Computations Underlying Confidence in Visual Perception

Humans intuitively evaluate their decisions by forming different levels of confidence. Despite being highly correlated, decisional confidence and sensitivity can be differentiated. The computational processes underlying this remain unknown. Here we find that, for visual judgments concerning global direction, signal range has a greater impact on confidence than it does sensitivity. We equated sensitivity for stimuli containing different degrees of directional variability. This failed, however, to equate confidence-participants were less confident when judging more variable signals despite constant sensitivity. When stimuli were instead calibrated to equate confidence, participants were more sensitive when judging more variable signals. Directional range had no impact on an unrelated judgment of brightness, helping to establish that these results cannot be attributed to a simple decisional confound. Our complementary results show that directional sensitivity and decisional confidence rely on independent transformations of sensory input. We propose that confidence will generally be shaped by the range of differently tuned neural mechanisms responsive to input during evidence accumulation, with this having a lesser impact on sensitivity.