Measuring affect-related attention bias to emotionally valenced visual stimuli in horses

Negative affect appears to mediate animals’ attention to competing emotional stimuli (e.g., threatening vs. non-threatening conspecific face images), similarly to anxiety-related enhanced attention to social threat reported in humans. To investigate this ‘attention bias’ (AB, i.e., the differential attention allocation to certain types of information over others) in horses, we developed a visual AB test assessing horses’ attention towards image pairs showing unfamiliar conspecifics’ facial expressions indicating, a) negative (social threat), b) more neutral (at rest), and c) positive (food anticipation) situations. We predicted that horses exhibit greater attention to negative compared to neutral or positive face images (as a normal adaptive response), and that horses in negative affective states (inferred from validated welfare indices comprising direct (health, behaviour) and indirect (housing, management) measures summarised as individual welfare scores and subscores reflecting health, social and environmental aspects) show greater AB to negative face stimuli and all images overall. Comparing AB to positive versus neutral social stimuli is rarely considered in AB studies, we therefore explored horses’ AB responses without a priori predictions. Over six trials, 44 horses from three facilities were shown stimulus pairs (negative/neutral, negative/positive, positive/neutral) presented simultaneously on two projector screens. Attention was assessed as absolute attention duration to each image, the proportion of time the negative/positive stimulus was attended to relative to the other stimulus, and overall attention (i.e., duration of head turns towards both stimuli combined). AB to stimulus type, side, effects of facility and individual characteristics (welfare and subscores, age) was analysed using linear and generalised mixed-effect models. Against our predictions, horses attended to the images within the three stimulus pairs for similar lengths of time (negative-neutral: W=1870.5, p=0.2572; negative-positive: W=2542.5, p=0.9296; positive-neutral: W=1762.5, p=0.1019). Due to Covid-19 interruptions, our sample size was lower than our estimated required number (N=113). Still, lower welfare (X^2₁=4.71, p=0.03) and health scores (X^2₁=4.13, p=0.04) significantly predicted shorter attention to the negative face stimuli, possibly reflecting threat avoidance previously reported in other animals. We found significant facility effects on overall attention to the stimuli (X^2₂=77.42, p<0.001), likely due to varying yard-specific conditions (e.g., lighting, noise). This highlights that external influences on visual attention require consideration when conducting cognitive tests at different testing sites. Further methodological investigation (e.g., test cue suitability, perceptual processing of computer-generated images; test stimuli familiarity; individual differences) is needed to evaluate the potential of AB as an indicator of affective valence in horses.