Abstract
The visual search literature consistently reports a "pop out" effect for targets defined by a single feature. For such targets, reaction time does not change as a function of display size. When targets are defined by a conjunction of features, however, reaction time increases as a function of display size (e.g. Treisman & Gelade, Cognitive Psychology,1980). Featural attention can also be studied using the centroid paradigm, in which the task is to estimate the center of mass of target items in a display (Sun, Chubb, Wright, & Sperling, Attention, Perception, and Psychophysics, 2015). Performance in the centroid task can be measured using the selectivity ratio (the relative weight of targets versus distractors in subjects' centroid estimations), and efficiency (the lower bound on the proportion of items processed). Method: In two sets of experiments, we compared performance in single-feature and conjunction conditions in both visual search (one target, 3 to16 distracters) and centroid tasks (4 targets, 4 or 12 distracters). The feature dimensions were size and color in Experiment 1 and luminance and shape in Experiment 2. Results: In both experiments, our visual search results replicated previous findings as expected. However, in the centroid task, we found improved performance on the conjunctive size-color conditions compared to the single-feature size conditions in Experiment 1, and improved performance on the conjunctive luminance-shape conditions compared to the single-feature shape conditions in Experiment 2. Conclusion: These centroid results are surprising given the visual search literature, which would seem to predict poorer performance on conjunction conditions compared to single-feature conditions. Rather, the centroid results suggest that it is possible to construct (and deploy over space) effective attention filters for conjunctive targets, and these attention filters for conjunctive targets can be even more effective than attention filters for a single constituent feature of the conjunction.
Meeting abstract presented at VSS 2016