Discrete confidence levels revealed by sequential decisions

Abstract

Humans can meaningfully express their confidence about uncertain events. Normatively, these beliefs should correspond to Bayesian probabilities. However, it is unclear whether the normative theory provides an accurate description of the human sense of confidence, partly because the selfreport measures used in most studies hinder quantitative comparison with normative predictions. To measure confidence objectively, we developed a dual-decision task in which the correctness of a first decision determines the correct answer of a second decision, thus mimicking real life situations in which confidence guides future choices. Whilst participants were able to use confidence to improve performance, they fell short of the ideal Bayesian strategy. Instead, behavior was better explained by a model with a few discrete confidence levels. These findings question the descriptive validity of normative accounts, and suggest that confidence judgments might be based on point-estimates of the relevant variables, rather than on their full probability distributions.

Suprathreshold perceptual decisions constrain models of confidence

“…It is unclear if the flat-prior variants of these models also conform to the definition of Bayesian Confidence, which is unspecific about whether a true stimulus prior is required. Overall, our results do not support the Bayesian Confidence Hypothesis, in agreement with some previous studies [ 11 , 28 , 39 ] but not others [ 12 , 32 ]. Despite this finding, the results of the current study should not be interpreted as against Bayesian computations in the brain more generally.…”

“…While the Probability-Difference model was third best in terms of the number of observers best fit, the LPR model was overall a superior model at the group level and in the model recovery analysis showed some fit similarity to the second best-fitting Scaled-Distance model ( Fig 4F ). Reviewing previous studies, a high resolution for extreme probabilities does not match with findings of compression of continuous confidence probabilities to a few discrete levels for perceptual confidence [ 39 ] or knowledge of motor uncertainty distributions (i.e., motor confidence, [ 54 ]). Further work is needed to understand the representation of confidence for extreme probabilities.…”

“…They found that an Evidence-Strength metric with quadratic sensory-uncertainty-dependent bounds best fits the behaviour in both tasks. Lisi et al [ 39 ] tested if confidence could be used in a subsequent decision where the perceptual evidence depended on the correctness of previous perceptual choice, and found that the model with a discrete Evidence-Strength metric outperformed the Bayesian Probability-metric model. Finally, Li & Ma [ 12 ] investigated various Probability and Evidence-Strength metrics in the context of a three-alternative forced-choice task.…”

A metacognitive approach to the study of motion-induced duration biases reveals inter-individual differences in forming confidence judgements

“…Our ability to reflect on our own performance, and somehow access the processes and operations which take place in our mind when we are executing a task, allows us to assign levels of confidence to our decisions. These self-evaluations help us guide our future behaviour 21 . In perceptual tasks, when we take a correct decision, our confidence in being correct will increase 22,23 .…”

Confidence guides priority between forthcoming tasks

“…Humans can usually evaluate the extent to which they believe their decision is correct and express this degree of belief as a confidence estimate. It has long been known that people possess this ability 1 , but the cognitive 2 , 3 , computational 4 and neural 5 , 6 underpinnings of confidence judgments are still being investigated. How individuals can make use of their confidence estimates to control behavior is also an important question in this topic 3 , 7 – 11 .…”