Computational characterization of metacognitive ability in subjective decision-making

Metacognition is the process of reflecting on and controlling ones own thoughts and behaviors. Metacognitive ability is often measured through modeling the relationship between confidence reports and choice behavior in tasks where performance can be objectively measured. Previous work has explored whether metacognitive ability is conserved across different types of tasks, and across different domains such as perception and memory. However, it is unclear whether approaches to the assessment of metacognitive ability that have worked in these contexts can be extended to value-based decision-making where objective accuracy cannot be evaluated. Here, we compare metacognitive ability across different tasks spanning perception and valuation, using Bayesian hierarchical estimation of the parameters of a computational process model of confidence. This model captures metacognitive ability as the uncertainty of an individuals own decision uncertainty, and can do so in the space of any decision variable, regardless of whether it indexes external features or internal, subjective states. We find that metacognitive ability can be reliably estimated in both objective and subjective decision-making tasks, and is relatively well conserved across tasks, especially within domains and similar confidence reporting paradigms.