Comparing the Clique Percolation algorithm to other overlapping community detection algorithms in psychological...

Area of Science:

• Psychological network analysis
• Network science
• Computational psychology

Background:

• Traditional community detection algorithms in psychological networks assign nodes (symptoms) to single communities, failing to identify overlapping symptoms.
• The clique percolation (CP) algorithm can detect overlapping symptoms, but its performance in psychological networks is not well-established.
• Existing methods lack robust evaluation for identifying overlapping symptoms within complex psychological network structures.

Purpose of the Study:

• To compare the performance of clique percolation (CP) with different parameter settings (CPMod, CPRat, CPEnt) against other methods for detecting overlapping symptoms in psychological networks.
• To evaluate the ability of these algorithms to accurately identify the number of latent factors (communities) and observed variables with cross-loadings (overlapping symptoms).
• To determine the most effective algorithm for identifying overlapping symptoms in psychological networks under diverse simulation conditions.

Main Methods:

• Simulations were conducted under 972 conditions, varying data categories, number of factors, variables per factor, factor correlations, factor loading sizes, proportion of overlapping variables, and sample sizes.
• Performance was assessed using metrics such as the Omega index, Mean Bias Error (MBE), Mean Absolute Error (MAE), sensitivity, specificity, and the mean number of isolated nodes.
• Compared clique percolation variants (CPMod, CPRat, CPEnt) with Exploratory Factor Analysis and the Walktrap algorithm modified for overlap (Walk-Ov).

Main Results:

• The Walktrap algorithm modified to consider overlap (Walk-Ov) demonstrated superior performance across the majority of simulated conditions.
• CP variants showed varying performance, with none consistently outperforming Walk-Ov in identifying both the number of communities and overlapping symptoms.
• The study identified specific conditions under which different algorithms performed better or worse, highlighting the complexity of psychological network structures.

Conclusions:

• The Walk-Ov algorithm is recommended for identifying communities with overlapping symptoms in psychological networks due to its robust performance.
• Accurate detection of overlapping symptoms is crucial for understanding the complex interplay of psychological constructs.
• Further research should explore the application of Walk-Ov in real-world psychological network data to validate simulation findings.