Cognitive biases as Bayesian probability weighting in context

Humans often exhibit systematic biases in judgments under uncertainty, such as conservatism bias and base-rate neglect. This study investigates the context dependence of these biases within a Bayesian framework.

Forty-eight participants made subjective probability judgments in 12 scenarios requiring the integration of prior probabilities and evidence likelihoods.

Results show that task context mediates the weighting of priors and evidence: small-world scenarios (e.g., urn problems) emphasize priors, thus amplifying the conservatism bias, whereas large-world scenarios (e.g., taxi problems) increase sensitivity to evidence, leading to base-rate neglect. Presenting probabilistic information as relative frequencies rather than probabilities did not attenuate these biases.

To explain these findings, we propose the Adaptive Bayesian Cognition (ABC) model, which describes how individuals dynamically adjust the weighting of priors and evidence. By integrating normative Bayesian principles with psychological insights, the ABC model recasts cognitive biases as adaptive strategies shaped by capacity constraints and meta-learning in specific contexts. These findings bridge cognitive psychology, behavioral economics, and computational modeling to provide a unified framework for understanding subjective probability, probability weighting, and decision making under uncertainty. This work also informs the design of decision support systems.