A rational model of sequential self-assessment

People's assessment of their ability varies in whether it is measured once following a task or sequentially via confidence judgments recorded throughout. Multiple models have been developed to predict one-off judgments of performance, which have often distinguished between peoples' biases about their general ability in a domain and their sensitivity to correctness. We propose a rational model of sequential self-assessment which allows us to make predictions about each individual separately-unlike in the one-off case which looks exclusively at the population level-and to identify, in addition to bias and sensitivity, the extent to which individuals' beliefs are responsive to their most recent evidence over the course of a task. We fit our model to data where participants solve algebraic equations and show that bias, sensitivity, and responsiveness vary meaningfully across participants.