|
Abstract:
|
Often researchers are interested in quantifying the association between pairwise relationships (e.g. interactivity or fertility relations) and attributes of the individuals or pair. For example, in a study of barn swallow reproduction, we are interested in how social interactions, individual ancestry, migratory behavior, and morphological variation are associated with fertilization networks. The inherent dependencies among relations involving the same individuals require special consideration when performing inference. Traditional nonparametric methods rely on permutation tests, which are known to suffer from low power and are unable to support more detailed inference. Additionally, existing regression approaches typically involve complex latent variable models, which are both computationally intensive to fit and difficult to specify. In this talk we propose an alternative regression approach that leverages an assumption of exchangeability, which underlies most latent variable models. We show that this assumption implies a parsimonious dependence structure among the relations, which can be used directly to make inference on the associations of interest.
|
