The Bayesian sampler: Generic Bayesian inference causes incoherence in human probability judgments.

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Decision Theory

Background:

• Human probability judgments exhibit systematic biases, contrasting with traditional Bayesian models of cognition.
• Existing models suggest explicit probability representation, which may not align with cognitive processes.
• Alternative approaches like sampling are used in statistical modeling for probabilistic calculations.

Purpose of the Study:

• To propose a novel cognitive model, the Bayesian sampler, for human probability estimation.
• To explain cognitive biases and heuristics within a unified computational framework.
• To reframe 'noise' in probability judgment models as a consequence of sampling with priors.

Main Methods:

• Development of the Bayesian sampler model, incorporating a generic prior with sample-based probability estimates.
• Reinterpretation of the probability theory plus noise model within the Bayesian sampler framework.
• Design and execution of two new experiments to test model predictions for conditional probabilities and judgment distributions.

Main Results:

• The Bayesian sampler model explains phenomena associated with conservatism and the conjunction fallacy.
• The model makes equivalent average predictions to the probability theory plus noise model for simple events, conjunctions, and disjunctions.
• Experimental results show the Bayesian sampler better captures mean probability judgments, with fit depending on assumed cognitive sample size.

Conclusions:

• The Bayesian sampler offers a unified framework for understanding human probability judgments and cognitive biases.
• The model provides a rational account of 'noise' in probability estimation, linking it to sampling processes.
• Empirical evidence supports the Bayesian sampler's ability to predict mean judgments, highlighting the role of sample size in cognitive processes.