Purchase this article with an account.
Daryl Fougnie, Anish Kanabar, Timothy Brady, George Alvarez; Asymmetric confidence intervals reveal hidden information in working memory. Journal of Vision 2016;16(12):34. doi: 10.1167/16.12.34.
Download citation file:
© 2017 Association for Research in Vision and Ophthalmology.
Studies have shown that working memory capability is limited, even for simple visual features. However, typical studies may underestimate the amount and richness of information in memory by relying on paradigms where participants only make a single recall response. To examine this possibility, we had participants memorize five briefly presented colored circles and then adjust a randomly probed item's color by selecting a color along a circular color space. Responses were noisy and many were far from the true color, suggesting poor memory. However, we found that this task significantly underestimates the information people remember about an item: even when participants responded inaccurately on an adjustment task, they performed reliably above chance when given the opportunity to make a second response (selecting which of two colors was present using a forced-choice task). Why do people seem to know more about items in memory then they convey in a single response? One possibility is that memory representations aren't well described by a single discrete color, but are rich and probabilistic. In a second experiment, we added a confidence judgement to the adjustment task. After selecting a color, participants drew arcs independently in the clockwise and counterclockwise direction to indicate their confidence in their response. Not only was reported uncertainty highly correlated with observed uncertainty (demonstrating metaknowledge), but the asymmetry in arcs predicted the direction of the correct answer (60.8 % of longer arcs were toward the correct answer). The information in the asymmetric arcs was also highly correlated (r=.80, p< .05) with participants' ability to choose the correct color on a subsequent forced-choice task, suggesting these measures reflect the same construct. We conclude that participants know significantly more than is reflected in a single response, perhaps because this knowledge is probabilistic and thus difficult to condense into a single report.
Meeting abstract presented at VSS 2016
This PDF is available to Subscribers Only