Confidence in the mind's eye.

Humans intuitively evaluate their decisions by forming different levels of confidence. It has been established that decisional confidence and accuracy are well-correlated, but can also deviate. Understanding of the computations underlying confidence, and how these differ from computations that limit sensitivity, is lacking. Here we show that, for visual judgments concerning global motion direction, confidence and accuracy are separable because the variance associated with a sensory signal undermines confidence to a greater extent than it does sensitivity. Stimuli consisted of dots moving in a uniform distribution of directions about a mean test value. The critical manipulation was the range of different directions contained in a test stimulus. Before testing, we calibrated different stimuli to equate sensitivity in fine direction discriminations (left / right of vertical), by manipulating the offset of the average direction from the vertical decisional boundary. This calibration failed to equate confidence - despite constant levels of sensitivity, people were less confident when judging more variable direction signals. When we instead calibrated stimuli to equate confidence, people were more sensitive when judging more variable signals. These complementary results reveal a differential weighting of signal variance in computations that limit visual sensitivity and decisional confidence. This means that the precision of perceptual decisions, and levels of confidence in those decisions, rely on independent transformations of sensory input. Meeting abstract presented at VSS 2015.