Sunbelt XXXI International Network for Social Network ... - INSNA

Cross‐sectional Approximations To Separable Temporal Ergm ParametersCarnegie, Nicole B.; Hunter, David; Goodreau, Steven M.Network DynamicsDynamic NetworksWED.PM1There has been a great deal of interest recently in the modeling and simulation of dynamic networks. One promising model is the separable temporal ERGM ofKrivitsky and Handcock, which treats the formation and dissolution of ties in parallel in each time step as independent ERG models. However, thecomputational burden for fitting these models can be substantial, particularly for large, sparse networks. Fitting cross‐sectional network models, while still anon‐negligible computational burden, is much more efficient than the full dynamic fit. In this paper we show that an analytic adjustment to the cross‐sectionalnetwork parameters based on the mean duration of relationships is an adequate approximation to the dynamic parameters for sparse networks withrelationships of moderate duration. We consider a variety of cases: Bernoulli formation and dissolution of ties, dyad independent formation and Bernoullidissolution, dyad independent formation and dissolution, and dyad dependent formation models.Cumulative Properties Of Elementary Dynamic NetworksGulyas, Laszlo; Khor, Susan; Legendi, Richard; Kampis, GeorgeNetwork DynamicsNetwork Dynamics, Dynamic Network Analysis, Centrality, Dynamic Networks, Clustering, Degree distributionsWED.PM1In this paper we create different elementary dynamic network models as an attempt to study the baseline properties of networks changing in time (i.e.,dynamic networks). Our goal is to understand the inherent properties brought about by the changes themselves (such as in degree distributions), and todevelop elementary models of these processes, similar in kind to the classic models for static networks (e.g., Erdos‐Renyi, Watts‐Strogatz and Barabasi‐Albert).Focusing on edge dynamics, we define 3 dynamic versions of "Erdos‐Renyi" models, one of which can also interpreted as a dynamic Watts‐Strogatz networkmodel, and we consider a variety of dynamic "Preferential Attachment" versions. Working with dynamic networks (either collecting longitudinal samples ofnetworks or trying to model the evolution of networks over time), we have to realize that sampling always involves an act of aggregation (thereby cumulatinginteractions that happened in a certain time window into one single static network instance). This issue is unavoidable, but poorly studied. Therefore, we payspecial attention to the choice of the selected cumulation time window and its effect to aggregated network properties (density, diameter, degree distribution,clustering, etc.).