Doubly Bayesian Analysis of Confidence in Perceptual Decision-Making

Area of Science:

• Cognitive psychology
• Neuroscience
• Decision-making

Background:

• Metacognition allows humans to report on the reliability of their cognitive processes.
• Confidence judgments are a common measure of metacognition, but the underlying computations are not fully understood.

Purpose of the Study:

• To develop and apply a Bayesian method for comparing computational models of confidence judgments.
• To investigate whether confidence reports in a visual task are Bayes optimal or heuristic-based.

Main Methods:

• A visual two-interval forced-choice task was employed.
• Two experimental designs were used: a standard sequential design (decision then confidence) and a simultaneous design (decision and confidence together).
• A fully Bayesian approach was used to compare heuristic and Bayes optimal models of confidence.

Main Results:

• In the standard sequential design, participants' confidence reports were largely Bayes optimal.
• In the simultaneous design, participants were equally likely to use Bayes optimal strategies or heuristic, suboptimal strategies.
• Even minor alterations in task complexity or design can shift participants away from Bayes optimal confidence computations.

Conclusions:

• Human confidence reports can align with Bayes optimal computations.
• Metacognitive judgments are sensitive to experimental design, with non-standard designs potentially eliciting heuristic strategies.
• Understanding the computational basis of confidence requires careful consideration of task parameters.