Assessment of test-retest reliability of animal-based measures on growing pig farms

Measures taken on animals are subject to variations and their reliability may represent one of their main limitations when used in a welfare assessment tool. Test-retest reliability refers to the likelihood that the same results will be obtained if the assessment is repeated by the same observer. This study discusses the test-retest reliability of quantitative and qualitative animal-based measures included into the Welfare Quality® protocol on a sample of 15 intensive conventional farms of growing pigs. An average of 12 months elapsed between the two visits and no changes in management routines or housing conditions were made by the farmers during this interval. A Wilcoxon signed rank test was used to test whether the mean results obtained during the two visits were significantly different. Different methods were used to assess repeatability of mean prevalences at farm level: Spearman's correlation coefficient (rs), intra class correlation coefficient (ICC) and limits of agreement (LoA). Moderate bursitis, widespread skin discolouration and positive social behaviour presented mean results significantly different between the two visits (P<0.05). From a discriminative point of view, only moderately soiled body (rs=0.7; ICC=0.8) and the panic response (rs=0.7; ICC=0.5) were consistent the 12 month period. However, inconsistencies between Spearman's correlation and ICC were found when interpreting the reliability of several quantitative measures such as severely wounded animal and severely soiled body. The Qualitative Behaviour Assessment (QBA) was subject to a significant effect of visit, however, the two sets of scores were moderately related on the first axis of the principal component analysis (rs=0.5). A more evaluative approach such as the limits of agreement can be useful to guide the interpretation of test-retest reliability of quantitative measures based on farm averages and to detect farms that present a persistent welfare problem. Unless the assessment relates only to the relative ranking of a farm within the same visit, the QBA is not consistent over the 12 month period.