• Social network analysis of small social groups: Application of a hurdle GLMM approach in the Alpine marmot ( Marmota marmota )

      editor: Ebensperger, Luis; Panaccio, Matteo; orcid: 0000-0002-1903-154X; email: matteo.panaccio01@universitadipavia.it; Ferrari, Caterina; Bassano, Bruno; Stanley, Christina R.; orcid: 0000-0002-5053-4831; von Hardenberg, Achaz; orcid: 0000-0002-9899-1687; email: a.vonhardenberg@chester.ac.uk (2021-03-24)
      Abstract: Social network analysis (SNA) has recently emerged as a fundamental tool to study animal behavior. While many studies have analyzed the relationship between environmental factors and behavior across large, complex animal populations, few have focused on species living in small groups due to limitations of the statistical methods currently employed. Some of the difficulties are often in comparing social structure across different sized groups and accounting for zero‐inflation generated by analyzing small social units. Here, we use a case study to highlight how Generalized Linear Mixed Models (GLMMs) and hurdle models can overcome the issues inherent to study of social network metrics of groups that are small and variable in size. We applied this approach to study aggressive behavior in the Alpine marmot (Marmota marmota) using an eight‐year long dataset of behavioral interactions across 17 small family groups (7.4 ± 3.3 individuals). We analyzed the effect of individual and group‐level factors on aggression, including predictors frequently inferred in species with larger groups, as the closely related yellow‐bellied marmot (Marmota flaviventris). Our approach included the use of hurdle GLMMs to analyze the zero‐inflated metrics that are typical of aggressive networks of small social groups. Additionally, our results confirmed previously reported effects of dominance and social status on aggression levels, thus supporting the efficacy of our approach. We found differences between males and females in terms of levels of aggression and on the roles occupied by each in agonistic networks that were not predicted in a socially monogamous species. Finally, we provide some perspectives on social network analysis as applied to small social groups to inform subsequent studies.