Collaborative Research: The Bipartite Temporal Event Network (BiTEN) Model

Ohio State University, The, Columbus OH

Investigators

Abstract

This research project will develop a powerful and flexible statistical modeling framework that facilitates the analysis of complex network data. This model, called the bipartite temporal event network (BiTEN) model, will allow researchers to tease apart finely grained temporal dynamics along with properties of network structure, spatial geography, and any given set attributes that describe actors and/or their relationships. Across a variety of disciplines, situations arise in which actors (e.g., individuals, organizations, states, etc.) form ties to entities (e.g., products, corporations, treaties). International environmental treaty ratification is an example of a complex process through which states connect to treaties via the decision to ratify. The questions surrounding treaty ratification lie at the heart of a science of networks and specifically speak to the phenomena of diffusion, contagion, and behavioral cascades. Understanding this process is critical to making progress on issues ranging from global warming to arms control. The researchers will apply this method to a number of substantive questions, including treaty ratification, the onset of war, transnational terrorism, the creation of legislation, consumer behavior, and peer influence among students. They also will develop and disseminate free and open source statistical software. The researchers will develop a temporal process for the modeling of network dependencies that can accommodate the vital actor- and relationship- level effects. The statistical challenge in this project stems from the fact that the structure in which the actors tie to the entities affects subsequent ties. That is, the structure of the network of relationships may drive its own evolution -- a phenomena familiar in network science. The researchers will develop two estimation strategies for the model: a Bayesian approach and a maximum likelihood approach. The models developed will reliably estimate what portions of a phenomenon are driven by latent state network structure, actor attributes, relational attributes, spatial geography, and temporal dynamics. The models also will leverage all of these dynamics to probabilistically predict link formation.