network models
Definition
Network models refer to a class of statistical and theoretical frameworks in which observed variables, such as individual symptoms, attitudes, or behaviors, are modeled as directly influencing one another rather than as reflections of an underlying latent construct. The core assumption is that high-level psychological attributes, including disorders and personality traits, emerge from dynamic systems of lower-level interactions that can settle into stable states, such as a depressive episode or a sustained level of neuroticism. Relations are estimated at the item level, so that, for example, insomnia is modeled as directly predicting fatigue rather than both being caused by a shared latent variable. Applications span depression, schizophrenia, post-traumatic stress disorder, personality, and several other domains, using data ranging from cross-sectional samples to intensive longitudinal designs employing Experience Sampling Methods. Cross-sectional implementations typically produce sparse partial correlation networks estimated via l1-regularized regression, while longitudinal extensions capture auto-regressive and cross-lagged pathways among items over time.
Sources: Wysocki et al. (2025)
Related Terms
- cross-lagged panel design (1 shared article)
- cross-lagged networks (1 shared article)
Applications
Network Models and Cross-lagged Panel Models
Cross-lagged panel models and network models address related questions about change over time but operate at different levels of analysis. Standard cross-lagged panel models estimate predictive effects between latent constructs across widely spaced measurement occasions, whereas the cross-lagged panel network applies the same longitudinal logic to individual items, estimating auto-regressive and cross-lagged pathways directly among observed components of psychological constructs. This item-level approach can reveal specific predictive bridges between constructs, such as a link between a school-performance item and a self-worth item, that aggregate latent variable models would obscure.
Sources: Wysocki et al. (2025)
Network Models and Latent Variable Models
Network models are explicitly positioned as an alternative to latent variable models for representing psychological constructs. Where a latent variable model treats observed items as passive reflections of a single underlying cause, a network model treats the covariance among those items as arising from direct relations between them. The choice between frameworks depends on whether theory and empirical evidence favor a direct-relation account or a common-cause account of the item covariance structure.
Sources: Wysocki et al. (2025)
Network Models and Longitudinal Panel Data
Network models were initially developed for cross-sectional data and intensive time-series designs, leaving a gap for the panel data structures common in developmental research. The cross-lagged panel network fills this gap by combining regularized regression estimation with structural equation modeling to recover auto-regressive and cross-lagged item-level pathways from data collected at a small number of discrete measurement occasions. Simulation work shows this two-step estimation approach performs particularly well when relationships are non-stationary across measurement occasions.
Sources: Wysocki et al. (2025)
