An Investigation of Bayesian Approximate Measurement Invariance with Longitudinal Data
Abstract
The Bayesian structural equation modeling approach was introduced by Muthén and Asparouhov (2013) to test measurement invariance (MI) across multiple groups or multiple time points. This approach estimates approximate measurement invariance by applying the zero-mean, small variance prior to the difference parameters between groups. The approach has been implemented in applied research. However, there is a lack of simulation research on the Bayesian Approximate MI with longitudinal data. Longitudinal MI is a critical assumption of modeling the underlying subjective growth or change over time. This study explored the performance of this approach with a comprehensive simulation study and provided practical suggestions in the choice of the methods when dealing with longitudinal data. Results suggested that the Bayesian A-MI approach is appropriate for testing longitudinal measurement invariance with continuous items, but not recommended for ordered-categorical data. This study also recommended using DIC, WAIC, and LOO for model selection, rather than Bayesian approximate indices.
Subject Area
Quantitative psychology|Psychology|Statistics
Recommended Citation
Feng, Ye, "An Investigation of Bayesian Approximate Measurement Invariance with Longitudinal Data" (2023). ETD Collection for Fordham University. AAI30426833.
https://research.library.fordham.edu/dissertations/AAI30426833
