Modeling Phase Transitions of Longitudinal Networks in the Behavioral Sciences: An Overview of the Literature

Understanding mental health as a complex system has motivated the adoption of time-varying network models to explore how the dynamics of psychological variables change over time. While existing time-varying network models assume gradual change, shifts in symptom dynamics such as transitions between healthy and disordered states may occur rapidly. This could manifest as abrupt, qualitative changes between distinct states, often labelled as regime switches. In this scoping review, we identified and synthesized statistical approaches for modeling such phase transitions in the behavioral sciences that could be applied to network models. Our review yielded 40 studies across different fields, featuring 56 statistical analyses in total. Most analyses (71.4%) were based on Hidden Markov Models, which can be extended flexibly to model unobserved, recurrent states in various scenarios. Other approaches included change point models (14.3%), which identify shifts directly in the observed time series, and threshold models (12.5%), which model regime switches triggered when a dedicated threshold variable is crossed. We describe these approaches, their applications, and their data requirements, and assess their strengths and limitations for modeling shifts in the dynamics of time series. Notably, several approaches already include multilevel or multivariate-extensions, making them promising candidates for modelling changes in network dynamics in mental health. In sum, our review highlights the broad potential of regime-switching models to enrich the understanding of dynamics of mental health processes.