Abstract
Recent research has made progress by estimating both the quantity and fidelity of working memory representations. For example, the precision of working memory representations decreases as more items are remembered, and the exact shape of this function has been used to distinguish between slot-models (Zhang & Luck 2008) and resource-models (Bays & Husain, 2008) of memory. Thus, it is important to validate the methods used to estimate the quality and quantity of memory representations. Here we compare estimates derived from mixture models of performance in two different behavioral paradigms: continuous report and change-detection. In the continuous report task, observers saw a set of colors, followed by a brief retention interval, and a cue to report the color of a randomly selected item. Observers could choose any color on the color wheel, and the quantity and quality of memory representations was estimated from the distribution of errors (Zhang & Luck, 2008). In the change detection task, observers saw a set of colors, and after a brief retention interval, a second display appeared in which all items were identical, or one item changed color by some amount on the color wheel (e.g., 5, 10, …180 degrees). By modifying a signal detection model (Wilken & Ma, 2004) to include a guessing parameter, we again estimated the quantity and quality of memory representations. The estimates from the two tasks were in high agreement for set size 1 and 3, but diverged at higher set sizes where the change detection task yielded higher estimates of the number of items remembered, and lower estimates of memory precision. These findings raise the possibility that the continuous report task underestimates the number of items stored and the decrease in memory precision as set size increases, which would have important implications for theories of working memory capacity.
Meeting abstract presented at VSS 2012