Keynotes | K&Ts | GACs | Talks | Posters | Search

Poster F in Poster Session F: Thursday, August 6, 1:45 – 3:30 pm, Kimmel Center, Shorin & Rosenthal Rooms

Adaptive distortions of confidence reflect environmental statistics and internal constraints

Shiyi Cao¹, Aurelio Cortese²; ¹Nara Institute of Science and Technology, ²Sung Kyun Kwan University

Presenter: Shiyi Cao

Confidence is thought to reflect the strength of internal evidence, a view largely resting on models that assume static, symmetric environments. Yet evidence reliability can shape confidence independently of evidence strength (Boldt et al., 2017), and natural input statistics account for several confidence biases (Webb et al., 2023). When category prevalence is unequal, a naïve equal-variance reading of the Bayesian-optimal criterion predicts higher second-order sensitivity for the more frequent category; however, prevalence is known to reshape decision boundaries non-trivially (Levari et al., 2018), suggesting the opposite may hold. Using SDT simulations with first order sensitivity(d’) fixed by design, we show that the Beyas-optimal criterion shifts opposite to the behaviourally observed prevalence-induced shift(Levari et al., 2018);only when the criterion is displaced in the non-optimal, Levari-consistent direction does the model reproduce that pattern-yielding higher Type-2 sensitivity for the rarer category at matched accuracy, counter to the naive prediction. Unequal evidence variance adds further asymmetry, and the two interact nonlinearly. Feedforward networks trained on skewed frequencies reproduce this while keeping accuracy stable, suggesting metacognitive sensitivity is jointly shaped by environmental statistics and decision policy.

Topic Area: Decision-Making, Cognitive Control & Event Cognition