Abstract
Recently seen items bias our perception ("serial dependence"), but the direction and magnitude of these biases vary. What determines this dependence between previous input and current percept? Attention and similarity (proximity in feature space) are two possibilities that have been discussed as potential influences on such biases. In four experiments, we investigated how attended (target) and ignored (distractors) visual search items bias the perceived orientation of an irrelevant test object while manipulating test-to-target and test-to-distractor similarity. In the first three experiments, participants searched for the line with the most distinct orientation from all other items (“odd-one-out” search) in blocks of 4-5 trials. To ensure that observers would have a precise representation of distractors, their orientations were repeatedly drawn from the same probability distributions within blocks. A test line was presented briefly after completing visual searches. Participants had to report its orientation by rotating an adjustment bar. We manipulated distance in the orientation space between the test line, the last target, and the mean distractor orientation in each block. In Experiment 4, target locations were additionally pre-cued to further decrease attention to distractors. The results indicate that ignored and attended items produce opposite (repulsive and attractive) biases upon the perceived orientation of an irrelevant item, but this effect was moderated by proximity in feature space. Attended items (targets) produced strong attractive biases when they were similar to the test stimuli but no biases when they were dissimilar. In contrast, ignored items (distractors) created weaker attractive biases when they are similar to the test items, but repulsive biases when they were dissimilar. Overall, we speculate that our results provide a glimpse into the bag of tricks that the visual system uses to stabilize visual perception over time and that both attention and similarity affect biases in perception.