Animal Welfare Assessment of Fattening Pigs: A Case Study on Sample Validity

SIMPLE SUMMARY: The welfare of farm animals is discussed in society and politics. In Germany, the Association for Technology and Structures in Agriculture developed a new guideline for the animal welfare assessment of fattening pigs. It is called ANIMAL WELFARE INDICATORS: PRACTICAL GUIDE – PIGS and contains 13 characteristics by which the welfare of an animal is assessed, so-called indicators, as well as instructions on how to collect those indicators. For reasons of feasibility, six of the indicators should not be collected for all fattening pigs in a herd, but for a sample. The question arises whether then the herd’s level of animal welfare is assessed with sufficient precision. For this reason, this study examines five strategies for collecting samples of the fattening pigs in a herd. The aim is to identify a feasible strategy that collects samples of high validity. However, the study shows that the result of the animal welfare assessment based upon samples can partly deviate considerably from the result of the assessment of the entire herd. Further studies are needed to identify the most feasible and valid method for collecting samples of pigs from a herd. ABSTRACT: A guide for animal welfare assessment of fattening pigs recommends recording some of the indicators for a sample of the animals from a herd. However, it is not certain whether the herd’s level of welfare can be correctly judged using a random sample. Therefore, both the true prevalences of welfare indicators in a full census and the estimated prevalences of the indicators based upon simulated samples taken according to five strategies (termed S1 to S5) were determined. Deviations from the true level of animal welfare in the herd due to the sampling were recorded and analyzed. Depending on the strategy, between 12% and 43% of the samples over- or underestimated the true prevalences by more than 50%. The validity of the sampling strategies was evaluated using the normalized root-mean-squared error (NRMSE) and the relative bias (RB). In terms of accuracy, the strategies differed only slightly (between NRMSE = 0.13 for S2 and NRMSE = 0.19 for S4). However, the strategies varied more obviously regarding the bias (between RB = −0.0002 for S1 and RB = −0.0370 for S5). The described results are the outcome of an initial case study on the sample validity of the indicators and have to be verified using the data of more herds.