What predominantly causes errors in visual working memory tasks? Items may sometimes fail to be represented, poor (and declining) representational precision may result in wrong judgements, and correspondence errors —comparisons between the wrong representation and an observed object— may lead to mistakes. If correspondence errors are a dominant cause, then increasing item discriminability (along task irrelevant dimensions) should improve performance. But typical laboratory tasks employ identical objects in all but the feature dimension tested, e.g. triangles of different orientations to test orientation. In Experiment 1, participants remembered the sizes of objects —triangles and circles— and each trial included one or two objects. Crucially, some two object trials comprised one of each shape, while others comprised identical shapes. At test, participants were probed with a single object changed in size by a variable amount, and they reported whether the change involved an increase or a decrease. Performance declined significantly for two object compared to one object trials when the two objects were from the same shape category. When they were different, however, there was no cost to performance. By fitting cumulative Gaussian functions to participant responses we could estimate representational precision, and we found no change in the precision of a size representation for a single item versus each of two differently shaped items. A second experiment found similar results in orientation memory for triangles and lamp-like objects (though with some costs associated with different item trials). Further experiments explored working memory for color. These results conflict with a central prediction of flexible-resource theories, that the biggest relative declines in representational precision should occur at a shift from remembering one item to two. We suggest that these results should catalyze further investigations into the possibility that item correspondence errors are the predominant cause of error in working memory tasks.

Meeting abstract presented at VSS 2012