A rational model of the Dunning–Kruger effect supports insensitivity to evidence in low performers
Abstract
Evaluating one's own performance on a task, typically known as 'self-assessment', is perceived as a fundamental skill, but people appear poorly calibrated to their abilities. Studies seem to show poorer calibration for low performers than for high performers, which could indicate worse metacognitive ability among low performers relative to others (the Dunning-Kruger effect). By developing a rational model of self-assessment, we show that such an effect could be produced by two psychological mechanisms, in either isolation or conjunction: influence of prior beliefs about ability or a relation between performance and skill at determining correctness on each problem. To disentangle these explanations, we conducted a large-scale replication of a seminal paper with approximately 4,000 participants in each of two studies. Comparing the predictions of two variants of our rational model provides support for low performers being less able to estimate whether they are correct in the domains of grammar and logical reasoning.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
Similar patterns are evidenced by prior research using AI-generated content, where users tend to overestimate their capabilities and those of AI.
“Similar patterns are evidenced by prior research using AI-generated content [38], where users tend to overestimate their capabilities and those of AI [28, 36].”
From How University Disability Services Professionals Write Image Descriptions for HCI Figures Using Generative AI (2026) · cites this paper as [36]
In Jansen et al. (2021) data, grammar test mean correct answers was 10.17 (SD=3.40) vs mean expectation 12.49 (SD=3.91); logic test mean performance 9.45 (SD=3.59) vs mean expectation 10.86 (SD=4.05).
“In the grammar test, the mean number of correct answers was 10.17 ( SD = 3.40) compared with a mean expectation of 12.49 ( SD = 3.91). In the logic test, the mean performance was 9.45 ( SD = 3.59), and the mean expectation was 10.86 ( SD = 4.05).”
From Talking the talk, not walking the walk: The coevolution of overconfidence and loss aversion (2026) · cites this paper as Jansen et al. (2021)
Regressing realized performance on expected performance shows self-beliefs are positively related to true ability (a increasing in b); expectations significantly predicted actual scores for grammar (β=0.25, 95%CI [0.22,0.28], t=16.26, p<.001) and logic (β=0.18, 95%CI [0.15,0.21], t=11.24, p<.001).
“The results of regressing realized performance on expected performance show that self-beliefs are positively related to true ability; that is, a is increasing in b . Speci fi cally, expectations signi fi cantly predicted actual scores for both the grammar ( β = 0.25, 95%CI [0.22, 0.28], t = 16.26, p < .001) and logic tasks ( β = 0.18, 95%CI [0.15, 0.21], t = 11.24, p < .001).”
From Talking the talk, not walking the walk: The coevolution of overconfidence and loss aversion (2026) · cites this paper as Jansen et al. (2021)
The data supporting these findings are from Jansen et al. (2021), publicly available on OSF; the complete analysis script is openly available; this study was not preregistered.
“The data that support these fi ndings are from Jansen et al. (2021), publicly available on the Open Science Framework at https://osf.io/ er9ms/. The complete analysis script to replicate the results is openly available in the Open Science Framework at https://osf.io/vpxqg (Dawson, 2026). This study was not preregistered.”
From Talking the talk, not walking the walk: The coevolution of overconfidence and loss aversion (2026) · cites this paper as Jansen et al. (2021)
Cites
Conducting behavioral research on Amazon’s Mechanical Turk
Mason, W., & Suri, S. (2012) · Behavior Research Methods · doi:10.3758/s13428-011-0124-6 · Cited by 3,172 · see citing works
Flawed Self-Assessment: Implications for Health, Education, and the Workplace
Dunning, D., Heath, C., & Suls, J. M. (2004) · Psychological Science in the Public Interest · doi:10.1111/j.1529-1006.2004.00018.x · Cited by 1,865 · see citing works
Flagged first, then by citations
Cited by
Sorted by citations
A foundation model to predict and capture human cognition
Binz, M., Akata, E., Bethge, M., Brändle, F., Callaway, F., Coda-Forno, J., Dayan, P., Demircan, C., Eckstein, M. K., Éltető, N., et al. (2025) · Nature · doi:10.1038/s41586-025-09215-4 · Cited by 119 · see citing works
Self-citationKnowing About Knowing: An Illusion of Human Competence Can Hinder Appropriate Reliance on AI Systems
He, G., Kuiper, L., & Gadiraju, U. (2023) · Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems · doi:10.1145/3544548.3581025 · Cited by 107 · see citing works
“Evaluating one’s own performance on a task, typically known as “self-assessment”, is perceived as a fundamental skill, but people appear to calibrate their abilities [39] poorly.”
BackgroundRaees, M., Khan, V.-J., Lykourentzou, I., & Papangelis, K. (2026) · Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems · doi:10.1145/3772318.3791467 · Cited by 5 · see citing works
Talking the talk, not walking the walk: The coevolution of overconfidence and loss aversion
Dawson, C., & de Meza, D. (2026) · Psychological Review · doi:10.1037/rev0000644 · Cited by 2 · see citing works
Raees, M., Iwamoto, Y., Papangelis, K., Heard, J., & Tigwell, G. W. (2026) · arXiv · doi:10.48550/arxiv.2602.08937 · Cited by 1 · see citing works
“Overall, HAI-Text was perceived as capturing essential visual details, such as differences or key quantitative values, though it very occasionally omitted some information.”
Performance and Metacognition Disconnect when Reasoning in Human-AI Interaction
Fernandes, D., Villa, S., Nicholls, S., Haavisto, O., Buschek, D., Schmidt, A., Kosch, T., Shen, C., & Welsch, R. (2024) · arXiv · doi:10.48550/arxiv.2409.16708 · Cited by 71
- Authors
- Jansen, R. A., Rafferty, A. N., & Griffiths, T. L.
- Affiliations
- University of California, Berkeley (US)
- Carleton College (US)
- Princeton University (US)
- Corresponding author
- Jansen, Rachel A.
- Venue
- Nature Human Behaviour
- Published
- 2021-02-25
- Volume
- 5
- Issue
- 6
- Pages
- 756-763
- ISSN
- 2397-3374 (ISSN-L)
- OpenAlex
- W3129534106
- Cited by
- 100
- References
- 2
- Citations by year
- 2026: 21 · 2025: 18 · 2024: 15 · 2023: 22 · 2022: 14 · 2021: 8 · 2020: 1 · 2019: 1
- FWCI
- 9.88
- Type
- journal article
- Language
- en
- Open access
- closed
- Licence
- Locations
- Nature Human Behaviour (publishedVersion)
- PubMed (publishedVersion)
- RePEc: Research Papers in Economics (submittedVersion)
- Topics
- Social Sciences › Decision Sciences › General Decision Sciences › Decision-Making and Behavioral Economics
- Life Sciences › Neuroscience › Cognitive Neuroscience › Memory Processes and Influences
- Social Sciences › Decision Sciences › Management Science and Operations Research › Psychometric Methodologies and Testing
- Keywords
- Dunning-Kruger effect, self-assessment, metacognition, calibration, rational model, low achievers, metacognitive judgments, grammar, logical reasoning
- MeSH terms
- Metacognition, Adolescent, Adult, Aged, Aged, 80 and over, Aptitude, Female, Humans, Male, Middle Aged, Models, Theoretical, Self Concept, Young Adult
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
