Application of social network analysis to understand acute and chronic post-mixing aggression in commercially reared pigs

Abstract

In commercial systems, pigs are routinely regrouped with unfamiliar conspecifics which leads to physical aggression in order to establish new dominance relationships. Post-mixing aggression lasts approximately 24 hours before steeply declining, although chronic aggression amongst familiar individuals is also observed. These aggressive encounters result in skin injuries commonly referred to as lesions, the number and location of which have been shown to correlate with the type and duration of aggression an individual has engaged in. Correlations between anterior injury rates at 24 hours post-mixing (24hr-PM) and 3 weeks post-mixing (3wk-PM) are inversely related, indicating that reciprocal aggression can only be delayed and not avoided altogether. In order to meaningfully improve animal welfare, a solution that leads to a reduction in aggression at both time points needs to be identified. For this to be achieved, the variation in skin lesions needs to be studied further and better understood. Previous analyses have focussed on interactions only at the dyadic level and failed to explain a large proportion of the variation in lesion scores. As interactions do not occur in isolation but rather as a part of a larger interconnected dynamic, this thesis aims to apply social network analysis to post-mixing aggression in order to examine this behaviour within the wider social context. Social network analysis may reveal group level and indirect behaviours that play an important role in post-mixing aggression that may otherwise be undetectable. In Chapter 2 the relationship between pen level network properties and pen level injury rates were established. This revealed that networks containing large fully connected subgroups (cliques) tend to have fewer injuries 3wk-PM, whereas highly divided networks (betweenness centralisation) have considerably higher lesions. Chapter 3 follows on from Chapter 2 by examining the effect on lesions for individual pigs when occupying different positions within a social network. The first part of the chapter focuses on individual position within the network structures identified in Chapter 2. This chapter also quantifies a variety of commonly studied individual network positions to examine how these relate to individual lesions, and compares the model fit to dyadic traits. This chapter presents evidence that occupying centralised network positions can be beneficial for the central pig, but at the expense of pen-mates who are at risk of elevated rates of aggression and injury. In contrast, in pens with large cliques, no significant difference was found in injury rates between clique members (those who are part of a fully connected sub-group) and non members. This chapter concludes that while direct engagement in aggression at 24hr PM remains a strong predictor of injury at 3wk-PM, individual injury rates can be indirectly affected by the behaviours of pen mates that occurred at 24hr-PM. Chapter 4 builds upon Chapter 2 by quantifying different dominance metrics to establish whether pen level network properties relate to variations in hierarchical structure. This chapter revealed that pens with large cliques tend to have well-defined hierarchies (quantified by linearity and steepness), whereas the hierarchical structure in highly centralised pens tend to be poorly defined. This indicates that certain network structures are associated with poor hierarchy formation, which may partially explain the difficulties certain groups have with achieving and/or maintaining long-term social stability. Together the results indicate that SNA can complement conventional analyses of dyadic interactions to predict and understand the outcomes of aggressive interactions.