Constructing and analysing time‐aggregated networks: The role of bootstrapping, permutation and simulation

Abstract

Abstract Animal social networks are often used to describe dynamic social systems, where individual behaviour generates network‐level structures that subsequently influence individual‐level behaviour. This interdependence between individual behaviour and group structuring is of central concern for questions concerning the evolution and development of social systems and collective animal behaviour more generally. Various statistical methods exist for estimating network changes through time. One approach, time‐aggregated networks, takes repeated snapshots of interactions within windows of time to generate a time series of networks. However, there remain many analytical hurdles when implementing the time‐aggregated approach. To ameliorate this, we introduce an r package netTS that focuses on three analytical steps for analysing time‐aggregated networks: choosing appropriate time scale using bootstrapping, comparing patterns to relevant null models using permutation and finally building and interpreting statistical models using simulated data. We use simulated data to first highlight these steps, then use observed grooming data from a group of vervet monkeys as an applied example. Our results suggest that the use of bootstrapping and permutation can accurately extract known patterns from simulated data. Using this approach with vervet data suggests that there is consistent social structuring, differing from what would be expected due to chance, and that some individuals are contributing to this structure more than others (i.e. keystone individuals). We demonstrate that bootstrapping, permutation and simulation can aid in constructing and interpreting time‐aggregated networks. We suggest that the use of time‐aggregated networks to quantify patterns of network change can be a useful tool alongside process‐based approaches that seek mechanistic descriptions. Ultimately, by looking at both patterns and processes, dynamic networks can be used to better understand how individual behaviour generates social structures, and in turn how individual behaviour can be influenced by social structures, ultimately leading to a better understanding of the evolution of social behaviour.