Influences of indication response requirement and target prevalence on dogs’ performance in a scent-detection task

Scent-detection dogs assist humans with many socially significant tasks and hold promise for assisting with many others. However, the methods used to train scent-detection dogs and the conditions under which they work are highly variable, and the influences of many relevant factors on scent-detection performance are poorly understood. Using an automated scent-detection apparatus that allowed the dogs to work independently from a handler, we evaluated the influence of two factors on scent-detection performance with amyl acetate as the target. In the first experiment in this study, we examined the influence of the indication response requirement on the performance of five dogs trained to perform a scent-detection task. The indication response consisted of the dogs breaking an infrared beam in a port through which they accessed samples. The response requirement was manipulated by adjusting the duration of the beam break that was required to activate food reinforcement if the target was present. As the indication response requirement increased, dogs’ ability to detect the target remained unchanged, as indicated by stable log d measures across durations (p = .09), but their response bias, represented by log B, shifted from a tendency to indicate that targets were present to a tendency to indicate that targets were absent (p < 0.001). In the second experiment, we examined the influence of target prevalence, or the proportion of samples that are target samples, on four dogs’ scent-detection performance. As target prevalence decreased, the dogs’ ability to detect the target remained unchanged (p = 0.13), but their response bias shifted from a tendency to indicate that the targets were present to a tendency to indicate that targets were absent (p < 0.001). This finding aligns with the “low-prevalence effect,” which is commonly observed in human signal detection research. The findings from both experiments have important theoretical and practical implications. For example, by adjusting response effort, we can alter dogs’ response bias toward or away from indicating the presence of targets and, by having a clear understanding of the influence of target prevalence on bias, we can make informed decisions about the need to artificially increase target prevalence when dogs are searching for rare targets.