Title: Comparing estimation methods for stochastic actor-oriented models
Authors: Viviana Amati - University of Konstanz (Germany) [presenting]
Felix Schoenenberger - University of Konstanz (Germany)
Tom Snijders - University of Groningen, University of Oxford (Netherlands)
Abstract: Stochastic actor-oriented models are models for dynamic network data, collected by observing a network and an actor-level variable (referred to as `behaviour') over time in a panel design. The method of moments (MoM) and the maximum likelihood estimation (MLE) have been developed to estimate the parameters of these models. Recently, focusing on dynamics of networks without a behaviour variable, we proposed an estimator based on the generalized method of moments (GMoM) and we described the algorithmic issues that had to be solved to obtain a stable estimation procedure. We apply this estimator to the interdependent dynamics of a network and a behavioural variable, complementing the usual cross-lagged statistics for the MoM by statistics that are contemporaneous combinations of network and behavioral data. We compare the efficiency of the GMoM with respect to the MoM and the MLE. Moreover, we discuss how differences in the estimated coefficients obtained by the three methods may be used to discover model misspecifications.