Moving from exploratory to confirmatory network analysis: An evaluation of SEM fit indices and cutoff values in network psychometrics

Psychometric networks allow for the investigation of inter-variable relations through uniquely identified multivariate models, without the assumption of existing latent variables. Thus, network models are well-suited for phenomena detection, and most empirical network studies have been exploratory in nature. Yet, due to the close connections between (Gaussian) network modeling and Structural Equation Modeling (SEM), the use of confirmatory modeling and SEM fit indices are readily applicable to network modeling as well. Therefore, despite the largely exploratory nature of network analysis research so far, confirmatory network psychometrics has been entirely feasible. However, no study to date has evaluated how SEM fit indices perform in confirmatory network analysis, and what criteria should be applied to such tests. In this study, we explored the possibility of utilizing SEM fit indices to detect misfit in confirmatory network analysis and the generalizability of conventional cutoff values that has been widely used in SEM. We use panel graphical autoregressive model as an illustrative example, for its generalizability to both models used in cross-sectional (Gaussian graphical models) and $N = 1$ time-series case (graphical autoregressive). Using simulations, we analyzed the performance of fit indices to test hypothesized network structures and evaluate stationarity, under varying number of nodes, sample sizes, and number of waves. Most fit indices performed well, except that Type~I incremental fit indices may reject good models at the absence of misspecifications. The conventional cutoffs in the SEM literature are largely generalizable to confirmatory network analysis as convenient assessment criteria when dynamical cutoffs are unavailable. The study validates the use of SEM fit criteria for confirmatory network psychometrics and encourages theory-testing and replication studies in network research, providing practical recommendations for the use of SEM fit indices in confirmatory network testing.