Howarth, E. R. I., Kemp, C., Thatcher, H. R. et al. 2021. Developing and validating attention bias tools for assessing trait and state affect in animals: A worked example with Macaca mulatta. Applied Animal Behaviour Science 234, 105198.
Attention bias is a new approach to assessing animal affect that has shown promising results in several animal species. It describes a tendency to preferentially attend to emotional compared to neutral cues and is influenced by underlying affect. It is important in the early days of this new field that we develop widely utilisable methods and incorporate lessons from the human literature from which tasks are adapted. This fundamental knowledge is critical to the development of standardised and sensitive tools, and the validation of experimental protocols to ensure best practice. Here, we describe protocols for two preferential-looking attention bias tasks. Study 1 involved a manual task using freely available low-cost materials. Study 2 used an automated task requiring specialist equipment and programming, but presumably less prone to noisy data. Tasks were tested with 109 socially housed rhesus macaques, Macaca mulatta, who had been trained to sit by a target, but received no other training. Tasks involved showing animals emotional face pairs (threat-neutral), and subsequent blind coding of video for duration of looking at either face. Three measures of social attention were examined: time spent looking at the threat face (THR), total time looking at the threat-neutral face pair overall (TL), and attention bias difference score (ABD) calculated as time spent looking at the neutral face subtracted from time spent looking at the threat face. Based on the human literature and early primate work, the influence of five potential confounding factors on attention was assessed: trial number, stimulus ID, previous testing experience, time of day and visual field to which the threat face was presented; as were several life history factors: sex, age, and social rank. Both tasks revealed stable individual differences in baseline social attention (THR and TL: effect sizes = 0.15−0.31; repeatabilities = 0.12−0.26; suggesting sensitivity to trait affect), but not ABD (which may be more sensitive to brief shifts in emotion state). All potential confounding factors had a significant effect on at least one measure of social attention. For a subset of monkeys who took part in both Study 1 and Study 2 several years apart (n = 18), there was significant reproducibility between tasks for all three measures (R = 0.15−0.63), supporting an argument for stable individual differences in baseline attention bias, and validating the two tasks for measuring the same trait. The attention bias method shows promise for further development of standardised protocols with animals. We provide framework and recommendations for future method development.