Adaptive cue combination in a visual estimation task
Abstract
Introduction. Uncertainty is a curse, and one antidote is the active collection of additional information. Information, though, can be costly. We asked subjects to attempt to hit a small, invisible target on a touch screen. They earned points for each hit. The only cues to the target's location were dots drawn from a Gaussian distribution centered on the target. The dots appeared one by one as the subject repeatedly pressed a key, but each dot cost a small amount subtracted from the subject's potential reward. Would subjects know when to stop "buying" information? Methods. Eight subjects participated in each of two conditions (200 trials per subject; conditions interleaved in short blocks). (1) Target value started at 40 points and decreased by two points per cue purchased. (2) Target value started at 60 points and decreased by six points per cue purchased. The optimal rule maximizing expected gain was to purchase seven and four cues, respectively.Results. (a) Subjects earned an average of 14.06 points per trial (SD = 2.19), which is 76% of the optimal expected gain of 18.55 points per trial. (b) All subjects correctly purchased more cues in the 40 point condition (M = 8.04, SD = 1.5) than the 60 point condition (M = 5.87, SD = 0.72), t(14) = 3.68, p < .003. (c) Statistical tests indicate that almost all subjects were risk averse in that they purchased more cues than dictated by the optimal rule. The exceptions were one subject who was risk seeking in the 40 point condition (M = 5.38, SD = 1.36), t(99) = 11.9, p < .001, and one other subject who was not significantly different from optimal in the 40 point condition (M = 7.03, SD = 1.55), t(99) = 0.19, p > .05.Conclusion. When balancing the costs and benefits of purchasing information to reduce visual uncertainty, subjects collected more information than predicted by the ideal observer.
Keywords
Bibtex entry:
@article{juni2011adaptive,
abstract = {Introduction. Uncertainty is a curse, and one antidote is the active collection of additional information. Information, though, can be costly. We asked subjects to attempt to hit a small, invisible target on a touch screen. They earned points for each hit. The only cues to the target's location were dots drawn from a Gaussian distribution centered on the target. The dots appeared one by one as the subject repeatedly pressed a key, but each dot cost a small amount subtracted from the subject's potential reward. Would subjects know when to stop "buying" information? Methods. Eight subjects participated in each of two conditions (200 trials per subject; conditions interleaved in short blocks). (1) Target value started at 40 points and decreased by two points per cue purchased. (2) Target value started at 60 points and decreased by six points per cue purchased. The optimal rule maximizing expected gain was to purchase seven and four cues, respectively.Results. (a) Subjects earned an average of 14.06 points per trial (SD = 2.19), which is 76% of the optimal expected gain of 18.55 points per trial. (b) All subjects correctly purchased more cues in the 40 point condition (M = 8.04, SD = 1.5) than the 60 point condition (M = 5.87, SD = 0.72), t(14) = 3.68, p < .003. (c) Statistical tests indicate that almost all subjects were risk averse in that they purchased more cues than dictated by the optimal rule. The exceptions were one subject who was risk seeking in the 40 point condition (M = 5.38, SD = 1.36), t(99) = 11.9, p < .001, and one other subject who was not significantly different from optimal in the 40 point condition (M = 7.03, SD = 1.55), t(99) = 0.19, p > .05.Conclusion. When balancing the costs and benefits of purchasing information to reduce visual uncertainty, subjects collected more information than predicted by the ideal observer.},
author = {Juni, Mordechai Z and Gureckis, T.M. and Maloney, Laurence T},
journal = {Journal of Vision (Proceedings of the Annual Vision Sciences Conference)},
number = {11},
pages = {1062--1062},
publisher = {The Association for Research in Vision and Ophthalmology},
title = {Adaptive cue combination in a visual estimation task},
volume = {11},
year = {2011}}QR Code:
Download SVG