← Did You Know?
📊Mind8 min read

The Dunning-Kruger Effect

The famous graph with 'Mount Stupid' on it does not appear in the study. Neither does the claim that incompetent people think they are brilliant. And a growing number of statisticians think the effect is largely an illusion produced by the way the data is plotted.

In plain English

The popular version goes like this: stupid people are too stupid to realise they are stupid, so they walk around brimming with confidence, while genuinely clever people are wracked with self-doubt.

That is not what the study found.

In 1999, David Dunning and Justin Kruger at Cornell ran a series of tests on undergraduates covering humour, logical reasoning and grammar. After each test they asked participants to estimate how well they had done relative to everyone else, as a percentile.

The pattern in the data was this. People in the bottom quarter scored around the 12th percentile and guessed they were around the 62nd. That is a very large gap. People in the top quarter scored around the 86th percentile and guessed around the 70th, so they underestimated themselves.

Two things follow that the meme gets wrong.

First, the worst performers did not think they were exceptional. They thought they were slightly above average. That is a much more ordinary and much more human failing.

Second, almost everyone in the study placed themselves above the middle, including people who were genuinely good. The bottom group was not uniquely arrogant. It just had the furthest to fall.

The proposed explanation was elegant: the skills needed to do something well are often the same skills needed to judge whether you have done it well. If you do not know grammar, you cannot spot your grammatical errors, and you also cannot spot that you cannot spot them. Dunning later called this a "double curse".

Then the statisticians arrived.

Five things to file under "wait, what?"

  • The famous curve is not from the paper. The chart everybody shares, the one with a towering "Peak of Mount Stupid" followed by a "Valley of Despair" and a slow climb to a "Slope of Enlightenment", appears nowhere in Kruger and Dunning's research. It is an internet creation, and the shape has clearly been borrowed from the Gartner hype cycle. The actual figures in the paper show two lines rising gently and converging. There is no peak. There is no valley.

  • The study was inspired by a bank robber and lemon juice. In 1995, McArthur Wheeler robbed two Pittsburgh banks in broad daylight with no disguise, having rubbed lemon juice on his face. He believed, apparently sincerely, that because lemon juice is used as invisible ink it would render his face invisible to security cameras. He was baffled when the footage was shown to him. Dunning read the news report and started wondering whether incompetence hides itself.

  • Random numbers reproduce the effect. This is the serious problem. Because participants are ranked by their test score and then compared to their self-estimate, and because both measures contain noise, regression to the mean guarantees the low scorers will look like over-estimators and the high scorers like under-estimators. Nuhfer and colleagues showed that simulated data, generated with no relationship at all between skill and self-assessment, produces the same characteristic crossing pattern. If pure noise draws the graph, the graph is not evidence of the psychological mechanism.

  • Nearly everyone thinks they are above average, and that alone explains a lot of it. The better-than-average effect is one of the most robust findings in social psychology. In one famous survey, 93% of American drivers rated themselves above the median. If everyone shifts their estimate upward by a similar amount, the people who are genuinely worst will automatically show the biggest gap between belief and reality, without any special inability to self-assess.

  • The finding may not travel. The original participants were Cornell undergraduates, a narrow slice of humanity. Attempts to replicate the pattern in East Asian samples have often found the opposite tendency, with participants under-rating rather than over-rating themselves. Self-enhancement is not obviously a human universal, and a supposedly cognitive effect that reverses across cultures is doing something other than what it says on the tin.

The full story

What the 1999 paper actually claimed

The paper is titled "Unskilled and Unaware of It: How Difficulties in Recognizing One's Own Incompetence Lead to Inflated Self-Assessments", published in the Journal of Personality and Social Psychology.

The central claim was about metacognition, meaning your ability to assess your own thinking. Dunning and Kruger argued that competence and the ability to evaluate competence draw on the same underlying knowledge, so a deficit in one produces a deficit in the other.

They supported this with a neat additional result. In one study, they trained the poor performers in logical reasoning. After training, those participants not only performed better, they also revised their earlier self-assessments downward. Gaining the skill improved their ability to see they had previously lacked it.

That last finding is genuinely difficult for the purely statistical explanation to account for, and it is the strongest thing the original research has going for it.

The statistical objection

The criticism is not that the researchers made an arithmetic mistake. It is that the analysis has a structural feature that manufactures the result.

Consider what happens when you sort people by their measured score. Any measurement includes some real signal and some random error. The people who land in the bottom group are disproportionately those whose random error happened to be negative that day, and the people at the top disproportionately got lucky. So the bottom group's true ability is, on average, higher than their measured score suggests, and the top group's is lower.

Now compare each group to their self-estimates. The bottom group will look like they over-estimated. The top group will look like they under-estimated. This happens automatically, from the sorting procedure alone, whether or not anybody is bad at self-assessment.

Ed Nuhfer and colleagues made this concrete in papers in 2016 and 2017, generating random data and showing it produced the familiar picture. Gilles Gignac and Marcin Zajenkowski published an analysis in 2020 concluding that once you model the data appropriately, most of the apparent effect is accounted for by the better-than-average effect plus regression to the mean, with little left over.

Where the argument stands

It would be wrong to say the effect has been debunked, and equally wrong to present it as settled science.

The strongest position for the sceptics is that the standard chart is an artefact and should never be used as evidence of anything. That case is solid.

The strongest position for the defenders is the training result, and the broader body of work on metacognition, which does find real relationships between expertise and accurate self-evaluation using methods that do not rely on the problematic comparison. Dunning has responded to the critiques and maintains that a genuine metacognitive effect remains once the statistical artefacts are stripped out.

What almost nobody defends is the popular version. Dunning himself has repeatedly pushed back on the idea that the research shows stupid people are confident and smart people are not, and has pointed out that the effect, properly understood, is about all of us. The domains in which you are unknowingly unskilled are invisible to you by construction, and everyone has some.

Why the myth is so sticky

The meme version is enormously more satisfying than the finding. It supplies a scientific-sounding label for an irritating person, and, crucially, it lets the person using it place themselves on the enlightened part of a curve that does not exist.

There is a tidy irony in a piece of research about the limits of self-assessment becoming famous mainly as a way of confidently diagnosing other people. It is probably the most-cited psychology finding that most of the people citing it have not read.

Go deeper

For the curious:

On YouTube:

Keep going

More things worth knowing

All topics →