Comparing the Clique Percolation with other overlapping community detection algorithms in psychological networks: A Monte Carlo simulation study

Resumen

In psychological networks, one limitation of the most used community detection algorithms is that they can only assign each node (symptom) to a unique community, without being able to identify overlapping symptoms. The clique percolation (CP) is an algorithm that identifies overlapping symptoms but its performance has not been evaluated in psychological networks. In this study, we compared the CP with model parameters chosen based on fuzzy modularity (CPMod) with two other alternatives, the ratio of the two large communities (CPRat) and entropy (CPEnt). We evaluated their performance to: (1) identify the correct number of latent factors (i.e. communities); and (2) identify the observed variables with substantive (and equally sized) cross-loadings (i.e. overlapping symptoms). We carried out simulations under 972 conditions (3x2x2x3x3x3x3): (1) data categories (continuous, polytomous and dichotomous); (2) number of factors (2 and 4); (3) number of observed variables per factor (4 and 8); (4) factor correlations (0.0, 0.5, and 0.7); (5) size of primary factor loadings (0.40, 0.55 and 0.70); (6) proportion of observed variables with substantive cross-loadings (0.0%, 12.5% and 25.0%); and (7) sample size (300, 500 and 1000). Performance was evaluated through the Omega index, Mean Bias Error (MBE), Mean Absolute Error (MAE), Sensitivity, Specificity and mean number of isolated nodes. We also evaluated two other methods, Exploratory Factor Analysis and the Walktrap algorithm modified to consider overlap (EFA-Ov and Walk-Ov, respectively). The Walk-Ov displayed the best performance across most conditions and is the recommended option to identify communities with overlapping symptoms in psychological networks.

Áreas y palabras clave

Cómo citar (APA 7)

Santiago, P. H. R., Soares, G. H., Quintero, A., & Jamieson, L. (2022). Comparing the Clique Percolation with other overlapping community detection algorithms in psychological networks: A Monte Carlo simulation study. https://doi.org/10.3758/s13428-024-02415-2