Decision Confidence As A Mapping Of Bayesian Posterior Belief

We study the confidence response distributions for several two alternative forced choice tasks with different structure, and assess whether their behavioral responses are accurately accounted for as a mapping from bayesian inferred probability of having made a correct choice. We propose an extension to an existing bayesian decision making model that allows us to quantitatively compare the relative quality of different function mappings from bayesian belief onto responded confidence. We find that a simple linear rescaling from bayesian belief best fits the observed response distributions. Furthermore, the parameter values allow us to study how task structure affects differently the decision policy and confidence mapping, highlighting a dissociable effect between confidence and perceptual performance.