Latent variable approaches to network modeling have great value as they are able to represent complex structure in an interpretable way. Initial approaches postulated the existence of a latent social space and used the geometry of that space to structure the propensity of social actors to have ties. These models have been extended to model networks that change over time and where the number and characteristics of the actors also changes. The complex interactions of these processes presents challenges both for parsimonious modeling and inference.
This talk will introduce spatial temporal exponential-family of point processes (STEPP) models to jointly represent the co-evolution of social relations and individual behavior in discrete time. The models use exponential-family models for the social structure conditional on latent variables. Temporal dynamics are modeled as a discrete Markov process specified through individual transition distributions for each actor in the system at a given time. We analyse the model and develop likelihood-based inference. Finally, we present a case-study by analyzing alcohol and drug use over time in the context of adolescent friendship networks.
|