Trial-by-trial fluctuations in decision criterion shape confidence.

Many decisions are accompanied by confidence. One influential view couched within signal detection theory views confidence as the distance between a decision variable and a static decision criterion. However, recent work suggests that the criterion undergoes trial-by-trial fluctuations. Here, combining theory and model simulations, we predict that fluctuations in the decision criterion shape confidence. In 15 datasets, trial-by-trial estimates of the decision criterion were obtained with the hierarchical model for fluctuations in criterion. Across datasets, we confirmed our pre-registered hypothesis that confidence is shaped by a single-trial criterion state. This effect was found in 14 out of 15 datasets, indicating a robust pattern across a variety of paradigms and confidence scales. Beyond self-reports, the effect was also observed in reaction times, pupil-linked arousal and a neural signature of confidence. Our results show that variability in confidence, often treated as noise, actually reflects genuine sensitivity to the current state of the (fluctuating) decision criterion.