Thesis (M.Sc.) - Cairo University - Institute of Statistical Studies and Research - Department of Statistics and Econometrics

Structural equation modeling (SEM) is a widely used statistical method in most of social science fields. Similar to other statistical methods, the choice of the appropriate estimation methods affects the results of the analysis, thus we found it important to examine the performance of SEM estimation methods under the different situations a researcher might face in reality. Thereby, an investigation on the performance of SEM estimation methods was held through an application study as well as simulation study, we mainly divided the studies into two sections: One under complete data analysis and the second is under incomplete (missing) data analysis. In simulation studies different conditions were imposed with respect to sample sizes and factor loading values, as well as misspecification but only under complete data. Both studies were executed through the statistical software R. Finally, the performances of the estimation methods were compared in terms of RMSEA, SRMR, CFI, TLI, and convergence rate (especially with missing data). Under complete data it was found that maximum likelihood, robust maximum likelihood, and diagonally weighted least squares gave better fit to the model than the other two methods. Under misspecified model, it was found that ML method was the most sensitive method to misspecification, followed by WLS and GLS. It was also concluded that the effect of factor loading was negative on the fit indices, which might be used as an indicator for misspecification. DWLS was the least method that showed sensitivity to misspecification

Issued also as CD