AniDomNet: A sequential pairwise model for inferring dynamic animal dominance hierarchies

Abstract

Inferring dominance hierarchies is key to quantifying social dynamics within animal groups. Observed dyadic agonistic interactions remain an important source of data for studying dominance hierarchies. As a result, numerous (statistical) approaches attempt to derive and characterize dominance hierarchies from dyadic interactions. However, most of them ignore the temporal component of these interactions. We introduce a novel model to characterize dominance hierarchies using a sequential pairwise relationship model called Animal Dominance Network (AniDomNet). This model is inspired by the Elo ranking model, yet relaxes several of the underlying assumptions and allows us to study the dynamics of hierarchy formation. While addressing certain shortcomings of the current sequential methods, AniDomNet also excels at predicting the outcome of future interactions. Moreover, we propose a social agony-based approach to obtain a directed acyclic graph (DAG) that represents the dominance hierarchy according to a fitted model. AniDomNet is shown to be a useful tool to detect mistakes (such as identity switches) made during the observation process.