Bayesian confidence in optimal decisions

Area of Science:

• Cognitive Psychology
• Computational Neuroscience

Background:

• The drift diffusion model (DDM) accurately models decision-making and response times.
• Current DDM-based confidence models have limitations, prompting exploration of extensions.
• Alternative decision models are often used for confidence, despite DDM's success.

Purpose of the Study:

• To investigate simple extensions of the DDM to better account for decision confidence.
• To determine if a single evidence accumulation process can inform both decisions and confidence.
• To test if the DDM framework can be adapted to explain confidence reports.

Main Methods:

• Developed and compared several DDM variants incorporating confidence.
• Ensured decisions and confidence relied on the same evidence accumulation process.
• Validated models against benchmark data and a new preregistered study.

Main Results:

• A subset of DDM variants successfully accounted for quantitative aspects of confidence data.
• Confidence appears to reflect evidence strength penalized by decision time (Bayesian readout).
• The time penalty in confidence reporting may not be perfectly calibrated.

Conclusions:

• The DDM framework can be extended to provide a robust account of decision confidence.
• There is no need to abandon DDM or single-accumulator models for confidence research.
• Findings support a Bayesian readout of confidence, incorporating decision time.