| Summary: | 碩士 === 國立臺北大學 === 統計學系 === 103 === The sampling design is getting more complex. Many large sample surveys are based on complex sampling methods to increase precision of estimation. The estimation may lower the accuracy if the data analysis ignores the complex sampling design but using simple random sampling method to simplify the estimation. This paper aims to compare the estimators of structural equation modeling under complex sampling design based upon a Monte Carlo approach. Five methods of structural equation modeling, namely, maximum likelihood (ML), unweighted least squares (ULS), generalized least squares (GLS), weighted least squares (WLS), and pseudo maximum likelihood (PML), are proposed in this study to compare their accuracy based upon mean square error (MSE).
In general, the simulation results show that there is no significant difference among the MSE of the five methods. All the MSE of the five methods are small, which indicates that the performances of the five methods are pretty good. Besides, the methods considering the complex sampling design (PML) have relative larger variance than those methods ignoring the complex sampling design (ML, ULS, GLS). This implies that the estimators ignoring the complex sampling design may underestimate the variance.
|
|---|
