Amplification from saliency affects explicit but not implicit ensemble representations

The visual system can encode multiple objects in the form of ensemble representations. Such representations can be accessed with either explicit or implicit reports, but depending on the type of report, the observed properties of the ensemble representation can differ in detail. Previous studies have suggested that the saliency of individual items biases the perceived mean of ensembles. It is unclear, however, whether saliency affects implicit representations of the whole feature distribution (beyond mean and variance). Our observers were presented with sets of lines varying in orientation and size where size was a task-irrelevant salient feature. To estimate explicit representations, observers adjusted the mean orientation and their accuracy was measured. To access the implicit representation of the feature distribution we used a visual search task for an oddly-oriented line among heterogeneous distractors and measured the search times. The results revealed a strong saliency-induced bias in the explicit report, with mean orientation estimates biased toward the more salient items. However, no such effect was observed in the implicit report, as search times were unaffected by the saliency of the items. Our results support the hypothesis that distinct mechanisms underlie the more detailed implicit representations of full distributions and explicit ensemble representations that are based on summary statistics.