Optimizing the structure of a path analysis model using a real-valued flexibly connected neural network

Abstract

The path analysis model (PAM) is a multivariate statistical modeling technique widely used in the behavioral and social sciences. Although some methods for optimizing the parameters and reducing the variables in PAM have been proposed, only a few studies have focused on flexible optimization of the structure and parameter in PAM. In this study, we used a real-valued flexibly connected neural network (RFCN) to construct PAM. Using survey data, we then confirmed the validity of our approach from two viewpoints. First, we assessed our approach using statistical fitness indices. Then, we compared the obtained results with those obtained from previous studies on consumer psychology. The results confirmed that our proposed approach offers a novel way of constructing PAM using RFCN.