The effect of repeated testing on judgement bias in domestic dogs (Canis familiaris).

Judgement bias paradigms are increasingly being used as a measure of affective state in dogs. Approach to an ambiguous stimulus is commonly used as a measure of affect, however, this may also be influenced by learning. This study directly measured the impact of learning on a commonly used judgement bias paradigm in the absence of an affective state manipulation. Dogs (N = 15) were tested on a judgement bias task across five sessions. The dogs' latency to approach a bowl placed in one of three ambiguous locations between non-baited (negative) and baited (positive) locations was measured. Results show that session number had a significant effect on the dogs' latencies to reach the ambiguous bowl locations, with post-hoc tests revealing that dogs were significantly slower to approach the locations as the number of sessions increased. Session number also had a significant effect on the number of times the dogs did not approach the bowl within 30 s of being released, with the number of no approaches generally increasing across sessions. When dog identity was included as a fixed effect, a significant effect on latency to approach was found, suggesting that some dogs were consistently faster than others across sessions. To assess whether the paradigm produced repeatable results, Intraclass Correlation Coefficients were used. A low degree of reliability was found between latencies to approach each bowl position across sessions. This study demonstrates that dogs learned that the ambiguous locations were not rewarded with repeated exposures, and that this impacted their responses. We conclude that this judgement bias paradigm may require further consideration if applied across multiple exposures and that repeated results should be interpreted with caution as they are likely impacted by learning.