Figure 1.

Experimental design and behavioral results. A, In discrimination trials, participants made discrimination judgments about clockwise (cw) and anticlockwise (acw) tilted noisy gratings, and then rated their subjective confidence by controlling the size of a colored circle. In detection judgments, decisions were made about the presence (Y) or absence (N) of a grating in noise. B, Mean confidence as a function of response for the 35 participants. Confidence in detection “yes” responses was significantly higher than in “no” responses. No significant difference was observed between confidence in discrimination responses. C, Response accuracy was not different between the two tasks. D, Decoding accuracy for a classifier trained to classify response (yes or no in detection, clockwise or anticlockwise in discrimination) based on confidence ratings alone. Decoding accuracy was significantly higher for detection than for discrimination. ***p < 0.001. Box edges and central lines represent the 25, 50, and 75 quantiles. Whiskers cover data points within four inter-quartile ranges around the median.