Bits of Confidence: Metacognition as Uncertainty Reduction

The ability of humans to assess the correctness of their own decisions via confidence
judgments is a form of metacognition. This self-reflective act is essential for learning,
memory, consciousness, group decision, and many other aspects of cognition. Researchers
evaluate the quality of metacognition according to bias, sensitivity and efficiency. These
are often measured with such quantities as meta - d’, or M-ratio by inferring the potential
accuracy of the primary task from the secondary confidence rating performance. In the
present study, I offer a comprehensive account of confidence judgments and
metacognition, in terms of a communication system between stimuli, actor, rater, and
experimenter. Several information theory techniques are harnessed to uncover the
underlying components of information transmission between stimuli, actor and rater.
Within this framework, I advance three independent measures of metacognitive
sensitivity: meta − U, meta − KL, and meta − J . These are based on multivariate
uncertainty analysis, and applications of the Kullback-Leibler and Jeffrey’s divergences to
confidence-accuracy distributions. I then demonstrate the various desirable
characteristics of these information-theory measures, and provide considerable evidence
for their construct validity. In addition, I outline the structural and individual sources of
metacognitive inefficiency, according to an information-theory model of communication.
Keywords: Metacognition, Confidence, Information-theory, meta-d’, M-ratio, meta - I