What the study found
The study found that community correlations can measure edge association between two binary graphs when vertex communities are given. These correlations are zero if and only if the two graphs are conditionally independent within a specific pair of communities. The authors also define a maximum community correlation and an overall graph correlation, with the overall graph correlation equal to zero if and only if the two binary graphs are unconditionally independent.
Why the authors say this matters
The authors say these measures provide a way to test conditional or unconditional independence between two binary graphs. They also state that the resulting test is fast, valid, and consistent.
What the researchers tested
The researchers proposed community correlations based on given vertex communities and computed sample versions using graph encoder embedding, a method for representing graph data. They proved the sample community correlations converge to their population versions and derived the asymptotic null distribution to support hypothesis testing for independence.
What worked and what didn't
The abstract says the sample community correlations converge to the corresponding population quantities. It also says the derived asymptotic null distribution enables a fast, valid, and consistent test for conditional or unconditional independence between two binary graphs. Theoretical results were validated through simulations, and two real-data examples were presented using Enron email networks and mouse connectome graphs.
What to keep in mind
The abstract does not describe limitations, but it does indicate the methods are developed for binary graphs and rely on given vertex communities for the community-correlation framework. The available summary does not state performance details beyond the validation through simulations and the two examples.
Key points
- Community correlations measure edge association between two binary graphs when vertex communities are known.
- A community correlation is zero exactly when the graphs are conditionally independent within a specific pair of communities.
- The overall graph correlation is zero exactly when the two binary graphs are unconditionally independent.
- Sample community correlations were computed using graph encoder embedding and shown to converge to their population versions.
- The abstract says the resulting test is fast, valid, and consistent, and it was checked in simulations and two real-data examples.
Disclosure
- Research title:
- Graph correlations test independence between binary networks
- Authors:
- Cencheng Shen, Jesús Arroyo, J. T. Xiong, Joshua T. Vogelstein
- Institutions:
- Microsoft (United States), Texas College, Texas A&M University, University of California, Berkeley, Johns Hopkins University
- Publication date:
- 2026-04-05
- OpenAlex record:
- View
Get the weekly research newsletter
Stay current with peer-reviewed research without reading academic papers — one filtered digest, every Friday.