Abstract
This study evaluated and compared one-step and two-step dynamic structural equation modeling (DSEM) approaches for analyzing intensive longitudinal data. In one-step DSEM, the within-person dynamic model and the between-person model are estimated simultaneously. In two-step DSEM, dynamic component estimates are first extracted from the within-person model, followed by estimation of between-person relations in a second step. We further examined whether including person-level variables as auxiliary variables in the first step affects the performance of two-step DSEM. The modeling approaches were evaluated by a simulation study and illustrated with a real data example. Results showed that two-step DSEM without auxiliary variables exhibited estimation bias, low coverage rates, and deflated Type I error rates. In contrast, one-step DSEM and two-step DSEM with auxiliary variables yielded satisfactory results when sufficient data were available (e.g., ≥ 30 time points and ≥ 100 individuals). Implications, modeling recommendations, and future research directions were discussed.