Comparison Of Ridge PLS-SEM And Ridge CB-SEM With Small Sample

Structural Equation Modeling(SEM)is a comprehensive research method based on statistical analysis technology,which includes many methods,such as factor analysis,path analysis,multiple regression,etc.It can deal with the correlation between multiple variables at the same time,and has been widely used in social and behavioral sciences.At present,there are two main structural equation modeling methods: CB-SEM(covariance-based SEM)and PLS-SEM(partial least squares SEM).When the sample size is small,CB-SEM method often does not converge,and ridge method can improve this problem.Compared with CB-SEM method,PLS-SEM method is more suitable for the construction of structural equation model of small sample investigation.On the basis of introducing the existing ridge CB-SEM method,this paper mainly studies the following three aspects:(1)The ridge method is applied to PLS-SEM.Through Monte Carlo simulation,the performance of the ridge PLS-SEM method and the ridge CB-SEM method in parameter estimation under small samples is compared,including the estimation deviation and accuracy.The research shows that the estimated values obtained by ridge CB-SEM method and ridge PLS-SEM method are generally more accurate than those obtained by conventional CB-SEM and PLS-SEM methods,but the deviation is not obviously improved.When the sample size is small,the accuracy of PLS-SEM and ridge PLS-SEM is better than CB-SEM and ridge CB-SEM.For endogenous latent variables affected by intermediate variables,ridge PLS-SEM is the most accurate in estimating the influence(path coefficient)of other latent variables.(2)Through the case of college students’ supermarket satisfaction,the estimation performance of Ridge PLS-SEM method and Ridge CB-SEM method in practical application is further studied.The standard error and confidence interval of parameter estimation are obtained by Bootstrap method,and the accuracy of estimation is compared,mainly comparing the path coefficients of endogenous latent variables affected by intermediary variables.It is found that among ridge CB-SEM,PLS-SEM and ridge PLS-SEM,the standard error of the estimated value obtained by ridge PLS-SEM is the smallest,and the confidence interval is also the shortest under the same confidence level,which shows that the estimation accuracy of ridge PLS-SEM is higher than that of ordinary PLS-SEM.(3)The variation of standard error of structural coefficient estimation in the model under different ridge constant settings when ridge CB-SEM method is applied is studied.Based on the structural equation model of college students’ supermarket satisfaction,it is found that under different ridge constants,the standard error of regression equation parameter estimation tends to be stable.