Title: Dynamic network models with global temporal dependence structure
Abstract: In this talk a dynamic network model is presented which allows connection structure to be governed by both global network statistics (such as centrality or graph triangles) as well as co-variate data that is known a priori. Moreover, node level classiﬁcation is allowed to link to an overall network regime that inﬂuences connectivity. In practice it will often be the case that agent classiﬁcations may or may not be observed before edges are observed. In this vein, we impose structure on the classes and a latent structure on edge probabilities. This ﬂexibility allows forecasting whether true node classiﬁcations are observed or unavailable at the time. We analyze a dynamic network that describes the co-voting patterns among U.S. Senators from 1867 to 2015. We model the co-voting tendencies of the Senators using our dynamic network where nodes represent Senators and an edges are formed between two nodes if the two Senators vote concurrently (both yay or both nay) on bills to which they were both present.