The spontaneous qualitative assessment of behavioural expressions in pigs: first explorations of a novel methodology for integrative animal welfare measurement.

Qualitative assessments of behaviour integrate and summarize the different aspects of an animal's dynamic style of interaction with the environment, using descriptors such as 'timid' or 'confident'. Although such qualitative terms are widely used in the study of animal temperament and personality, their use in relation to questions of animal welfare has yet to be formally explored. The terms used in integrative assessment (e.g., content, distressed) tend to have expressive, welfare-related connotations, and lie at the heart of the lay public's concern for animal suffering. For this reason they are frequently dismissed as 'anthropomorphic' and unscientific. However, in theory it is possible that these terminologies reflect observable aspects of behavioural organization. They may therefore be liable to scientific analysis, and be of use as integrative welfare measurements. A first step in investigating this hypothesis is to examine the inter-observer reliability of assessments of behavioural expression. This study investigated the extent to which 18 naive observers showed agreement when given the opportunity to qualitatively describe, independently and in their own words, the behavioural expressions of 20 individual growing pigs. Pigs were brought singly into a test pen and given the opportunity to interact with a human squatting in the centre of the test pen. Observers were instructed to first observe each pig and then to write down terms which adequately summed up the emergent qualities of that pig's behaviour. Data thus consisted of 18 sets of individually generated descriptive terms, attributed to 20 pigs. This procedure was repeated a month later with the same observers but using a new group of 20 pigs. To analyze the resulting 36 sets of descriptive terms, pigs in each set were given a score for each term. This score was either 0 (term not used for that pig) or 1 (term used for that pig). These data were analyzed with Generalized Procrustes Analysis (GPA), a multivariate statistical technique which finds a consensus between observer assessment patterns (the 'pig consensus profile'), and provides a measure of observer agreement. Results show that for each group of 20 pigs, the 'pig consensus profile' differed significantly from an analysis of the same data in randomized form (p<0.001), indicating that the consensus profiles were not artifacts of the GPA procedures. It can therefore be concluded that observers showed significant agreement in their spontaneous assessment of pig expressions, which suggests that these assessments were based on commonly perceived and systematically applied criteria. The extent to which these shared criteria reflect observable aspects of behaviour now requires further study.