Abstract
At VSS14, we reported an experiment where observers performed change detection with natural scenes. On 25% of trials, observers were asked to mark 12 locations where they thought they had searched in the last three seconds. Given the close relationship between fixation and attention, one might assume that they would know where they looked. However, observers subsequently viewed 10 more scenes and marked 12 locations where they thought a hypothetical observer would search in three seconds. Observers were no better at locating their own fixations than when they guessed someone else's, though in both cases, they were well above chance. Are observers bad at knowing their own fixations or good at guessing others' fixations? To test this, we replicated the experiment using artificial "Where's Waldo" scenes, since these make guessing others' eye-movements much harder. Participants previewed the Waldo display for 3 seconds. They were then given a non-Waldo search target, and searched the scene for another 3 seconds before making a target present/absent response. On 25% of trials, after the preview, there was no search. Observers were queried on where they had looked during preview. They were asked to place 12 clicks marking their fixations (3 seconds * 4 fixations/second). At the end of the experiment observers viewed 10 new Waldo displays and marked 12 locations where they thought someone else would have fixated. Here it seems much harder to guess where someone else will look and, indeed, performance was worse than in the previous experiment. Still, we found memory for one's own fixations (M=24%) was no more accurate than guesses about the fixations of a hypothetical observer (M=27%, p=.07) and performance was much worse than an ideal observer model (M=67%, p < .001). This replicates the previous experiment, suggesting that we have poor memory for where we just looked.
Meeting abstract presented at VSS 2016