Exploratory Graph Model

Network psychometric models have emerged as an alternative to latent variable frameworks for conceptualizing psychological phenotypes. Despite their growing popularity, the formalization of network models as a measurement model has been largely unexplored. Recent efforts have integrated network models within the structural equation modeling (SEM) framework, but no standalone framework exists to specify network models as a measurement model, particularly with consideration of dimensions. This paper introduces the Exploratory Graph Model (EGM), a mathematical framework that formalizes networks as a measurement model that includes community structures where dimensions arise from the mutual interactions between observed variables rather than caused by latent constructs. After formalizing the model, a data generation approach is parameterized and a Monte Carlo simulation study is performed to demonstrate the differences in parameter estimates between EGM and exploratory factor analysis (EFA) when data are generated from EGM and latent factor models. We further explore whether the data generating model can be determined based on traditional SEM and likelihood statistics. An empirical example is presented to showcase how EGM can be applied and compared against EFA and alternative EGM structures. EGM broadens traditional views on how psychological phenotypes can be conceptualized, providing a network-based understanding of psychological measurement.