Short communications: a methodological pilot comparing facial, endocrine, and observer-based welfare measures in...

Service dogs perform complex tasks in cognitively demanding environments, yet their welfare during training is often inferred from performance rather than measured directly. Several behavioral and physiological measures characterize how dogs respond to training, but researchers rarely apply them concurrently to the same animals. For this reason, we compared three such measures in seven purpose-bred Labrador Retrievers from a single litter during standardized early-stage training: facial behaviors associated with negative affective contexts (coded using DogFACS), urinary cortisol, and observer-based behavioral assessments by experienced trainers and non-expert observers. For each measure, dogs were ranked within the cohort according to that measure's directional convention, and all pairings were compared using Kendall's tau (τ). Rankings varied substantially across measures. Expert assessments agreed most strongly with urinary cortisol (τ = 0.67), non-expert assessments agreed with cortisol at a similar level (τ = 0.60), and expert and non-expert assessments agreed strongly with one another (τ = 0.65). By contrast, facial-behavior rankings showed no agreement with cortisol (τ = 0.00) and weak to very weak agreement with the observer measures (τ = 0.25 and 0.14). Because this was a small, closely related cohort assessed without a non-training control condition, these results cannot establish the validity of any measure. They instead show that measures can characterize the same dogs differently, highlighting the methodological challenges of welfare assessment in working dogs and the need for larger, adequately designed studies.