A preliminary comparison between proximity and interaction-based methods to construct equine (Equus caballus) social networks

Abstract

Evidence suggests that keeping horses in groups may be beneficial, as increased opportunities for social interaction have been linked to improved welfare and trainability. Considering social structure within these groups is also recommended, so attempts may be made to minimize inter-horse aggression and subsequent injury risk, thus encouraging more owners to adopt this management practice. Manual observation of dyadic interactions is often considered the most reliable way to determine group structure. However, alternative methods, such as the use of inter-individual proximity, may be more practical but first requires validation in the species of interest to ensure reliability. Four interaction-based methods, which considered (1) ‘All observed’, (2) ‘Affiliative’, (3) ‘Allogrooming’ and (4) ‘Agonistic’ equine interactions, were used to construct social networks for 3 small domestic horse groups following 20-hour of observation. Horses also wore Global Positioning System units, so distance between group members could be calculated every 10-minute, with this information used to create proximity networks for each group. Mantel tests were run in Socprog 2.9 to determine if networks based on observed interactions are structurally similar to those based on inter-individual proximity. Accelerometers were also used to monitor horse activity, to investigate the effect that filtering proximity data by activity level has on its agreement with interaction-based methods. Mantel tests identified that proximity networks were similar to networks based on affiliative interactions between horses, with positive but non-significant agreement seen in all 3 groups (‘Group A’: Z = 0.85438, n = 4, P = 0.05; ‘Group B’: Z = 0.61582, n = 3, P = 0.475; ‘Group C’: Z = 0.88925, n = 4, P = 0.05) Proximity was not seen to be significantly associated with any other methods. These findings suggest that Global Positioning System-derived proximity may be a viable alternative to manually collected data when affiliative interactions are of interest. Although, more work is warranted to establish how generalizable these results are in larger groups, and how variables, such as field size, group composition and resource provision, influence method agreement. Ultimately, this study has assessed agreement between existing social network techniques, whille also considering the costs associated with each, the results of which are of use to inform both equine management and future studies.