There are two statistic analysis methods to solve the statistic problems with latent variables. One is the partial least squares analysis(PLS),using principal components analysis.The other is the structural equation model(SEM),using the maximum likelihood covariance structure analysis and the software LISREL is the most popular software to analyze SEM.In this paper,the parameter estimations are computed for two applied problems:Education function on personal income and Satisfaction analysis.We use the popular software LISREL and make a program by using the PLS to solve the two problems.The first problem can be solved by the two methods,and the parameter estimation is better by LISREL.The second problem cannot be solved by LISREL,as the covariance matrix is singular.We make a program using the modified BFGS algorithm and the maximum likelihood covariance structure analysis,the initial parameters are obtained by PLS.Our program can solve the second problem.The numerical experiment shows that the parameters obtained by our program is better than the parameters by PLS. |