all the analyses were performed with the MCMC. Simulations were performed under the conditions of normal and non-normal distribution with various factor loadings(.20, .50, .80) and sample sizes(50, 100, 200, 300). Four different estimation methods such as bayesian, bootstrap, MLMV, and ML were used for estimate the simulated data. Results obtained from the normal distribution showed that bayesian outperformed over bootstrap MLMV, and ML in terms of BIC values across all different factor loadings(.20, .50, .80) and sample sizes(50, 100, 200, 300). In addition, similar values of BIC were obtained among bootstrap, MLMV, and ML. Performances of all four estimations under the normal distribution were different across various factor loadings. With the factor loadings of .20, bayesian estimation methods outperformed over three estimation methods(bootstrap, MLMV, ML) when compared their performances with BIC. Under the factor loadings of .50 and .80, however, bayesian estimation method obtained higher values of BIC compared to the other three estimation methods. Considering the availability of SEM, this study should contribute some valuable information about behaviors of different estimations under the small sample sizes and various factor loadings.