A0959
Title: Statistical inference for subgraph frequencies under exchangeable hyperedge models
Authors: Ayoushman Bhattacharya - Washington University in St. Louis (United States)
Nilanjan Chakraborty - Missouri University of Science and Technology (United States) [presenting]
Robert Lunde - Washington University in St Louis (United States)
Abstract: The problem of statistical inference is considered for subgraph counts under an exchangeable hyperedge model. Various classes of subgraph statistics for hypergraphs are introduced, and inferential tools are developed for a notion of subgraph frequency that takes into account the multiplicity of an edge. It is further shown that a certain subclass of these subgraph statistics is robust to the deletion of low-degree nodes, facilitating inference in settings where low-degree nodes are more likely to be missing. A more traditional notion of subgraph frequency is also considered, which does not take into account multiplicity. It is demonstrated that while statistical inference based on limiting distributions for these statistics is possible in certain cases, a proper limiting distribution does not even exist in certain cases. The finite sample properties of the procedures are studied in both simulation studies and real-world datasets. For data analysis, new hypergraph datasets involving academic and movie collaborations are collected, and it is found that the inferential tools for hypergraphs have more power to distinguish between networks than traditional approaches based on binary adjacency matrices.
