Towards a Network Psychometrics Approach to Assessment: Simulations for Redundancy, Dimensionality, and Loadings

Abstract

Research using network models in psychology has proliferated over the last decade. The popularity of network models has largely been driven by their alternative explanation for the emergence of psychological attributes-observed variables co-occur because they are causally coupled and dynamically reinforce each other, forming cohesive systems. Despite their rise in popularity, the growth of network models as a psychometric tool has remained relatively stagnant, mainly being used as a novel measurement perspective. In this dissertation, the goal is to expand the role of network models in modern psychometrics and to move towards using these models as a tool for the validation of assessment instruments. This paper presents three simulation studies and an empirical example that are designed to evaluate different aspects of the psychometric network approach to assessment: reducing redundancy, detecting dimensions, and estimating loadings. The first simulation evaluated two novel approaches for determining whether items are redundant, which is a key component for the accuracy and interpretation of network measures. The second simulation evaluated several different community detection algorithms, which are designed to detect dimensions in networks.The third simulation evaluated an adapted formulation of the network measure, node strength, and how it compares to factor loadings estimated by exploratory and confirmatory factor analysis. The results of the simulations demonstrate that network models can be used as an effective psychometric tool and one that is on par with more traditional methods. Finally, in the empirical example, the methods from the simulations are applied to a real-world dataset measuring personality. This example demonstrated that these methods are not only effective, but they can validate whether an assessment instrument is consistent with theoretical and empirical expectations. With these methods in hand, network models are poised to take the next step towards becoming a robust psychometric tool.

A Psychometric Network Perspective on the Validity and Validation of Personality Trait Questionnaires

“…Because of this, we connect network models to latent variable models (where applicable) and highlight the substantive differences that these models imply. As a general point, we recommend at least 500 cases when performing these analyses, which is based on previous simulation studies (Christensen, 2020;.…”

Associations between symptoms of problematic smartphone, Facebook, WhatsApp, and Instagram use: An item-level exploratory graph analysis perspective

“…After fitting the model, EGA implements a community detection algorithm in order to specify the number of dimensions. Recently, it has been shown that one of the bestperforming algorithms, in terms of higher accuracy and less bias, in polytomous (e.g., ordinal) and multidimensional data is the Louvain algorithm (Christensen, 2020). It tends to perform better than, e.g., Spinglass algorithm, especially in this type of data (Christensen, 2020).…”

“…Recently, it has been shown that one of the bestperforming algorithms, in terms of higher accuracy and less bias, in polytomous (e.g., ordinal) and multidimensional data is the Louvain algorithm (Christensen, 2020). It tends to perform better than, e.g., Spinglass algorithm, especially in this type of data (Christensen, 2020). The Louvain algorithm uses the modularity statistic to optimize its partitions (Blondel, Guillaume, Lambiotte, & Lefebvre, 2008).…”

“…The Louvain algorithm uses the modularity statistic to optimize its partitions (Blondel, Guillaume, Lambiotte, & Lefebvre, 2008). Firstly, as it aims to identify hierarchical structures in large networks, it exchanges nodes between communities/dimensions iteratively and evaluates the change in modularity (Christensen, 2020). Then, by creating latent nodes representing a collection of nodes, and identifying edge weights with other observed and latent nodes, the Louvain algorithm creates smaller networks (Christensen, 2020;Gates, Henry, Steinley, & Fair, 2016).…”

“…Firstly, as it aims to identify hierarchical structures in large networks, it exchanges nodes between communities/dimensions iteratively and evaluates the change in modularity (Christensen, 2020). Then, by creating latent nodes representing a collection of nodes, and identifying edge weights with other observed and latent nodes, the Louvain algorithm creates smaller networks (Christensen, 2020;Gates, Henry, Steinley, & Fair, 2016). The community to which a node belongs to is formed deterministically (e.g., without the researcher's direction) by the proportion of connected edges between nodes: densely connected edges form a community (Christensen & Golino, 2019).…”