bioRxiv (Cold Spring Harbor Laboratory), Год журнала: 2024, Номер unknown
Опубликована: Ноя. 1, 2024
Abstract Low-density social networks can be common in animal societies, even among species generally considered to highly social. Social network analysis is commonly used analyse societal structure, but edge weight (strength of association between two individuals) estimation methods designed for dense produce biased measures when applied low-density networks. Frequentist suffer data availability low, because they contain an inherent flat prior that will accept any possible value, and no uncertainty their output. Bayesian alternative priors, so provide more reliable weights include a measure uncertainty, only reduce bias sensible values are selected. Currently, neither accounts zero-inflation, estimates towards stronger associations than the true network, which seen through diagnostic plots quality against output estimate. We address this by adding zero-inflation model, demonstrate process using group-based from population male African savannah elephants. show approach performs better frequentist caused these problems, though requires careful consideration priors. recommend use framework, with conditional allows modelling zero-inflation. This reflects fact derivation two-step process: i) probability ever interacting, ii) frequency interaction those who do. Additional priors could added where biology it, example society strong community such as female elephants kin structure would create additional levels clustering. Although was inspired reducing observed sparse networks, it have value all densities.
Язык: Английский