ABSTRACT
In broadening the scope of network analysis beyond the conventional paradigm, the framework of edge exchangeability put forward in Chapter 9 takes a small step toward realizing the broader vision of complex data analysis laid out in Chapter 1. This chapter expands upon the previous toward a theory for network data constructed from a sample of arbitrary relations. In Section 9.1, for example, an edge‐labeled network is constructed out of pairwise interactions (i.e., phone calls) between callers and receivers sampled from a phone call database. More generally, networks can be built from repeated observations of any relational structure, including pairwise interactions (i.e., edges) as in Figure 9.5, multiway interactions (i.e., hyperedges), paths (i.e., ordered multisets), or even networks (i.e., graphs) or arbitrary relational structures (Xl, ⋯, X r ) as in Section 8.5. Networks constructed from an exchangeable sequence of such relations are called relationally exchangeable [53]. I focus here on a few special cases of relationally exchangeable network models.
