Downloads & Free Reading Options - Results

Unidimensionality is fundamental to psychometrics. In the last decade, network analysis in psychology has skyrocketed in popularity. Dimensionality assessment has been a strong focus of network psychometrics yet unidimensionality assessment remains a challenge. Community detection algorithms are the most common approach to estimating dimensionality in networks. Most of these algorithms optimize an objective function called modularity. A limitation of modularity is that it penalizes unidimensional structures and tends to favors two or more communities (dimensions). In this study, I perform a Monte Carlo simulation using one and two factor models to determine what community detection algorithms work best. Based on previous simulation studies, I compare the best performing community detection algorithms: Leading Eigenvalue, Leiden, Louvain, and Walktrap. A grid search is performed on these algorithms parameters to determine the optimal trade-off between unidimensional and bidimensional detection. The best performing parameters for each algorithm are then compared against each other. This study advances undimensionality detection in network psychometrics, further developing the psychometric toolbox of network analysis in psychology.