Computing a ranking network with confidence bounds from a graph-based Beta random field.

Area of Science:

• Computational Social Science
• Network Analysis
• Statistical Modeling

Background:

• Determining accurate social hierarchies is crucial but challenged by data limitations and uncertainty.
• Existing methods often overlook network relevance and the impact of chance on ranking outcomes.
• Understanding dominance and conflict in social networks requires robust analytical frameworks.

Purpose of the Study:

• To develop a novel computational approach for constructing robust ranking hierarchies from pairwise conflict data.
• To address the fundamental issues of identifying relevant network information and quantifying the effect of chance.
• To provide a method that accounts for uncertainty arising from limited data and data heterogeneity.

Main Methods:

• Constructed a random field with Beta distributions from pairwise conflict outcomes to model uncertainty.
• Evaluated network information using information transitivity and synthesized transitive dominance odds.
• Fused direct and indirect dominance information using coupled random matrices and simulated annealing for optimal ranking networks.

Main Results:

• Developed a computational approach suitable for large, heterogeneous datasets of conflict outcomes.
• Demonstrated the infeasibility of classical maximum-likelihood approaches due to ignored network information.
• Derived conditional statistical inferences to manifest the effect of uncertainty on network features.

Conclusions:

• The proposed method effectively integrates relevant network information and accounts for chance, leading to more accurate social hierarchy rankings.
• This approach offers a significant improvement over traditional methods by explicitly handling uncertainty and data heterogeneity.
• Validated on real-world datasets from college football and rhesus macaque societies, showcasing its practical applicability.