Kai Xue¹ (), Herrick Fung, Dobromir Rahnev; ¹Georgia Institute of Technology

Stimulus manipulations in perceptual decision-making tasks can be categorized into two broad categories: manipulating the quality of the sensory input (stimulus reliability) and manipulating the distance between the sensory feature of interest and the decision boundary (boundary distance). However, despite the fundamental difference between these two manipulations, it remains controversial whether they lead to dissociable behavioral effects. Here, we demonstrate robust dissociations between these manipulations across two large experiments (N = 78). In Experiment 1, subjects indicated the orientation of a Gabor patch relative to a decision boundary set at 45°, with stimulus reliability and boundary distance manipulated through contrast and the offset from 45°, respectively. Experiment 2 replicated this design using dot motion stimuli, where stimulus reliability and boundary distance were manipulated through motion coherence and the offset from 45°, respectively. Across the two experiments, we found that, the boundary distance manipulation, as compared to stimulus reliability, had a larger effect on accuracy but a smaller effect on confidence. This led to some of the strongest confidence-accuracy dissociations observed in perceptual tasks. Specifically, high-reliability/low-distance conditions produced much higher confidence but lower accuracy compared to low-reliability/high-distance conditions. We also found a qualitative dissociation in confidence patterns for correct and error trials. Specifically, boundary distance manipulations produced the "folded-X pattern" where easier conditions increased confidence for correct trials but decreased it for error trials. In contrast, stimulus reliability manipulations violated the folded-X pattern, such that where easier conditions increased confidence for both correct and error trials. These results demonstrate that manipulations of stimulus reliability and boundary distance have fundamentally different effects on the relationship between accuracy and confidence, suggesting that the mapping between confidence and accuracy depends critically on how task difficulty is manipulated.

Acknowledgements: This work was supported by the National Institute of Health (award: R01MH119189) and the Office of Naval Research (award: N00014-20-1-2622).