|
Abstract:
|
Networks have become a quite popular tool for representing relational data in many applied settings, but further statistical development of easily interpretable and computationally feasible models that can deal with multiple networks is greatly needed. We motivate embedding networks in a non-Euclidean latent space, based on typical observed characteristics of real world network data. We develop a methodology to compare features across networks that may vary in size (number of nodes), and also outline a framework for multilevel modeling of networks, where networks are nested within higher-level observations. We demonstrate the use of this non-Euclidean latent space model for both simulated and real world network data, illustrating the types of substantive findings that this approach allows.
|
