Abstract
It has been shown that the recalled size of an individual item is systematically biased towards the mean size of the set in visual working memory (VWM), suggesting hierarchical encoding (Brady & Alvarez, 2011). Here, we investigated whether hierarchical representations in VWM are spatially local or spatially global. We showed participants six circles of various sizes. The circles were spread across two quadrants aligned vertically or horizontally relative to eye-tracked fixation, with three circles per quadrant. Unbeknownst to participants, the mean sizes differed between the quadrants (the size gradient was masked by irrelevant filler circles). For each particular display, we tested memory for exactly the same circle size twice: when it was presented in a quadrant with a larger mean size vs. a quadrant with a smaller mean size. We calculated the bias towards the quadrant mean by computing the ratio of observer’s responses to the tested item in these two responses. Although participants were unaware of the mean size manipulations across quadrants, we found a strong bias toward the quadrant mean in both vertical and horizontal alignments. In Experiment 2, we investigated whether this effect is quadrant-based or caused by any neighboring items. We tested a circle near a between-quadrant meridian in half of the trials. On these trials, the mean sizes of the two same-quadrant circles were always opposite (relative to the target item) those of the two nearest circles from a different quadrant, such that taking into account the entire local region of the target would result in no bias in any direction. The results showed the same strong bias toward the own quadrant mean in both horizontal and vertical alignments. Overall, the experiments showed that hierarchical representations in VWM are substantially space-based with a specific role played by horizontal and vertical meridians of the visual field.