Hierarchical, Continuous-Time Models for Network-Based Event Data

• Activity Number: 578
• Type: Contributed
• Date/Time: Wednesday, August 1, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Christopher L DuBois*+ and Padhraic Smyth and Carter T. Butts
• Companies: University of California at Irvine and University of California at Irvine and University of California at Irvine
• Keywords: dynamic networks ; hierarchical Bayesian models ; time-to-event data

Abstract

Interaction within small groups can often be represented as continuous-time relational data, where each observed event involves a sender and a recipient. Recent methods model the mechanisms guiding a sequence of such events by parametrizing the rate at which individuals interact in terms of the previous history of events and actor covariates (Butts 2008). We present a class of hierarchical Bayesian models for the situation where one observes many such sequences. In addition to facilitating inferences about event-level dynamics and their variation across sequences, this approach helps share information between sequences. An application to high school classroom dynamics is used to illustrate the benefits of this approach. Covariates about the students, teacher, and classroom settings are included in the model, as well as indicators for typical participation shifts common to human conversation (Gibson 2003). We present estimates from the model and describe their implications. Finally, we describe adequacy checking and model selection methods for this class of models.