The Study On The Problems Of Structural Equation Modeling

In the fields of social sciences, economics, management and marketresearch, we need to deal with relationships between multiple causes andmultiple results, encounter variables which can not be directly observed(Latent variable). Since the1980s, Structural Equation Analysis developsrapidly, it makes up the deficiencies of Traditional Statistical Methods,and becomes important tool for multivariate statistical analysis.Structural Equation Modeling (SEM) is a research methodology. It isbased on the statistical analysis techniques. It can be regarded as thesynthesis of different technologies and research methods, such as pathanalysis, canonical correlation analysis, factor analysis, discriminantanalysis, multivariate analysis of variance and multiple regressionanalysis. Compared with the traditional approaches, SEM is ameasurement technology, it combines "measurement" and "analysis". Ithas a lot of advantages, for example, it can estimate the measurementindex and the latent variables in the model; it can estimate themeasurement error of index variable in the forecasting process; it can alsoassess the measurement reliability and validity.In this paper, we do the following work:（1）Compare some fitting indexes; discuss the similarities and thedifferences between Maximum Likelihood and Bayesian Approaches inanalyzing Structural Equation Modeling; Through a simulation, we getthe conclusion: values of Bayesian estimation are more accurate thanMaximum Likelihood; when the sample volume is increased, theaccuracy of these two estimate methods will improve, and thedifferences will be smaller as well.（2）Research: If adding or removing certain factor, the factor loadings ofthe corresponding latent variable and the others will change or not. Byresearching, we get the conclusion: the former almost changes, while thelatter does not change. （3）Research: whether there are certain relationships between the correla-tion values of first-order factors and the estimations of the second-orderfactors. By researching, we get the conclusion: when the correlationvalues of the first-order factors are almost the same, the estimations of thesecond-order factors are almost the same too; In addition, if thecorrelation values between some first-order factor and the others arerelatively larger, the second-order parameters of this first factor will berelatively larger.（4）Based on the model for Zigong tourist satisfaction by Factor analysis,we use SEM to do further validation on the model. The relevant dataindicates this model is good. We give the SEM diagram, it canintuitively show the relationships between the variables. In the factoranalysis model, the latent variables are not relevant. But in the real life,the three latent variables influence each, so the structural model canreflect the correlation between the latent variables. According to thisrespect, the structural model is better. In addition, on the basis of themodel, we give some suggestions, hoping to provide some usefulreferences for the tourism industry.