Abstract This study introduces an innovative approach for analyzing longitudinal behavioral data with hidden patterns in mean (location) and intraindividual variability (scale) trajectories, using location-scale regressions with latent classes in both the location and scale parts of the model. A full Bayesian approach using Stan is adopted for the estimation of the model parameters. Using simulation studies, we demonstrate that our latent class model yields more precise and informative results, especially regarding the scale, in data exhibiting hidden patterns. Simulation results also show that our model can achieve unbiased parameter estimates as well as a high correct classification rate without over-identifying latent classes in data lacking hidden heterogeneity. Our study equips researchers with a practical tool for subgrouping subjects based on both mean and within-subject variability trajectories of longitudinal outcomes. As an illustration, the latent class model is applied to calorie intake data from a weight loss management study. The integration of latent classes into intraindividual variability trajectories of calorie intake facilitates an understanding of dietary behavior consistency, aiding in personalized weight management interventions.
Zhang et al. (Tue,) studied this question.