A0457
Title: Dynamic network models with time-varying nodes
Authors: Luca Gherardini - University of Klagenfurt (Austria) [presenting]
Monia Lupparelli - University of Florence (Italy)
Mauro Bernardi - University of Padova (Italy)
Abstract: Dynamic networks offer a versatile framework for examining temporal dependencies among statistical units, yielding insights into the stochastic mechanisms that drive interactions over time. An often-overlooked complexity - population dynamics - is addressed, whereby individuals may enter or exit the network during the observation window, potentially biasing inferential outcomes if unaccounted for. Focusing on binary temporal networks with evolving topologies, it distinguishes between two scenarios: (i) fully observed network evolution and (ii) partially observed topology requiring inferential reconstruction. In both contexts, a hierarchical mixed-effects model that jointly characterizes uncertainty in node-set evolution and dyadic links is proposed. To enable scalable inference, a Bayesian conjugate algorithm grounded in Polya-Gamma data augmentation is developed. The principal inferential properties of the model are derived, establishing its theoretical validity. Through simulation experiments and an application to social interaction data from a school setting, the approach is demonstrated to not only mitigate biases associated with time-varying populations but also enhance the interpretability of dynamic network patterns.
