Abstract
In standard visual search tasks, targets appear on 50% of trials. This contrasts with real-world searches in which targets can be very rare indeed, such as airport screening. Previous work examining target prevalence (defined as the proportion of trials that contain a target) has found that, as prevalence increases, participants become increasingly likely to detect targets, while increasing their reaction times on target-absent trials. Eye-movement analyses have found that this pattern is not only due to an increase in the proportion of objects fixated as prevalence increases, but also to an increase in the likelihood that distractors will be revisited. Since search has a limited-capacity memory record for already-inspected objects, in the present study, we asked whether the increased revisitation rate in higher-prevalence searches results from memory capacity limitations, or from a tendency to revisit objects as prevalence increases. To address this, we engaged participants in a search for simple targets (T-shapes amongst L-shaped distractors) while tracking their eye-movements. Prevalence was set to 10%, 50% or 90%. We replicated standard prevalence effects: as prevalence increased, participants were more likely to detect targets. Revisitation rates also increased with prevalence. Across all prevalence levels, we found that participants were more likely to revisit objects that were first visited early in a trial, compared with objects first visited later in a trial: a finding consistent with a limited-capacity memory record. In addition to this rise in revisits for objects visited early during trials, we also found a general increase in revisitation rates for all objects as prevalence levels increased, regardless of when those objects were first visited. Our results suggest that the increased revisitation rates in higher-prevalence searches are not only caused by the limited-capacity memory for already-inspected objects. Rather, there is also tendency to revisit all objects as prevalence increases.
Meeting abstract presented at VSS 2016