Abstract
Understanding the factors that mediate visual working memory (VWM) performance is of great interest. For instance, recall for dual-feature stimuli tends to be less accurate than for single feature stimuli. Although the cost of an additional feature is not large enough to support a pure feature-load account of WM storage limits, this empirical pattern challenges models that posit no cost for doubling the feature load in a VWM task. We examined this question while taking advantage of a new analytic technique that distinguishes between the maximum number of items that can be stored in VWM (max storage parameter: Kmax) and the probability that one will achieve that maximum (attention parameter: a). We employed a whole-report task in which subjects attempted to recall the features of all six stimuli in each trial. Subjects stored either color alone, orientation alone, or both the color and orientation of each stimulus. Replicating past work, subjects made more errors in the dual-feature condition compared to single-feature conditions. However, the total number of features stored was approximately 40% higher in the conjunction condition, suggesting an “object-based” benefit for storage. Moreover, the maximum number of items for which subjects stored at least one feature was equivalent in the single feature and conjunction conditions, in line with the hypothesis that VWM storage is limited by the total number of items rather than by total feature load. Finally, we examined whether storage of multiple features within a single object was associated, such that storage of one feature predicted successful storage of the other. Given one feature was reported accurately, the other feature was correctly reported approximately two-thirds of the time. Increased numbers of features stored from dual-feature objects, and covariations in the storage of each feature within an object, point towards a central role of object-based representations in VWM.