551 – Statistical Methodologies for Identifying and Modeling Structure
Edge Exchangeability: A New Foundation for Modeling Network Data
Harry Crane
Rutgers University
Walter Dempsey
University of Michigan
We provide a non-technical description and motivation for the edge exchangeable framework of network modeling. The discussion here complements our prior work, in which we introduced and developed the basic theory and philosophy of edge exchangeability in detail, and is intended to provide further details on our seminar presentations on the topic. The need for edge exchangeability as a new foundation for network modeling arises out of the observation that traditional exchangeable models for vertex labeled graphs cannot replicate the large sample behaviors of sparsity and power law degree distributions observed in many network datasets. Beyond addressing this basic issue, the edge exchangeable framework promotes a mindset that better explains key features of network analysis. It also clarifies why the conventional approach is untenable for networks constructed from processes of interactions.