Date created: | Last Updated:

Category: Project

Description: Pairwise Markov random field networks-including Gaussian graphical models and Ising models-have become the "state-of-the-art" method for psychopathology network analyses. Recent literature has focused on the reliability and replicability of network characteristics (e.g., edges and node centrality), with a corresponding rapid evolution in the methods for quantifying the stability and consistency of these characteristics in PMRF networks (e.g., the new bootnet and NetworkComparisonTest packages in R). In the present study we examined the stability and consistency of two PMRF networks of depression and anxiety symptoms estimated in two waves of data from an observational study (n = 403). We assessed the networks using both the latest toolbox of methods and a variety of complementary metrics. All methods highlighted the unreliability of node centrality estimates. Taken on face value, the latest toolbox of methods suggested that edge estimates were reliable and stable, but the complementary metrics suggested that this was not the case. For example, there were 97 edges estimated between the two waves, but only 55 (56.7%) were replicated; 38 (39.2%) were unique to a single network; and 4 (4.1%) reversed sign (e.g., from positive to negative) between waves. We discuss reasons for these apparently contradictory results, and conclude that the limited reliability of the detailed characteristics of networks observed here is likely to be common in practice. Poor replicability of these detailed characteristics underpins our concern surrounding the use of these increasingly popular psychopathology network methods, given that generalizable conclusions are fundamental to the utility of their results.