Research On The Improvement Of Parameters Estimation In Structure Equation Model

Since the Structural Equation Modeling (SEM) technology was developed by Joreskog and his partners in1973, and also with the development of LISREL software, this technology have been used in different sectors of social science field. Due to the overly strict assumptions, many researchers have to abandon the Structural Equation Modeling technology. Fortunately, the partial least squares path modeling (PLS Path Modeling) give researchers another convenient choice to construct the SEM theoretical model and get the accurate estimated parameter values. This is because the PLS Path Modeling technology do not need too much strict assumptions. However, the accuracy of estimated parameter values that calculated from PLS is still not meets the requirements of many researchers. In this situation, we need to develop a new parameter estimate method, which can remedy the drawbacks of PLS. In this paper, we developed this new method based on the Double Chains Quantum Genetic Algorithm (DCQGA), which is the combination of Quantum Computing and Genetic Algorithm. The DCQGA is an advanced Quantum Genetic Algorithm (QGA) which is more effective and powerful than the common genetic algorithm. However, the DCQGA focuses on the fitness function optimization but not the actual theoretical significance of the research model. So we proposed two methods to limit the solution space of DCQGA. At first, we add prior knowledge as much as possible and transform them into different restrained conditions; then we take the results of SmartPLS as the initial parameter values for DCQGA. The new method we proposed here is called New-DCQGA. A series of simulation tests suggest that, these two methods reduced the optimal solution search space of DCQGA effectively and increase the probability to find the ideal parameter values. Moreover, we used the New-DCQGA to solve the structural equation model which includes the mediator. After a series of simulation tests which contain four kinds of hypothesis, we find that the New-DCQGA is more efficient than the SmartPLS.