How disconfirmatory evidence shapes confidence in decision-making.

When assessing our decisions, the normative strategy involves giving equal weight to each evidence sample when computing confidence. However, recent findings suggest that the brain tends to overweight decision-congruent information when forming confidence judgements (i.e., positive-evidence bias; PEB). Here, we re-analyzed nine datasets (total N = 176) from human participants who judged the average color of eight shapes and gave their confidence. This task precisely allowed us to disentangle the impact of choice-confirming and choice-conflicting evidence on the formation of confidence. Strikingly, participants overly relied on evidence that conflicts with their choice, contrary to the normative model and the PEB. To explain this response-incongruent evidence effect in the computation of confidence, we fitted an extended log-posterior-ratio for confidence model to our data and show that the same robust averaging principle that influences decisions also accounts for these confidence effects: incongruent evidence receives a stronger weight in the computation of confidence because it lies closer to the category boundary around which there is heightened sensitivity. In a preregistered experiment (N = 32), we then empirically demonstrate that an experimentally induced shift in the category boundary affects the computation of confidence in otherwise identical stimuli. We conclude that confidence depends on dis-confirmatory evidence due to downstream consequences from decision-making mechanisms.