Skip to main content

A Unified Complex: Conditionally Independent Dyadic Models for Multiple Complex Networks

The usefulness of generative models for complex networks is felt in many different disciplines, including psychology, biology, education and business, as a means of capturing important considerations in how individual units interact. Choosing the appropriate model family for such a network, or a series of networks, is a more difficult task, particularly as the complexity of these models increases. To make this task both theoretically clear and practical to execute, I expand the existing family of network models into a single family that allows for both the specification of multiple networks, the combination of multiple types of dependence structure, and clean model selection using in- and out-sample methods. I demonstrate these methods on several network data sets using our software in development.

This work is joint with Tracy Sweet, Brian Junker, Beau Dabbs and Mauricio Sadinle.