Feedback reduces but does not eliminate confirmation bias

Poster Presentation: Saturday, May 18, 2024, 8:30 am – 12:30 pm, Pavilion
Session: Decision Making: Perceptual decision making 1

Marvin R. Maechler1, Tobias H. Donner2, Alan A. Stocker1; 1University of Pennsylvania, 2University Medical Center Hamburg-Eppendorf

Previous studies have demonstrated that a categorical choice can bias the subsequent integration and evaluation of sensory evidence: evidence that is congruent with the choice is given more weight in a subsequent perceptual judgment than incongruent evidence. However, participants in these studies did not receive feedback about the accuracy of their perceptual judgments and therefore produced these biases unknowingly. Thus it is unclear whether confirmation bias is the result of a suboptimal or uncalibrated inference strategy, and thus might be malleable by proper feedback. Or alternatively, whether confirmation bias is resistant to feedback and possibly the result of a normative strategy. To resolve this, we conducted a psychophysical experiment in which participants (N=8) had to estimate the unknown mean angular position from a stimulus sequence consisting of eight normal distributed samples of this mean (fixed variance; 11 different mean positions, angular positions on a circle equidistant [5dva] from fixation). Every trial required two responses from the participants. First, participants had to report whether the mean was clockwise or counterclockwise of a reference position after either seeing four or all eight samples (prospective and retrospective condition, respectively; randomly interleaved). Then they were asked to provide an estimate of the mean either directly after their categorical choice or after seeing the remaining samples of the sequence (prospective condition). Crucially, participants were shown the true mean position immediately after reporting their estimate, and also received monetary reward for estimates close to the true mean. We found that the incentivized feedback substantially reduced confirmation bias but without eliminating it. This shows that previously reported confirmation biases are, in part, caused by suboptimal decision strategies due to the missing feedback. Our results suggest that the remaining confirmation bias is the result of a deliberate decision process that considers other objectives in addition to estimation accuracy.

Acknowledgements: This work was supported by the NSF CRCNS grant IIS-1912232 to A.A.S., and Federal Ministry of Education and Research (BMBF) Project Nr. 01GQ1907 to T.H.D.