Agent‑based null models  for examining experimental social  interaction networks

We consider the analysis of temporal data arising from online interactive social experiments, which  is complicated by the fact that classical independence assumptions about the observations are not  satisfed. Therefore, we propose an approach that compares the output of a ftted (linear) model from  the observed interaction data to that generated by an assumed agent-based null model. This allows  us to discover, for example, the extent to which the structure of social interactions difers from that  of random interactions. Moreover, we provide network visualisations that identify the extent of  ingroup favouritism and reciprocity as well as particular individuals whose behaviour difers markedly  from the norm. We specifcally consider experimental data collected via the novel Virtual Interaction  APPLication (VIAPPL). We fnd that ingroup favouritism and reciprocity are present in social  interactions observed on this platform, and that these behaviours strengthen over time. Note that,  while our proposed methodology was developed with VIAPPL in mind, its potential usage extends to  any type of social interaction data.