The role of weighted and topological network information to understand animal social networks: a null model approach

Abstract

Network null-models are important for drawing conclusions about individualand population-(or graph) level metrics. While the null-models of binary networks are well-studied, recent literature on weighted networks suggests that: i) many so-called weighted metrics do not actually depend on weights, and ii) many metrics that supposedly measure higher-order social structure actually are highly correlated to individual-level attributes. This is important for behavioural ecology studies where weighted network analyses predominate, but there is no consensus on how null-models should be specied. Using real social networks, we developed 3 null-models that address two technical challenges in the networks of social-animals: i) how to specify null-models that are suitable for proportion-weighted networks based on indices such as the half-weight index; and ii) how to condition on the degree-and strength-sequence and both. * Corresponding author. E-mail: robertw.rankin@gmail.com Animal Behaviour (2016) 113:215-228 doi:10.1016/j.anbehav.2015 We compared 11 metrics with each other and against null-model expectations for 10 social networks of bottlenose dolphin (Tursiops aduncus) from Shark Bay, Australia. Observed metric values were similar to null-model expectations for some weighted metrics, such as centrality measures, disparity and connectivity, whereas other metrics such as anity and clustering were informative about dolphin social structure. Because weighted metrics can dier in their sensitivity to the degree-sequence or strength-sequence, conditioning on both is a more reliable and conservative null model than the more common strength-preserving nullmodel for weighted networks. Other social structure analyses, such as community partitioning by weighted Modularity optimisation, were much less sensitive to the underlying null-model. Lastly, in contrast to results in other scientic disciplines, we found that many weighted metrics do not depend trivially on topology; rather, the weight distribution contains important information about dolphin social structure.

A guide to null models for animal social network analysis

“…An issue arising with these models is that they work directly with the observed networks, without consideration of underlying uncertainty or data collection methodology. Although ERGMs have been used with animal social networks (Ilany, Booms & Holekamp 2015;Rankin et al 2016), they also do not incorporate a permutation procedure for testing hypotheses.…”

“…Although ERGMs have been used with animal social networks (Ilany, Booms & Holekamp ; Rankin et al . ), they also do not incorporate a permutation procedure for testing hypotheses.…”

Network measures in animal social network analysis: Their strengths, limits, interpretations and uses

“…social system and group size) and because several other network measures appear correlated with it. Density is correlated with degree distribution (see corresponding section), geodesic distances (see corresponding section) and the frequency of micro‐motifs, like closed triplets* and thus clustering coefficients (see corresponding section; Rankin et al, 2016). These correlations between density and other global network measures make it necessary to control for network density when comparing global network measures from different groups, conditions or species.…”

Indirectly connected: simple social differences can explain the causes and apparent consequences of complex social network positions

“…Many questions remain regarding the importance of indirect network connections and what these metrics can tell us about animal social behaviour [13,17]. Indirect connections are, by definition, an emergent feature of associations between pairs of individuals.…”