Study On Application Problems Of Structural Equation Model

Now Structural Equation Modeling(SEM) has been widely used to evaluate causal and predictive hypothesis in Behavioral sciences and Sociology research, which is an econometric analysis technology that combines measurement and analysis. But in the current research and application, there are still some problems with it. First, it is not based on the theory of normal distribution assumption in the application of structural equation modeling; Second, there is no correct understanding to the problem of minimum sample; Third, the study of structural equation model using simulated data is in the majority; Fourth, the application of structural equation model is wrong, not following the reasonable logic analysis strategy. In the part of theory analysis part, for normal distribution problem, this paper reiterated the normal distribution assumption’s importance to Structural Equation Modeling, and summarized several correction methods of non-normal distribution, and suggested that using Bootstrap method estimated the model parameter and standard error; And about the minimum sample size, summarizes that the minimum sample size is influenced by five factors and several rules of thumb to determine the minimum sample. The minimum sample size should be determined according to the influencing factors and the rule of thumb; Finally, based on the essence of structural equation model, there was erroneous zone in the current application of structural equation model. Then put forward a reasonable logical strategy in applying the Structural Equation Modeling, which combine exploratory and confirmatory analysis, as well as three principles to build path of structural model. In the part of empirical analysis part, using the data of the national happiness survey of Guangdong province in 2014 to conduct the empirical analysis, the results are well validated. To ignore the normal distribution can lead to a higher Chi-square value. We need to correct the non-normal distribution; Influenced by the factor of normal distribution, the minimum sample of the happiness influencing factors model is 352; To solve the problems of normal distribution and minimum sample simultaneously, using the Bootstrap method can get the asymptotic refinement.