Modeling Intensive Longitudinal Data: A Tutorial Comparing Dynamic Structural Equation Modeling and Psychometric Network Analysis

Intensive longitudinal data (ILD) enable modeling of within-person dynamics and their between-person differences. Dynamic structural equation modeling (DSEM) and psychometric network analysis (mlGVAR-PNA) are two prominent frameworks for this purpose. Although both build on the multilevel vector autoregressive model, they differ deeply in construct conceptualization, estimation, and the parameters obtained. This tutorial applies both frameworks to the same data, examining where their results converge or diverge and what each uniquely reveals. Ecological momentary assessments of self-esteem, life satisfaction, and optimism (the three facets of positivity), together with positive and negative affect, were collected from 166 adults three times daily for 28 days. For each framework, the workflow, assumptions, and interpretation of within-person (lagged and contemporaneous) and between-person parameters are described. In DSEM, with the facets averaged into a positivity composite, within-person increases in positive affect predicted higher positivity at the next occasion, whereas negative affect did not. In PNA, with the facets as separate nodes, they predicted one another over time, and negative affect predicted lower subsequent life satisfaction, but not optimism or self-esteem, effects the composite did not detect. At the between-person level, positive and negative affect were negatively correlated, yet their partial correlation conditional on the positivity facets was positive, showing that marginal and conditional parameters answer different questions. The frameworks should be viewed as complementary: DSEM suits testing directional, theory-driven hypotheses, whereas PNA suits mapping mutual dynamics among multiple constructs. Guidelines informing the choice between DSEM and PNA, as well as reproducible code and data, are provided.