Abstract
When performing data analysis using the structural equation model, one of the main concerns is the estimation and testing of path coefficients. As a test method for path coefficients, a method using t-value, a method using a scaling test statistic of Satorra-Bentler, and a bootstrap method of Bollen-Stine are used. All of these methods are approximate testing methods, and in some cases provide different test results depending on the type of data and model given. In this paper, the type I error was calculated under special circumstances by using bootstrap simulation for the main test methods, and through this, the performance of these testing methods was evaluated. In terms of estimation and testing, the ML method was found to have a robust property in deviating from the normal distribution. On the other hand, in the case of the WLS method, the non-convergence rate was relatively high when the sample size was small, and the type I error was also relatively large. Therefore, it can be seen that the WLS method is useful when the sample size is quite large. In the case of the bootstrap methods, it can be seen that the type I error tends to increase as the sample size increases. Therefore, it is considered that a supplement to improve this phenomenon is necessary when applying the bootstrap methods.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have