Robustness, Generalizability, and Heterogeneity of Dynamic Networks of Psychopathology

The network perspective of psychopathology proposes that mental disorders arise from dynamic interactions between psychopathology-related variables. This study explored the robustness, generalizability, and heterogeneity of dynamic networks of psychopathology using Ecological Momentary Assessment data of 176 university students with varying degrees of subclinical psychopathology (M = 21.9 years, SD = 2.8; 83.3% female).

Robustness-i.e., how precisely model parameters are estimated-of nomothetic networks was assessed via case-dropping bootstrapping. Heterogeneity was analyzed using the Individual Network Invariance Test (INIT) per pair of individuals. Generalizability (i.e., how much group-derived estimates reflect individual processes) was evaluated by comparing freely estimated idiographic networks (i.e., graphicalVAR) to idiographic networks where significant effects from the nomothetic network (i.e., mlVAR) were constrained to be present. This was done with 3 different γ values (i.e., 0.5, 0.25, and 0).

Results suggest that robustness was acceptable overall. Using a γ = 0.5, the nomothetic network generalized well to the majority of individuals, but not to a substantial minority. Specifically, the group model generalized well to 127 participants (73%) participants, but not to 47 (27%). However, with lower γ parameters, the group model generalized to less participants-161 participants (7.47%) at γ = 0.25 and 174 participants (0%) at γ = 0. Finally, evidence for relevant levels of inter-individual heterogeneity was found. Concretely, 3793 out of 14.878 (24.49%) pairs of individuals displayed different network structures according to the INIT test.

This heterogeneity is a partial explanation of why treatments may not work for everyone and the individual networks provide a possible point of entry to determine more personalised treatments based on homogeneous groups. Recommendations to find such groups combining data-driven and theory-driven approaches with a focus on single-case research are discussed.