Iterated community detection in psychological networks.

Area of Science:

• Network analysis
• Psychometrics
• Computational statistics

Background:

• Psychological network models often exhibit community structures.
• The Spinglass algorithm detects communities but is nondeterministic, leading to variable results.
• Current methods lack ways to determine optimal solutions or assess instability in network psychometrics.

Purpose of the Study:

• Introduce and evaluate Iterated Community Detection: Spinglass (IComDetSpin).
• Aggregate multiple Spinglass iterations to find the most frequent solution.
• Quantify and visualize solution instability and community boundary fuzziness.

Main Methods:

• Developed IComDetSpin to aggregate results from multiple Spinglass algorithm runs.
• Conducted two simulation studies to evaluate IComDetSpin's performance.
• Compared IComDetSpin against single-iteration Spinglass, Walktrap, and Exploratory Graph Analysis.

Main Results:

• IComDetSpin outperformed single-iteration Spinglass in identifying the true number of communities.
• IComDetSpin performed comparably to Walktrap in Study 1.
• In Study 2, IComDetSpin outperformed Exploratory Graph Analysis under specific conditions (high inter-community correlation, fewer nodes per community).

Conclusions:

• IComDetSpin enhances Spinglass performance for community detection in psychological networks.
• The method provides valuable insights into the stability of community detection results.
• IComDetSpin offers unique information on the fuzziness of community structures, unavailable with single-iteration methods.