A goodness of fit framework for relational event models

Abstract

Abstract We introduce a novel procedure to assess the goodness of fit in relational event models. Building on existing auxiliary variable approaches developed in network modelling, the procedure involves a comparison between statistics computed on observed relational event sequences and statistics calculated on event sequences simulated from the fitted model. We argue that the internal time structure of the relational mechanisms assumed to generate the observations under the model is an important aspect of the fit of a model to observed relational event sequences. We establish the empirical value of the proposed goodness of fit approach in an analysis of data that we collected on collaborative patient-referral relations among healthcare organizations. The illustrative case study that we develop reveals distinctive features of relational event models that have been ignored or overlooked in received empirical studies.