Date of Award
Summer 2025
Document Type
Open Access Thesis
Department
Psychology
First Advisor
Amanda Fairchild
Abstract
Cross-loadings are a common phenomenon in psychological research. Extant studies have focused primarily on measurement models, like confirmatory factor analysis (CFA), and have shown that omitted cross-loadings can bias the inter-factor correlation. These results suggest that omitted cross-loadings might likewise bias structural paths in other structural equation models (SEMs), but this area remains under-investigated. Such an effect could impact the evaluation of third variable effects, like mediation, when estimated in a SEM framework.
This study addresses this gap by examining the effects of cross-loadings in single, latent mediator model in a simulation study. The performances of conventional SEM and Bayesian SEM (BSEM) approaches were compared in the presence of omitted cross-loadings on the mediator through a simulation study with manipulation on the location, number, and magnitude of cross-loadings, as well as the magnitude of structural paths in the model. Study outcomes are model fit, estimation bias, Type I error rates, and power in relation to making mediation effect inferences. Results show that the additional modeling flexibility afforded by BSEM yields better statistical performance. Results provide insights into recommended strategies for managing cross-loadings in latent mediator models, as well as suggest directions for future research in more complex mediator models within simulation studies.
Rights
© 2025, Qiulin Lu
Recommended Citation
Lu, Q.(2025). Investigating the Impact of Cross-Loadings on Model Fit and Mediation Inferences in the Latent Mediator Model. (Master's thesis). Retrieved from https://scholarcommons.sc.edu/etd/8427