Abstract
Humans can adjust decision criteria to incorporate the probability of an event and the potential rewards or cost of the choice. Often, this adjustment is smaller than optimal, which is called conservatism. Self-assessment of the quality of these decisions is called confidence. While the probability of an event should affect criterion placement for both the decision and confidence assessment, the reward structure should only impact the former. We tested whether humans are optimal or conservative in perceptual judgments and whether they incorporate prior probabilities, potential rewards, or neither in their confidence reports. On each trial, participants performed an orientation-discrimination task (left vs. right), followed by a confidence report in their orientation judgment (high vs low). The probabilities of the two orientations and the rewards for a correct answer for each orientation varied independently across sessions. Participants received explicit instructions on the values of the priors and rewards and performed a practice block before testing. We used Signal Detection Theory to model the discrimination and confidence criteria. We modeled confidence as a region bounded by symmetric criteria centered on a hypothetical discrimination criterion that accounts for the prior and reward, the prior only, or neither. The hypothetical criterion could share conservatism with the discrimination criterion or have no conservatism. As in previous studies, participants were conservative in the orientation-discrimination task. Preliminary results show that participants correctly ignored potential reward when making the confidence judgment. Importantly, confidence judgments were not affected by conservatism. That is, the hypothetical criterion for the orientation-discrimination task is not shifted by conservatism. Thus, conservatism is introduced to decision making during perceptual decisions but not during evaluations of those decisions. This has implications for understanding the basis of different components and processes in human decision making.
Meeting abstract presented at VSS 2018