-
The Development of Multidimensional Latent Variable Modeling: Dual Perspectives of Hierarchy and Overlapping
- GU Honglei, WEN Zhonglin
-
Psychological Development and Education. 2025, 41(5):
750-760.
doi:10.16187/j.cnki.issn1001-4918.2025.05.15
-
Abstract
(
)
HTML
(
)
PDF (1862KB)
(
)
Save
-
References |
Related Articles |
Metrics
Empirical psychological, educational, and managemental researches are frequently confronted with multidimensional latent variables (e.g., emotional intelligence, job burnout). The measurement model of multidimensional latent variables is established through confirmatory factor analysis (CFA), bifactor models, exploratory structural equation models (ESEM), bayesian structural equation models (BSEM), bifactor exploratory structural equation models (bifactor-ESEM), and bifactor Bayesian structural equation models (bifactor-BSEM). BSEM and bifactor-BSEM models are developed within the Bayesian statistical framework, while other models are proposed on the base of frequency statistics. Although domestic scholars are familiar with CFA and have a certain understanding of bifactor model, ESEM and BSEM, they know little about bifactor-ESEM and bifactor-BSEM. The latter two models not only allow the coexistence of global and specific constructs (i.e., both general and specific factors in the models), but also incorporate the effects of non-target subdomains on items (i.e., cross-loadings between specific factors). First, this study reviews and comments on the traditional modeling methods of multidimensional latent variables. Second, we introduce the characteristics of bifactor-ESEM and bifactor-BSEM and their applications in social science research (e.g., psychology). Third, a flow chart of multidimensional latent variable modeling is presented, which balances the simplicity and the precision. Finally, an empirical example is illustrated to demonstrate the use of various modeling methods of multidimensional latent variables.