The Metacognitive Sensitivity of Verbal and Numerical Confidence Reports

This study demonstrates that verbal confidence reports are more sensitive than numerical ones in distinguishing correct from incorrect perceptual decisions, while both formats prove equally effective in facilitating accurate judgment revisions during dyadic collaboration.

The Gist

Imagine you are playing a game where you have to guess which way a blurry arrow is pointing. After you make your guess, you have to tell a friend how sure you are about it. You can do this in two ways: you can give a number (like "I'm 80% sure") or use words (like "I'm very likely right").

This paper is like a detective story investigating which of these two methods actually tells the truth better about how well your brain is working.

The Big Surprise: Words vs. Numbers
Usually, we assume that numbers are the "gold standard" for precision. We think that saying "80%" is more exact and honest than saying "very likely." It's like thinking a digital scale is always more accurate than a person's gut feeling about weight.

However, this study found the opposite. When people used words to describe their confidence, those words were actually better at separating the "good guesses" from the "bad guesses."

• The Analogy: Think of verbal confidence as a high-quality spotlight. When someone says "very likely," that spotlight shines brightly on the correct answers and dimly on the wrong ones.
• The Contrast: Numerical confidence (the percentages) was more like a flickering, dim light. It didn't distinguish between right and wrong answers as clearly as the words did. Even though we think numbers are more precise, our brains seem to express their true certainty better through language.

The Teamwork Test
The researchers also set up a scenario where two people had to work together. One person made a guess and shared their confidence, and the other person had to decide whether to stick with their own guess or change it based on their partner's input.

Here, the "spotlight" and the "flickering light" performed equally well. Whether the partner used words or numbers to share their confidence, the team was just as good at correcting mistakes and finding the right answer. It didn't matter which language they spoke; the message got through effectively in both cases.

The Bottom Line
The main takeaway is that we shouldn't underestimate the power of simple words. When it comes to telling someone how sure we are about a decision, saying "I'm very confident" might actually be a more honest and accurate signal than trying to pin down a specific percentage. While numbers are great for math, words seem to be the superior tool for sharing the feeling of certainty in our everyday decisions.

Technical Summary

Technical Summary: The Metacognitive Sensitivity of Verbal and Numerical Confidence Reports

Problem Statement
This study addresses a fundamental question in the psychology of decision-making: how the format of confidence communication influences the accuracy of metacognitive judgments. While numerical probability reports (e.g., "80% certainty") are commonly assumed to offer more precise and granular representations of uncertainty compared to verbal expressions (e.g., "very likely"), this assumption has not been rigorously tested regarding their ability to distinguish between correct and incorrect perceptual decisions. The paper investigates whether numerical formats genuinely provide superior metacognitive sensitivity or if verbal expressions are equally, or even more, effective in signaling decision accuracy.

Methodology
The research employed a multi-phase experimental design involving perceptual decision-making tasks.

1. Individual Confidence Reporting: Participants performed perceptual tasks and subsequently reported their confidence in their choices using two distinct formats: verbal expressions (categorical linguistic labels) and numerical probabilities (percentage estimates). The study measured the relationship between these reports and the actual correctness of the decisions to assess metacognitive sensitivity.
2. Dyadic Decision-Making: The researchers extended the investigation to a social context involving pairs of participants. In this phase, individuals made initial decisions and confidence reports, then had the opportunity to revise their judgments after observing a partner's choice and their expressed confidence. This setup tested whether the format of confidence (verbal vs. numerical) influenced the efficacy of social learning and the accuracy of judgment revisions.

Key Contributions and Results
The study yields two primary findings that challenge prevailing assumptions about numerical precision in uncertainty reporting:

• Superiority of Verbal Expressions in Metacognitive Sensitivity: Contrary to the hypothesis that numerical reports provide finer-grained data, the results indicate that verbal expressions more reliably distinguish between correct and incorrect choices. Participants' verbal confidence ratings showed a stronger correlation with decision accuracy than their numerical probability estimates.
• Equivalence in Dyadic Revision: In the collaborative setting, the format of the confidence report did not significantly alter the outcome of social interaction. Both verbal and numerical reports were found to be equally effective in facilitating accurate revisions of initial judgments when participants incorporated a partner's input.

Significance
The paper concludes that verbal expressions are a highly effective, and in some contexts superior, means of conveying decision confidence. These findings underscore the functional utility of natural language in metacognition, suggesting that the common preference for numerical probabilities as the "gold standard" for precision may be misplaced when the goal is to accurately signal the likelihood of being correct. The results contribute to a more nuanced understanding of how uncertainty is represented and communicated, highlighting that the modality of expression (verbal vs. numerical) plays a critical role in the fidelity of metacognitive signals.