A0457
Title: Bayesian mixture SEM for ordinal categorical data
Authors: Hiroki Takeshima - Doshisha University Graduate School (Japan) [presenting]
Jun Tsuchida - Kyoto Womens University (Japan)
Hiroshi Yadohisa - Doshisha University (Japan)
Abstract: To compare the relationships of constructs between different groups, data from each group is collected using a common questionnaire, such as five-point Likert scale questions, and applied multi-group structural equation modeling (MGSEM). However, when the sample size of each group is small, the parameter estimation of the MGSEM tends to be unstable. In addition, the parameter estimation performance of the MGSEM tends to decrease when it is applied to ordinal categorical data. To address these issues, in this report, an MGSEM is proposed for ordinal categorical data with clustering of groups. Specifically, a mixture model is used to represent the differences in the relationships of constructs between groups. The mixture model allows the incorporation of information on parameters from groups within the same cluster. This is expected to stabilize the parameter estimation, even for groups with small sample sizes. To treat ordinal categorical variables as continuous, a continuous latent variable behind each ordinal categorical variable is assumed. Using these continuous latent variables for continuous SEM is expected to improve the estimation performance. The results of numerical experiments show that the parameter estimation performance of the proposed method is superior to that of the MGSEM.