Dissecting Bayes: Using influence measures to test normative use of probability density information derived from a...

Area of Science:

• Cognitive Psychology
• Computational Neuroscience
• Decision Science

Background:

• Bayesian decision theory (BDT) models normative performance in decision-making tasks involving uncertainty and value.
• Normative models dictate optimal information encoding and combination to maximize expected reward.
• Standard BDT computations involve probabilities, but real-world tasks often use probability density functions (PDFs) from samples.

Purpose of the Study:

• To investigate human ability to perform individual computations within a BDT framework for visual cognitive tasks.
• To assess human adherence to normative principles of accuracy, additivity, and influence when using sample-derived PDFs.
• To compare human decision-making strategies against normative BDT predictions and explore alternative models.

Main Methods:

• Deconstructing Bayesian decision theory (BDT) into sequential computations for isolated testing.
• Evaluating human performance on tasks requiring the use of probability density functions (PDFs) derived from samples.
• Measuring influence to quantify the weighting of individual sample points in decision-making and comparing it to normative standards.

Main Results:

• Participants systematically violated normative accuracy and additivity principles in PDF-based decision tasks.
• While accuracy and additivity deviations had minor impacts, sample point influence weighting significantly differed from BDT predictions.
• Human decision-makers failed to utilize geometric symmetries of PDFs, unlike the normative BDT model.

Conclusions:

• Human decision-making in tasks with sample-derived PDFs deviates from normative Bayesian decision theory (BDT) predictions.
• The normative BDT model's assumption of utilizing geometric symmetries is not reflected in human behavior.
• An alternative model, prioritizing decisions based on a single extreme sample point, offers a more accurate account of observed human data.