The Dunning-Kruger effect says the least competent people are the most overconfident. A new study from Bath and LSE found it is actually the other way around
In 1999, two psychologists at Cornell University published a paper that would eventually become one of the most-cited findings in popular psychology. David Dunning and Justin Kruger gave participants a series of tests on humor, logical reasoning, and grammar, then asked them to estimate how well they had done. The people who scored in the bottom quarter of performance dramatically overestimated their scores. The people who scored highest were more modest, sometimes even underestimating how well they had done.
The conclusion felt both profound and immediately recognizable: incompetent people lack the skills to recognize their own incompetence. They are, as the paper’s title put it, “unskilled and unaware of it.” The Dunning-Kruger effect entered the culture as shorthand for a specific and deeply satisfying observation about human behavior. Bad managers who think they are visionary leaders. Novices who argue with experts. People who know just enough to be dangerously wrong. The effect seemed to explain all of them.
A new paper published in Psychological Review, the flagship journal of the American Psychological Association, argues that the entire phenomenon is a statistical artifact. When the mathematical flaw at the heart of the original methodology is corrected, the pattern does not merely shrink. It reverses.
What the original studies got wrong
The critique is not new. Researchers have been questioning the statistical foundations of the Dunning-Kruger effect for over a decade. But the new paper from Professor Chris Dawson at the University of Bath and Professor David de Meza at the London School of Economics is the most comprehensive test of the claim to date, applying advanced statistical modeling to massive replication datasets rather than the small samples of the original studies.
The core problem is what happens when you sort people into groups purely based on their test scores. Test performance is inherently noisy. Whether you get a question right or wrong depends not only on your underlying ability but on whether you happened to know that particular question, whether you misread the phrasing, whether you were having a good day. Someone who genuinely belongs in the middle of the ability distribution might, through bad luck alone, end up in the bottom quarter of a given test. And when researchers then compare what those bottom-scorers predicted about their performance, they find a predictable mathematical distortion: because their actual score is unusually low due to bad luck, the gap between their prediction and their score looks enormous.
The same distortion operates in reverse for high scorers. Someone who ends up in the top quarter partly through lucky question selection will have a score that overestimates their true ability. Their prediction will look modest by comparison.
None of this has anything to do with how aware people are of their own competence. It is a consequence of the statistical method, not a psychological phenomenon.
“Our research demonstrates that the original methodology fell victim to a statistical illusion,” said Dawson. “The popular image of the incompetent person brimming with confidence tells us less about human psychology than it does about the hidden traps in statistical analysis.”
What the corrected data shows
When Dawson and de Meza applied statistical models that properly accounted for this noise, sorting participants by their estimated true ability rather than their raw test score, the classic Dunning-Kruger pattern vanished. In its place was something different: overconfidence was distributed across all ability levels, but it was most pronounced among the highest performers.
The smartest people, on the corrected analysis, were the most likely to think they had done even better than they actually had. The least competent people showed less overconfidence than the original data suggested, not because they had unusual insight into their limitations, but because the original analysis had mathematically inflated the apparent gap between their scores and their predictions.
“When we clear away the statistical problems, the apparent pattern disappears,” Dawson said. “Instead, we see that overconfidence is a universal human trait, but it is the most capable among us who exhibit it the most.”
This is not a minor revision to a secondary finding. It is a direct inversion of one of the most widely replicated and most confidently cited conclusions in modern psychology.
Why high performers might genuinely be more overconfident
Dawson and de Meza do not claim that the corrected result is random noise. They propose a specific explanation for why high-ability people might be authentically more overconfident, independent of the statistical correction.
Their argument draws on signaling theory from economics. Overconfidence, on this account, is not primarily a cognitive failure. It is a social strategy. When people need to communicate their competence to others, particularly in contexts where their actual ability is not directly observable, projecting confidence serves as a credible signal. And the logic of signaling suggests that high-ability individuals have the most to gain from this strategy. They can afford to be overconfident because they are more likely to be right, or at least to be right often enough that the occasional costly failure does not undermine the signal.
“It is plausible that poor performers may sometimes lack the cognitive ability to recognise their own incompetence,” the researchers write. “There are, though, other reasons why overconfidence arises. It can result from attempts to signal otherwise hidden ability, a mechanism that high-ability individuals have the most incentive to engage in.”
This is a fundamentally different model of overconfidence. In the Dunning-Kruger account, overconfidence is a deficiency, something that happens to people who are not smart enough to know what they do not know. In Dawson and de Meza’s account, overconfidence is adaptive behavior. It is what rational people with genuine competence do when they are trying to communicate that competence in a world where you cannot directly show others how good you are.
What this means for how you use the concept
The practical implications depend on how honest you want to be about what the Dunning-Kruger effect has been doing in public discourse. It has been used, almost universally, as a way of explaining other people’s overconfidence. Politicians who talk with certainty about topics they do not understand. Internet commenters who argue with specialists. Colleagues who overestimate their own contributions. The effect has provided a psychologically credentialed way of concluding that confident but wrong people are, at some level, too incompetent to know they are wrong.
If the corrected analysis holds up, that conclusion requires revision. Overconfidence appears to be a human universal, present across all ability levels, but correlated with competence rather than with its absence. The most confident people in the room may be the ones who have the best reasons to be confident, and who are deploying that confidence strategically.
Whether that makes overconfidence more or less forgivable depends on your point of view. What it makes considerably harder is using the Dunning-Kruger effect to diagnose other people’s incompetence while exempting yourself from the same analysis.
The study “Talking the Talk, Not Walking the Walk: The Coevolution of Overconfidence and Loss Aversion” was authored by Chris Dawson and David de Meza and published in Psychological Review on July 27, 2026.
Source: University of Bath and London School of Economics. DOI: 10.1037/rev0000644
