Mapping the dynamics of idiographic network models to the network theory of psychopathology using stability landscapes

The network theory of psychopathology posits that mental disorders are stable states of symptom activation that arise from networks of causally interrelated symptoms. Based on this theory, it can be inferred that, ceteris paribus, a person with a more strongly connected symptom network (i.e., a network characterized by more and stronger causal relationships), may have an increased risk of developing a mental disorder. Researchers have started to take an interest in idiographic networks to evaluate this theoretical position. However, there are good reasons to question whether the dynamics implied by these models make them suitable for evaluating the network theory. In this paper, we map the parameters of different types of idiographic network models to the network theory of psychopathology. In particular, we use stability landscapes to map the dynamics implied by the Ising model with a $\{0, 1\}$ and $\{-1, 1\}$ encoding, and the graphical vector autoregressive (GVAR) model onto symptom severity and the variability in symptom activation. Only the Ising model with the $\{0, 1\}$ encoding shows behavior consistent with the network theory. The Ising model with the $\{-1, 1\}$ encoding only partially aligns with the network theory, as increased network connectivity leads both the healthy and unhealthy state to become more stable. Contrary to popular belief, we illustrate that, in the GVAR model, temporal network connectivity is independent of symptom severity. Instead, in temporal network connectivity strongly relates to variability in symptom activation: more strongly connected networks are associated with more variance around the stable state. Our results suggest that more systematic investigations should precede linking psychological theory to statistical network models for empirical data. To this end, we present stability landscapes as a tool to gain insight into the dynamics of idiographic network models and as a means to link model-implied to theory-implied behaviour.