Research On Structural Equation Model: Structure And Partial Least Square Algorithm

Structural equation model is a kind of statistical analysis technique which adopts the according linear systems of equations to express the theories of cause and effect based on the known ones, aiming at exploring the causality among various things and describing them with causality mode and path figure. It is a booming branch of applied statistics field and has become a very popular data-analytic technique. As customer satisfaction index was required by a series of ISO9000 criterions, the calculation plays an important part in the applications of structural equation model. At present, partial least square is the most important algorithm for structural equation model. Structural equation model is not only applicable to the customer satisfaction index analysis, but also to other psychoanalyses and various evaluations besides those. So it plays an important part in practical application.This paper mainly studies some problems on structural equation model and the partial least square algorithm. It improves the partial least square approach for structural equation model based on the previous theory and widely applies the theory analysis united practice in psychology and sociology as well as other fields, especially in customer satisfaction index model. It firstly proposes an improved partial least square algorithm in structural equation model using a suitable iterative initial value with constraint of unit vector. The algorithm enhances the convergence rate greatly and its convergency is ensured. According to the character of models proposed, the union equation models are transformed into common regression models and the convergency of PLS algorithm is proved. For the final computational formula for CSI, we present a robust and comprehensive one. When we consider multi-level SEM and take the lower level structural variables as independent variables, our algorithm is suitable for the model too. Then multi-group SEM is investigated and distributed computing is adopted to calculate all the coefficients in SEM of each group which will be regarded as observation variables. Furthermore, a uniform model is built using the generalized linear model with convex constraint and an algorithm for the multi-group SEM is presented. The programs are completed.This paper takes technology commercialization success index model for example to show the validity and optimality of the improved approaches using the simulated results. The computation adopts ordinary least square, partial least square with an arbitrary iterative initial value and partial least square with a suitable iterative initial...