Spatial transposition gradients in visual working memory

Abstract

In list memory, access to individual items reflects limits of temporal distinctiveness. This is reflected in the finding that neighbouring list items tend to be confused most often. This article investigates the analogous effect of spatial proximity in a visual working-memory task. Items were presented in different locations varying in spatial distance. A retro-cue indicated the location of the item relevant for the subsequent memory test. In two recognition experiments, probes matching spatially close neighbours of the relevant item led to more false alarms than probes matching distant neighbours or non-neighbouring memory items. In two probed-recall experiments, one with simultaneous, the other with sequential memory item presentation, items closer to the cued location were more frequently chosen for recall than more distant items. These results reflect a spatial transposition gradient analogous to the temporal transposition gradient in serial recall and challenge fixed-capacity models of visual working memory (WM).

Keywords:

• Transposition errors
• Spatial distinctiveness
• Cue-based retrieval

This research was supported by a grant from the Swiss National Science Foundation (Project 100014_126766/1). We thank Iliana Karipidis and Matthias Hartmann for collecting the data.

Notes

¹ Analyses including all participants showed the same data pattern as the analyses with subjects excluded.

² One reviewer asked why we controlled for chance by dividing the observed proportion of errors by the proportion expected by chance, rather than subtracting the latter from the former. We opted for a ratio rather than subtractive correction because the former is more conservative. For illustration, imagine a simplified scenario: Across all trials, a participant had 100 opportunities for committing a transposition error (i.e., there were 100 colours in the response panel that, if chosen, would constitute a transposition error), and 100 opportunities to commit an extralist intrusion. Imagine further that the person has a higher probability of committing a transposition error (p = .3) than committing an extralist intrusion error (p = .1), perhaps because list objects are more familiar than extralist items, regardless of their spatial proximity to the tested object. Let the transposition error opportunities be unevenly distributed across distances to the tested object's location: close (50), intermediate (30), and far (20). The person would be expected to commit the four kinds of errors with the following frequency: close (15), intermediate (9), far (6), extralist (10). By chance, the person would be expected to commit each kind of error with an equal probability, so that their expected frequencies are a constant proportion of the opportunities (i.e., p = .2): close (10), intermediate (6), far (4), and extralist (20). If we subtracted the frequencies expected by chance from the observed frequencies, we would obtain an artefactual spatial gradient: close (5), intermediate (3), far (2), extralist (–10). If instead we correct by calculating the ratio of observed to expected frequencies, we would obtain an elevated level of transpositions independent of distance: close (1.5), intermediate (1.5), far (1.5), extralist (0.5).

³ We used linear regression as a proxy to capture the hypothesized monotonic trend without committing to a particular function form, for which we see no theoretical basis.