Is the Dunning-Kruger Effect Real? What the IQ Data Actually Shows
In the famous 1999 study, students who scored at the 12th percentile guessed they ranked at the 62nd. When researchers later tested the effect against measured IQ, with statistics built to avoid a well-known trap, most of it disappeared. Not quite all of it.
What you need to know
- The 1999 paper did not measure IQ. It tested Cornell undergraduates on humour, logic and grammar. In the logic study, the bottom quarter scored at the 12th percentile and estimated its test score at the 62nd.
- Low scorers overestimated, but they never ranked themselves above high scorers. In the same study the top quarter scored at the 86th percentile and put its test score at the 68th, still above the bottom quarter's guess.
- Regression to the mean plus the near-universal habit of rating oneself above average reproduces the same quartile chart even when everyone judges themselves equally well. In a 2020 study of 929 adults tested on Raven's matrices, the chart appeared, but analyses built to avoid that trap found no Dunning-Kruger effect.
- Self-knowledge is real but rough: across 41 studies, self-estimates correlate about .33 with tested ability. It is also inflated almost everywhere. In two US surveys, 65% of adults agreed they were more intelligent than average.
The Dunning-Kruger effect is one of the most quoted findings in psychology. It usually arrives as a slogan — the less people know, the more confident they are — and often with a curve whose early peak is labelled "Mount Stupid". Neither the slogan nor the curve is what the original study reported. The real finding is narrower, it was never about IQ, and when researchers finally tested it against measured intelligence using statistics designed for the job, most of the effect went away. Most, not all, which is why the honest answer to "is it real?" takes more than one word.
What the 1999 study actually measured
Justin Kruger and David Dunning published "Unskilled and unaware of it" in the Journal of Personality and Social Psychology in 1999. It reported four studies, all run on Cornell University undergraduates taking part for extra credit. Each student completed a short test, then estimated how their ability and their score ranked against the other students as a percentile: the percentage of peers they believed they had outperformed. The results for the bottom quarter of scorers were striking:
- Humour, 65 students: rating 30 jokes against the judgements of professional comedians, the bottom quarter scored at the 12th percentile and placed its ability at the 58th
- Logical reasoning, 45 students: on 20 items taken from the LSAT, the bottom quarter scored at the 12th percentile, rated its ability at the 68th and estimated its test score at the 62nd
- Grammar, 84 students: on a 20-item test, the bottom quarter scored at the 10th percentile and put its test score at the 61st
- Logic problems, 140 students: on 10 Wason selection-task problems, the bottom quarter scored at the 13th percentile and put its test score at the 53rd
The authors' explanation was that the skill needed to get an answer right is the same skill needed to recognise a right answer, so people who lack it also lack the means to notice their mistakes. Two parts of the paper rarely survive the retelling. Low scorers never placed themselves above high scorers: in the logic study the top quarter scored at the 86th percentile and put its test score at the 68th, still above the bottom quarter's 62nd. And the effect responded to teaching. In the fourth study, bottom-quarter students given a short logic lesson cut their test-score estimate from the 51st percentile to the 32nd.
The famous curve is not in the paper
The graph that usually illustrates the effect — confidence rocketing up with a little knowledge, crashing into a valley, then climbing slowly with expertise — does not appear in the 1999 paper. Its figures show something much plainer: for each quarter of scorers, one point for the average actual percentile and one for the average self-estimate. The actual-score line climbs steeply from the bottom quarter to the top. The self-estimate line climbs too, only far more gently. The mountain-and-valley curve is a later drawing that has become attached to the study's name.
The bottom quarter overestimated itself, yet still ranked itself below the top quarter. The curve everyone shares shows a peak the data never contained.
Why the same chart appears from arithmetic alone
The quartile chart has a statistical weakness, and the original authors acknowledged part of it. Sort people by a single noisy test score and the bottom group fills up with people who had a bad day: their measured score sinks below their usual level while their self-estimate does not. That is regression to the mean. Combine it with the better-than-average effect, the near-universal habit of placing oneself above the middle, and the bottom quarter must look wildly overconfident while the top quarter looks modest, even if everyone judges themselves exactly as well as everyone else.
The size of that illusion is easy to calculate. Suppose self-estimates correlate .33 with tested ability, the figure a large meta-analysis reports, and that everyone overestimates by the same half a standard deviation. In a simulation of 400,000 people, the bottom quarter scores at the 12.5th percentile on average and places itself at about the 52nd; the top quarter scores at the 87.5th and places itself just under the 75th. That is the Dunning-Kruger pattern, produced with no difference at all in self-knowledge between low and high scorers. It is our own illustration rather than a published result, and it rests on the same normal-curve arithmetic as our IQ percentile calculator.
Published work reached the same point by other routes. Krueger and Mueller showed in the same journal in 2002 that the gap between low and high scorers disappeared once regression and the better-than-average effect were taken into account. Burson, Larrick and Klayman found in 2006 that on hard tasks the best performers misjudged their standing more than the worst, the reverse of the popular story. And a 2016 paper in the journal Numeracy by Nuhfer and colleagues showed that the standard quartile graph produces the pattern even from random numbers.
What happened when IQ was measured properly
The most direct test for intelligence came in 2020. Gignac and Zajenkowski, writing in the journal Intelligence, tested 929 adults in Warsaw on Raven's Advanced Progressive Matrices, a nonverbal reasoning test, and asked each person to rate their own intelligence. The average self-assessed IQ was 123.76; the average measured IQ was 101.70. Self-ratings and scores correlated .28, which is related but loose. Our piece on what a Raven's score actually is covers the test they used.
Plotted the classic way, the data produced a textbook Dunning-Kruger chart: the lowest-scoring quarter overestimated its IQ by 32.63 points, the highest by 12.68. But two tests designed to sidestep the regression artefact, one asking whether self-estimates become less accurate as ability falls and one asking whether the relationship bends at the low end, found nothing statistically significant. The authors put their verdict in the paper's title: the effect is "(mostly) a statistical artefact".
A 2023 replication in the same journal, by Dunkel, Nedelec and van der Linden, used the far larger US Add Health survey, 13,977 people with a vocabulary-based measure of ability. It did find a statistically significant effect, but a minimal one: the bend it detected explained about 1% of the variation in self-estimates. How much genuinely survives is still argued. A 2021 study in Nature Human Behaviour by Jansen, Rafferty and Griffiths, built on two large online replications with grammar and logical reasoning tests, concluded that low performers really are less sensitive to whether their answers are right. A pre-registered 2022 study led by McIntosh found that poor performers were appropriately less confident than good ones.
What people actually know about their own intelligence
Set the quartile debate aside and a sturdier picture emerges. People are rough but not useless judges of their own ability. A 2012 meta-analysis in Psychological Bulletin by Freund and Kasten, pooling 154 effect sizes from 41 studies, put the correlation between self-estimated and tested ability at .33, and a 2024 update across 115 samples and 36,833 people found .30. A correlation of that size says something real about groups while leaving a wide margin for any one person.
People are also inflated judges, and not only at the bottom. In two nationally representative US surveys reported in PLOS ONE in 2018, 65% of adults agreed that they were more intelligent than the average person, against 23% who disagreed, and men were more likely to agree than women. In the Warsaw data even the top quarter rated itself above its measured score. Overestimation is the rule across the ability range; what the careful studies dispute is whether it is really worse at the bottom.
- A confident self-rating is weak evidence in either direction, because self-estimates track tested ability at only about .3
- A low scorer who overrates themselves is doing what almost everyone does, not revealing a special blind spot
- High ability does not protect against misjudgement, and our article on why smart people make bad decisions covers the reasoning errors intelligence does not prevent
- Claims about "the incompetent" built on this study stretch four university experiments far beyond what they tested
Where would your own score land?
A self-estimate of intelligence correlates only about .33 with a tested score. If you want to know where you stand, measure it rather than guess.
Find your IQ score now! →The fairest summary is less quotable than the slogan. People are rough judges of their own ability, most of them believe they are above average, and the image of the unskilled person who thinks they are brilliant owes more to the arithmetic of sorting noisy scores than to human nature. If you would rather have a number than an impression, our IQ test reports a score with its percentile attached, and our note on why a score is a range, not a number explains how precisely to read it.
Common questions
Is the Dunning-Kruger effect real?
Partly. The basic pattern, low scorers overestimating their rank, replicates, but much of it follows from regression to the mean and the tendency to rate oneself above average. Studies that test it against measured IQ with methods designed to avoid those artefacts find either no effect or a very small one.
What did the original Dunning-Kruger study find?
In four studies of Cornell undergraduates tested on humour, logic and grammar, the bottom quarter of scorers overestimated its standing. In the logic study it scored at the 12th percentile but estimated its test score at the 62nd, still below the top quarter's estimate of the 68th.
Is the Mount Stupid graph from the Dunning-Kruger paper?
No. The 1999 paper plotted the average actual percentile and the average self-estimated percentile for each quarter of scorers. The confidence curve with an early peak and a valley is a later illustration that became attached to the study's name.
Do people with low IQ overestimate their intelligence?
Most people overestimate, whatever their score. In a 2020 study of 929 adults, the average self-assessed IQ was 123.76 against a measured 101.70, and analyses built to avoid statistical artefacts found no sign that low scorers misjudge themselves more than high scorers do.
Sources for this story
- Unskilled and unaware of it: How difficulties in recognizing one's own incompetence lead to inflated self-assessments — Kruger, J., & Dunning, D. (1999), Journal of Personality and Social Psychology, 77(6), 1121-1134
- Unskilled, unaware, or both? The better-than-average heuristic and statistical regression predict errors in estimates of own performance — Krueger, J., & Mueller, R. A. (2002), Journal of Personality and Social Psychology, 82(2), 180-188
- The Dunning-Kruger effect is (mostly) a statistical artefact: Valid approaches to testing the hypothesis with individual differences data — Gignac, G. E., & Zajenkowski, M. (2020), Intelligence, 80, 101449
- Reevaluating the Dunning-Kruger effect: A response to and replication of Gignac and Zajenkowski (2020) — Dunkel, C. S., Nedelec, J., & van der Linden, D. (2023), Intelligence, 96, 101717
- How smart do you think you are? A meta-analysis on the validity of self-estimates of cognitive ability — Freund, P. A., & Kasten, N. (2012), Psychological Bulletin, 138(2), 296-321
- 65% of Americans believe they are above average in intelligence: Results of two nationally representative surveys — Heck, P. R., Simons, D. J., & Chabris, C. F. (2018), PLOS ONE, 13(7)
Corrections: spotted an error? Email corrections@iqmetrics.org and we will update this story and note the change here.
Related stories
All news →Does Music Raise IQ? What the Mozart Effect Study Actually Measured
The 1993 experiment behind three decades of "classical music makes babies smarter" ran for ten minutes, used college students, measured a single spatial task, and never mentioned IQ at all.
The Left-Brain/Right-Brain Myth: What Brain Scans Actually Show
The idea that everyone leans on one dominant hemisphere — logical on the left, creative on the right — traces back to a real, Nobel Prize-winning discovery. It is not what that discovery found.
Is IQ Debunked? What Critics Get Right and Wrong
The score is stable, predicts exam results and survival, and is not debunked. What critics have challenged, with more success, is the idea that it is fixed, complete or explained by a single cause.
Read the research.
Then find your own number.
Our IIF-certified assessment reports your score with its scale, percentile and confidence range — and a breakdown of the cognitive domains behind it.
Start IQ Test →