Measuring performance in multiple-target visual search

Abstract

Naturalistic visual search tasks, such as radiographic image screening, often contain more than one target and require the searcher to not only report the presence of anomalies, but to also indicate their specific locations. The requirement to localize multiple targets precludes the use of common signal detection performance measures that assume only a single response per display. Chakraborty (2008) has validated two measures of sensitivity, A’1 and A1, for use in free response localization tasks like radiology. These measures, however, require the searcher to report confidence ratings for each anomaly, a demand that is not always practical or naturalistic. The current work examines whether these measures would be effective in a case in which confidence ratings were not reported. Experiments 1 and 2 examined the effects of target discriminability and instructional bias on A’1, A1, and d’eq. Experiment 3 examined the effect of bias imposed by differences in target base rate. Across three experiments, neither A’1 nor A1 was entirely resistant to shifts in bias, but the effect of the bias manipulations on A’1 was small in Experiments 1 and 2 ( ω2 ≤ .05) and non-significant in Experiment 3. Although d’eq was not significantly affected by the bias manipulation in Experiments 1 & 2, it was strongly affected by target base rate in Experiment 3. The results suggest that, of the three measures, A’1 can serve as the most useful measure of sensitivity in multiple-target search tasks.