Abstract
At VSS-2022, we reported that following a brief exposure of an array of 30 dots composed of 6 dots of each of 5 colors, randomly interleaved, all our subjects could concurrently compute at least five centroids, one for each color. This conclusion was based on the accuracy of centroid judgments in a post-cue, partial report procedure. Here, we propose a computational model of performance to compare ideal components versus our subjects' observed performance to estimate the extent to which each of the component processes limits performance. We find that that channel loss (the perceptual difficulty in associating so many differently colored dots with their locations) and memory loss (the ability to remember so many color-labeled locations until the required report) contribute approximately equally to response limitations. (The exact ratio of error contributions depends on the whether error is measured by mean-square distances of responses from centroids or by the number of stimulus items an ideal detector needs to process in order to match subjects' performances.) Conversely, the ability to extract and group dots according to their color (attention filters), centroid computation errors, and motor error (the ability to accurately output a recalled dot location) are minor factors in limiting performance. Only when the required number of concurrent centroid computations is greater than 5 (for two subjects) or greater than 7 (for one subject), is the limited number of available salience maps a possible factor limiting performance.