Abstract
What attentional cues are most effective in complex, dynamic scenes, and how might cueing effects differ in real-world scenes and situations like driving? To test this, we had 100 online participants (all licensed drivers) detect dangerous events in videos of near-collisions across four cue conditions: no-cue (baseline), temporal (a bar at the bottom of the screen, localizing the hazard in time, but not in space), spatiotemporal-valid (a box indicating the time and location of a dangerous object in the scene) and spatiotemporal-invalid (a box indicating a distracting object simultaneous with hazard onset). On each trial, participants indicated whether the hazard was on the left or the right half of the display as quickly as possible following hazard onset. Relative to the no-cue baseline (542 ms, 87% correct), we found significantly faster RTs with spatiotemporal-valid (479 ms, p=0.03 vs. no-cue) and temporal (481 ms, p=0.04 vs no-cue) cues, and significantly slower responses with spatiotemporal-invalid cues (600 ms, p<0.0001, vs. no-cue). Hazard detection performance was not significantly different between no-cue (87%), temporal (85%) and spatiotemporal-valid (88%), but dropped significantly versus baseline in spatiotemporal-invalid (76%, p<.0001). In this real-world setting, attentional cueing shows similar effects to those observed in the lab, suggesting that they translate to time-critical settings in the world. Dangerous situations develop quickly on the road, and cueing drivers to them can buy them critical time needed to respond. Our results suggest that a simple cue that only temporally localizes a hazard may be more useful for drivers than a complex cue that localizes in time and space, because inaccurate spatiotemporal localization brings substantial RT and accuracy penalties (a +58 ms increase in RT, and a 12% reduction in accuracy). Our results extend attentional cueing into driving, and suggest how attention can be studied in real-world settings.