Abstract
Humans can rapidly and accurately compute summary statistical information about features of the environment. However, little is known about how ensemble representations are affected when task-irrelevant and distracting feature information is also present. Here we tested how effective feature-based attention –when tuned to a specific color– is in filtering out irrelevant orientation information in two intermixed ensembles of colored lines. Participants (N=40) viewed a stimulus display consisting of eighteen target and eighteen nontarget lines randomly positioned within an 8x8 grid. Targets and nontargets were colored red and blue respectively, counterbalanced across participants. The orientations of target and nontarget lines were randomly generated from a uniform distribution with a range of +/-30°, and either matched in average orientation or were offset by +/-25°, 50°, or 75°, allowing us to assess the role of orientation similarity during color-based attentional selection. The display appeared for 250ms, after which participants indicated the average orientation of the target ensemble by adjusting a single probe line using the left and right arrow keys. Participants performed overall well in the task (12.9° average error, SD=16.0°). When target and distractor ensembles matched in average orientation, participants’ average errors did not significantly differ from 0°. At +/-25° and +/-50° offsets, orientation reports were systematically biased in the direction of the distractor ensemble by 2.8° on average (average p-value = 0.00018), with 22% of participants’ average error in those conditions (13.0°) accounted for by this bias. At +/75° offset, however, errors did not significantly differ from the matched orientation condition. These data suggest that feature-based attention is an imperfect filter for segmenting ensembles that overlap or are similar to one another on a separate task-relevant feature. When ensembles are sufficiently distinct in the relevant feature dimension, feature-based filtering is either more effective, or potentially redundant with more basic grouping mechanisms.