Abstract:
|
Most social network models describe relations using a single level of analysis, or scale. Existing multiscale methods, such as the multiresolution stochastic block model, are built by repeating similar models at increasingly finer scales. However, we argue the questions researchers pose about social networks are fundamentally different depending on the level of analysis. At the finest level, researchers are interested in relationships among specific nodes, relational patterns like transitivity or reciprocity, and dependence on node- and edge-level covariates. These questions are addressed by a variety of models, including ERGMs and latent space models. At a higher level, researchers often focus on understanding how a network partitions in to communities and how these communities relate to one another. These are the types of questions addressed by blockmodels. Finally, at the highest level, researchers care about whether the communities themselves fall in to clusters defined by similar network structure. We propose a new class of models to simultaneously and parsimoniously capture the network structure of interest at each scale by using different network models at different scales.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.