Is the Dunning-Kruger Effect Real
The idea that the least competent people are the most confident has become shorthand for a whole kind of person. The research is less settled than that. The Dunning-Kruger effect describes a pattern in grouped data, and twenty-five years of argument have been about how much of that pattern is psychology and how much is arithmetic.
Answer to the question
Is the Dunning-Kruger effect real?
Partly. People do misjudge where they stand, and in grouped data low scorers overshoot the most. But a test on 929 adults found the pattern mostly statistical, and a sample of 13,977 found a significant but very small effect. The claim that the least competent are the most confident goes beyond the evidence.
Key takeaways
- Kruger and Dunning (1999) ran four studies with Cornell undergraduates; in the logic study the bottom quarter scored at the 12th percentile and put their test performance at the 62nd.
- Their fourth study, with 140 students, found that a short logic training packet made low scorers judge their own answers more accurately, while untrained low scorers did not improve.
- Burson and colleagues (2006) ran 12 tasks across three studies and found that on harder tasks the best performers judged their standing less accurately than the worst, so error was not confined to low scorers.
- Gignac and Zajenkowski (2020) tested 929 adults: grouping by quarters reproduced the pattern, two tests that avoid grouping did not, and a simulation produced the pattern from data that had none built in.
- Dunkel and colleagues (2023) repeated the analyses on 13,977 Americans and found a significant but small effect, with the curved term adding about 1 per cent to the variation explained.
Justin Kruger and David Dunning argued that a lack of skill hides itself. The knowledge that answers a logic question correctly also recognises a wrong answer, so the lowest-scoring 25 per cent are badly placed to notice. The Dunning-Kruger effect names the result: low scorers who rate themselves well above where they stand.
What the original studies reported
The 1999 paper rests on four studies of Cornell undergraduates, covering humour, logical reasoning, grammar and logic again. In the reasoning study, 45 students sat a 20-item test drawn from an LSAT preparation guide. Those in the bottom quarter scored, on average, at the 12th percentile and put their test performance at the 62nd. The abstract compresses the pattern into one sentence: "Although their test scores put them in the 12th percentile, they estimated themselves to be in the 62nd."
The grammar study added a second phase. Weeks later, the top and bottom scorers graded five of their peers' tests and rated themselves again. Top scorers raised their estimates, which the authors read as a false-consensus effect, while the bottom scorers did not significantly revise theirs.
The fourth study went further than measurement. Of 140 students taking a ten-item logic test, a randomly chosen half read a short training packet on logical reasoning and the other half did an unrelated task. Trained low scorers judged their own performance more accurately and were better at telling which of their answers were right. Untrained low scorers did not improve. For the authors, that was the mechanism: give people the skill and you give them the means to see the gap.
Adrian Furnham, in his chapter on the effect in The New Psychology (2024), keeps the reading of the Dunning-Kruger effect narrow. It describes unskilfulness in a particular domain, not low intelligence, and everyone is unskilled at most things. He also notes that the concept "has attracted a great deal of critical appraisal".
Objections that began as arithmetic
The first appraisal came in 2002, within three years of the paper, from Joachim Krueger and Ross Mueller, and it was statistical. Self-estimates are noisy, and people on average rate themselves above average. Any group picked for an extreme score, such as the bottom quarter, will sit closer to the middle on a second measurement. Regression to the mean and the better-than-average effect together could produce the familiar chart of the Dunning-Kruger effect with no deficit in insight at all.
Both objections accept the gap in the chart and dispute its cause.
Their own replication supports the argument. Errors were no longer asymmetric when either regression or the better-than-average effect was statistically removed, and high performers, not low ones, were the more error prone, because they were more likely to neglect their own estimate of how others had done.
A second objection arrived from a different direction. Katherine Burson and colleagues ran 12 tasks across three studies in which people judged how they ranked against others. On moderately difficult tasks, the best and worst performers differed little in the accuracy of their judgments. On harder tasks, the best performers were less accurate than the worst. Judges at every skill level, they proposed, make similar errors, and how difficult a task seems pushes everyone's estimate in the same direction. Both objections accept the gap in the chart and dispute its cause.
The statistical objection to the Dunning-Kruger effect got a direct test in 2020, from Gilles Gignac and Marcin Zajenkowski. They had 929 adults from the community rate their own intelligence on a single scale, then sit the Advanced Progressive Matrices, a standard reasoning test scored against age norms. People rated themselves far above the measured result, at about 124 on an IQ-like scale against a measured average of about 102, and the link between the two was modest.
Sorted into quarters, as the original studies did it, the data gave the familiar picture. The gap between belief and measurement shrank from about 33 points in the bottom quarter to about 13 in the top. Two analyses that do not depend on grouping found nothing: one asked whether errors changed with ability, the other whether a curved line fitted better than a straight one. In simulated data the quartile method produced a Dunning-Kruger-like pattern where none had been built in. The authors concluded that the effect is mostly, "rather than entirely", a statistical artefact.
In 2023 Avram Hiller objected, in a published comment, that the equal errors across ability levels may come from how the self-ratings were converted to a linear scale. Gignac and Zajenkowski re-ran the analysis with the transformation he suggested and again found no support for the effect.
Why a small effect survived the rebuttals
A different line of work asked what a perfectly rational judge would do. Rachel Jansen, Anna Rafferty and Thomas Griffiths, writing in Nature Human Behaviour in 2021, built a rational model of self-assessment and showed that two mechanisms, alone or together, could produce the effect. One is the pull of prior beliefs about one's own ability, which shrinks surprising results towards expectation even in a judge with the same insight at every skill level. The other is a link between performance and skill at telling whether each answer is correct.
They then ran large replications of the original paper, with about 4,000 participants in each of two studies. Comparing two variants of the model, they reported support for low performers being less able to tell whether they were correct, in grammar and logical reasoning. Matan Mazor and Stephen Fleming, commenting on that work, report that the model with performance-dependent noise fitted the online data better than shrinkage alone. So the pattern is not only shrinkage.
They also ask whether the extra noise deserves the word metacognitive, since standard models of decision-making predict more uncertainty in low performers with no added burden. The effect, they suggest, may carry the signature of a single burden rather than a double one.
A nationally representative American sample of 13,977 people gave the 2020 analyses a second run, led by Curtis Dunkel. His team pointed out that more than half of the original sample were undergraduates, which may have narrowed the range of ability. With a curved term in the model, they found a statistically significant Dunning-Kruger effect, but a small one: the curve added about 1 per cent to the variation explained. They judged it "premature" to call the effect mostly an artefact.
Which claims does the evidence still support?
The three claims in circulation have different records. That people are only loosely accurate about where they stand holds in every dataset above. That low scorers overshoot by more than everyone else is partly a product of how the data are grouped, and a test built to avoid the grouping found little in one large sample and a small effect in another. That incompetence causes confidence, which is how the effect is usually repeated, does not follow from any of it.
Trained and untrained low scorers were selected on the same test, so regression to the mean applies to both groups equally.
The training study is the piece the statistical objections reach least directly. Trained and untrained low scorers were selected on the same test, so regression to the mean applies to both groups equally, and the difference between them is not regression. That reading is ours, and one experiment with undergraduates is a narrow base for a mechanism. That makes calibration, which can be measured and trained, the better question to ask about it.
A reader meeting the claim in the wild has one practical check. Ask whether people were sorted by score and then compared with their own estimates. If they were, part of the gap was built in by the sorting.
Common questions
Is the Dunning-Kruger effect real?
Partly. People do misjudge their standing, and low scorers overshoot most in grouped data. But a 929-person test found the pattern mostly statistical, and a sample of 13,977 found a significant but very small effect. The popular claim that the least competent are the most confident goes beyond the evidence.
Does the Dunning-Kruger effect mean unintelligent people are overconfident?
No. The effect concerns skill in one domain, such as logic or grammar, not general intelligence. Everyone is unskilled at most things. In Burson and colleagues' 12 tasks, the best performers judged less accurately than the worst on harder ones, so confidence errors are not confined to low scorers.
Can training reduce the Dunning-Kruger effect?
In the 1999 fourth study it did. Low scorers who received a short logic training packet judged their own performance more accurately, while untrained low scorers did not. It was one experiment with 140 undergraduates.
Sources
Kruger, J., & Dunning, D. (1999). “Unskilled and unaware of it: How difficulties in recognizing one’s own incompetence lead to inflated self-assessments.” Journal of Personality and Social Psychology, 77(6), 1121-1134. doi:10.1037/0022-3514.77.6.1121
Krueger, J., & Mueller, R. A. (2002). 'Unskilled, unaware, or both? The better-than-average heuristic and statistical regression predict errors in estimates of own performance.' Journal of Personality and Social Psychology, 82(2), 180-188. doi:10.1037/0022-3514.82.2.180
Burson, K. A., Larrick, R. P., & Klayman, J. (2006). 'Skilled or unskilled, but still unaware of it: How perceptions of difficulty drive miscalibration in relative comparisons.' Journal of Personality and Social Psychology, 90(1), 60-77. doi:10.1037/0022-3514.90.1.60
Gignac, G. E., & Zajenkowski, M. (2020). 'The Dunning-Kruger effect is (mostly) a statistical artefact: Valid approaches to testing the hypothesis with individual differences data.' Intelligence, 80, 101449. doi:10.1016/j.intell.2020.101449
Hiller, A. (2023). 'Comment on Gignac and Zajenkowski, The Dunning-Kruger effect is (mostly) a statistical artefact: Valid approaches to testing the hypothesis with individual differences data.' Intelligence, 97, 101732. doi:10.1016/j.intell.2023.101732
Gignac, G. E., & Zajenkowski, M. (2023). 'Still no Dunning-Kruger effect: A reply to Hiller.' Intelligence, 97, 101733. doi:10.1016/j.intell.2023.101733
Jansen, R. A., Rafferty, A. N., & Griffiths, T. L. (2021). 'A rational model of the Dunning-Kruger effect supports insensitivity to evidence in low performers.' Nature Human Behaviour, 5(6), 756-763. doi:10.1038/s41562-021-01057-0
Mazor, M., & Fleming, S. M. (2021). 'The Dunning-Kruger effect revisited.' Nature Human Behaviour, 5(6), 677-678. doi:10.1038/s41562-021-01101-z
Dunkel, C. S., Nedelec, J., & van der Linden, D. (2023). 'Reevaluating the Dunning-Kruger effect: A response to and replication of Gignac and Zajenkowski (2020).' Intelligence, 96, 101717. doi:10.1016/j.intell.2022.101717