Abstract
The visual system can rapidly extract ensemble statistics from multiple objects: e.g., we might evaluate the average size or color of raspberries on a bush. To perform this task, observers must often select subsets of objects, rather than summarizing all visible information: we wish to evaluate raspberries while ignoring leaves. What is the representational basis of ensemble subset selection? What kind of visual information makes ‘good’ ensembles that can be attended and not confused with other ensembles to provide accurate summary statistics? Participants were shown spatially intermixed sets of objects and reported the average tilt (2AFC method, Experiments 1-2) or the average size (adjustment method, Experiment 3) of a target subset. We tested three subset definitions. In unique color subsets, one subset of items was the target ensemble (e.g., light red-dark red triangles). Distractors were light-dark triangles in green, blue, and yellow. Results were comparable to the baseline without distractors. If the target subset was defined by a conjunction of two colors partly shared with distractors (e.g., red-green targets among yellow-green, red-blue, and yellow-blue distractors), subset averaging was also possible but less accurate than in baseline and unique color conditions. Performance was very poor if the distractor subset includes specific items having both target colors but in different spatial combinations (e.g., red-green targets among green-red distractors as well as yellow-blue, and blue-yellow distractors). Experiment 2 provided evidence that participants did not base answers on a single selected target in unique color and conjunction conditions. Overall, these results suggest that observers can use color and, to a lesser degree, conjunction of colors but not their relative spatial positions to select subsets for the computation of ensemble statistics.