Attention bias assessment

Attention bias was assessed as described by Verbeek et al.³⁵, using a similar method to that validated by Lee et al.²⁴. The test was conducted in a 4 × 4 m arena surrounded by a 1.5 m high fence covered in opaque black cloth (Fig. 5). The concrete floor was divided into 12 rectangular zones (1 × 1.2 m) marked on the ground with white paint. A familiar bowl containing 200 g of concentrate pellets was placed in the centre of the arena. A small window (77 × 58 cm) was located on one wall behind which a dog was sitting quietly (different from the dog used in judgement bias training). Each sheep was tested individually for a total of 3 min. The dog was visible to the sheep for the first 30 s of testing, then the window was covered by a retractable opaque cover and the dog was removed for the remainder of the test. All tests were continuously recorded by a video camera.

The following behaviours were recorded in real time or from video footage; time spent looking towards the dog or covered window with binocular vision (attention to dog²⁴), duration with the head at or above shoulder height (vigilance^24,82), latency to eat the food, number of zones crossed with both front legs and duration spent immobile with no movements of the head or body for greater than 3 s. Latency to eat was recorded from the beginning of the test. All other behaviours were recorded separately for the durations with and without the dog present. Only data without the dog present are presented, as this is the period where responses are to a perceived threat, as opposed to an actual threat posed by the presence of a predator, and thus better reflect anxiety states assessed using attention bias^24,35. Between animals, all uneaten food was removed and the bowl refilled with new pellets so that the amount of food eaten by each sheep could be calculated. The arena was also cleaned after eliminations to avoid the influence of odour cues on further subjects. Vocalisations, eliminations and foot stamps were also recorded, however these data were not further analysed due a low number of occurrences during the test.