Heterogeneity of rules in Bayesian reasoning: A toolbox analysis

Area of Science:

• Cognitive Psychology
• Decision Making
• Bayesian Inference

Background:

• Understanding how individuals compute Bayesian posterior probabilities is crucial for fields like medicine and law.
• Existing research often fits aggregate data to single-process models, potentially masking underlying cognitive heterogeneity.

Purpose of the Study:

• To investigate whether people use a single cognitive process or a toolbox of strategies for Bayesian inference.
• To compare the predictive power of single-process theories against a toolbox approach using extensive data.

Main Methods:

• Analysis of response distributions from laypeople and professionals across 106 Bayesian tasks.
• Simulations to evaluate the fit and predictive accuracy of the weighing-and-adding model.
• Testing a 'Five-Plus toolbox' of rules against over 10,000 individual inferences.
• Validation through experiments measuring response times, self-reports, and strategy use.

Main Results:

• Single-process theories showed limited support when analyzing response distributions.
• The weighing-and-adding model best fit aggregate data and predicted out-of-sample results, yet failed to predict individual inferences.
• A toolbox of five non-Bayesian rules plus Bayes's rule explained 64% of individual inferences.
• Experimental validation supported the toolbox model.

Conclusions:

• Fitting single-process theories to aggregate data can misrepresent the actual cognitive processes involved in Bayesian inference.
• Cognitive process heterogeneity is evident, suggesting individuals employ a flexible toolbox of reasoning strategies.
• Future research should focus on analyzing individual differences and strategy use to accurately model cognitive processes.