Abstract
There are two predominant models of visual working memory (VWM) organization: one stipulates that VWM represents information into discrete slots and the other states that there is a single continuous pool of resources. However, one key aspect of the models has gone unnoticed, the speed of processing at the moment in-memory items should be assessed for their accuracy. Specifically, the slot model attributes incorrect memory performance for over-capacity memory setsizes to storage failure of test-relevant items, rather than failure of precisely storing items per se. According to this model, over-capacity memory assessment will be little hesitant regardless of setsize. Contrarily, the resource model attributes incorrect performance to setsize-dependent decline in the precision of in-memory items, although compensated by storing as many items as possible. Thus, the resource model expects that memory assessment eventually to be suffered as a result of an extensive coverage of imprecise in-memory items as setsize increases, consequently predicting proportional RT delays. To test this, we measured RTs in a change detection task across memory setsizes of 2, 4, 6, or 8 items. We model-fitted RT distribution from each setsize with an ex-Gaussian function under a maximum-likelihood method. We estimated three parameters from the Gaussian components, μ and σ (representing conventional central-tendency estimates) and from the exponential component, τ (representing a skewedness estimate), together providing a powerful comparison of the distributions across the setsizes. The analyses showed that the μ and σ were comparable across setsizes 4, 6, 8, except the lower μ and σ in setsize 2. The τ was constant regardless of setsize, indicating the distributions across over-capacity setsize conditions were virtually identical. The results suggest that about 4 items with good precision determined the changes detection performance, and support the slot model's idea that VWM represents information as a set of discrete high-resolution items.
Meeting abstract presented at VSS 2014