Factor or Network Model? Predictions From Neural Networks

Abstract

The nature of associations between variables is important for constructing theory about psychological phenomena. In the last decade, this topic has received renewed interested with the introduction of psychometric network models. In psychology, these models are often contrasted with latent variable (e.g., factor) models. Recent research has shown that the differences between the two tend to be more substantive than statistical (i.e., equivalent statistical representations but differing data generating hypotheses). Still, some researchers have shown that there may exist differences when data are generated from these models (e.g., sparse network models and unidimensional factor models). One recently developed algorithm called the Loadings Comparison Test (LCT) takes advantage of the loading structures of network and factor models to predict whether data were generated from a random, factor, or network model. A significant limitation in the current implementation of the LCT is that it’s based on heuristics that were derived from descriptive statistics of a limited set of data structures. In the present study, we used artificial neural networks to replace these heuristics, and develop a more robust and generalizable algorithm. We performed a simulation study that compared neural networks to the original LCT algorithm as well as logistic regression models that were trained on the same data. We found that the neural networks performed as well as or better than both methods, demonstrating generalizablity across data generating models. We echo the call for more formal theories about the relations between variables and discuss the role of the LCT in this process.

Centrality measures in psychological networks: A simulation study on identifying effective treatment targets

“…These results suggest that different properties might be of interest according to the nature of the network (nomothetic or idiographic). In fact, previous research has also pointed to this need of further exploration and clarification of the properties of interest in psychological network [57] and the impact of these networks' structural properties in the selection of centrality measures [58].…”

Identifying Treatment Targets in Idiographic Psychopathological Networks: A Simulation Study

“…These results suggest that different properties might be of interest according to the nature of the network (nomothetic or idiographic). Previous research has also pointed to this need for further exploration and clari cation of psychological network structure 57 and the impact of the networks' structural properties in the selection of centrality measures 58 .…”

Exploratory Graph Analysis in context

“…They showed simulation study that network loadings are akin to factor loadings, opening a new line of research within the EGA approach. These loadings have opened to door to assessing measurement (metric) invariance (Jamison, Golino, and Christensen (2022)) from the network perspective (Jamison et al, 2022) as well as determining whether data are generated from a factor or network model (Christensen & Golino, 2021b).…”