Evaluating Cross-Lagged Models: A Comparative Analysis Using Simulated and Empirical Data

• In Event: Statistical Frontiers: Bayesian Inference, Machine Learning, and Longitudinal Modeling

Abstract

Understanding reciprocal relationships over time is essential in developmental and psychological research. However, traditional models like the cross-lagged panel model (CLPM) conflate within- and between-person variance, limiting causal inference. This study compares four longitudinal models—CLPM, random intercept cross-lagged panel model (RI-CLPM) with observed and latent variables, and the dynamic panel model (DPM) using simulated and real-world panel data on psychological distress and life satisfaction. Results show that RI-CLPM with latent variables and DPM outperform CLPM in model fit and parameter accuracy, while RI-CLPM with observed variables tends to underestimate effects. These findings underscore the importance of accounting for unobserved heterogeneity and offer practical guidance for selecting appropriate longitudinal models to better capture temporal dynamics.

Authors

• Xiaoyu Yang, Texas A&M University

• Lu Chen, Texas A&M University

• Su Jiang, University of Houston Downtown

• Oiman Kwok, Texas A&M University