Community correlations and testing independence between binary graphs

Key findings from this study

This research indicates that:

• Community correlations equal zero if and only if two graphs are conditionally independent within specific community pairs.
• Sample community correlations computed via graph encoder embedding converge to population parameters with known asymptotic distributions.
• The framework enables fast, valid, and consistent tests for conditional or unconditional independence between binary graphs.

Overview

Graph-structured data violates standard statistical assumptions, requiring tailored analytical methods. This work introduces community correlations to measure edge association between two binary graphs conditional on vertex communities. These correlations equal zero if and only if the graphs are conditionally independent within specific community pairs.

Methods and approach

The authors propose community correlations derived from vertex communities to assess edge association. Sample community correlations are computed via graph encoder embedding. Asymptotic null distributions are derived to enable hypothesis testing. Theoretical convergence properties establish that sample estimates converge to population parameters.

Results

Sample community correlations computed through graph encoder embedding converge to their population counterparts. The framework yields the maximum community correlation, which indicates conditional independence across all community pairs, and the overall graph correlation, which equals zero if and only if the two graphs are unconditionally independent. Asymptotic null distributions enable fast, valid, and consistent testing for conditional or unconditional independence between binary graphs.

Comprehensive simulations validate the theoretical results. Empirical applications demonstrate utility: an Enron email network analysis and a mouse connectome graph study illustrate the correlation measures in realistic network settings. The proposed statistics maintain validity under the derived asymptotic framework.

Implications

Valid statistical inference on graph independence requires methods accounting for network structure. Community correlations provide a principled approach to measuring conditional edge association, extending classical correlation concepts to structured data. This framework enables researchers to test independence hypotheses in biological networks, social networks, and other graph-based systems.

The availability of fast, consistent tests expands the analytical toolkit for comparative network analysis. Practitioners can now assess whether two observed networks exhibit conditional or unconditional independence at the community level or globally. These methods support rigorous hypothesis testing in domains where network comparison is substantive.

Scope and limitations

This summary is based on the study abstract and available metadata. It does not include a full analysis of the complete paper, supplementary materials, or underlying datasets unless explicitly stated. Findings should be interpreted in the context of the original publication.