Investigating the Performance of Fit Indices for Testing Longitudinal Measurement Invariance in Latent Transition Analysis: A Monte Carlo Simulation Study
Korean Society for Educational Evaluation, Jaeuk Kim, Wonsook Sohn · 교육평가연구 · 2023
In latent transition analysis (LTA), the assumption of longitudinal measurement invariance (LMI) carries significant advantages, such as simplifying interpretation of transition probabilities, aiding model identification, and reducing bias in parameter estimation. Despite it’s importance, there’s a notable lack of practical methodologies for LMI testing in LTA models. This study aims to address this gap by investigating and comparing the performance of four commonly used fit indices such as AIC, BIC, SABIC, and LRT-for structural LMI tests in LTA under various conditions. This will provide empirical guidelines for researchers conducting LMI tests in the context of LTA. To this end, a simulation study was designed and conducted using the Monte Carlo method. In total, 560 conditions were manipulated, and data were generated from 500 iterations for each condition. Mplus 8.3 and the R package ‘MplusAutomation’ were used for data generation and model analysis, while the Python libraries ‘pandas’ and ‘pdfminer’ were used for preprocessing the results. The results are as follows. First, all model fit indices demonstrated favorable performance when the LMI was satisfied completely or partially. Second, the performance of the fit indices was weaker when the LMI was not met, due to the small sample size and the subtle degrees of LMI violation to be detected. Third, the AIC showed stable performance in detecting LMI violations across all conditions, followed by LRT, SABIC, and BIC. Finally, the practical implication of the results were discussed.