Cognitive biases
The Dunning–Kruger effect: what the research actually showed
You have probably seen this picture: a curve shoots up to “Mount Stupid,” where a beginner feels sure of themselves, then falls into a “valley of despair” and slowly crawls toward a “plateau of wisdom.” It is often captioned “the Dunning–Kruger effect.” That graph does not appear anywhere in Dunning and Kruger’s research.
The Dunning–Kruger effect describes something else: people who score low on a test, on average, greatly overestimate their result, and people who score high slightly underestimate theirs. That observation itself has been reproduced. Scientists argue about why it appears and how much of it is explained by plain statistics.
Below, we look at what the original study found, why it is criticized, and what is useful for everyday life. In the middle of the article there is a simulation in which the famous graph appears out of random numbers.
Contents
The short answer: the popular version and the data
What the 1999 study found
Justin Kruger and David Dunning gave students tests of logical reasoning, grammar, and sense of humor. After the test, participants rated how well they had done compared with others.
Participants in the bottom quarter by actual results landed, on average, at about the 12th percentile, which means they did worse than almost nine out of ten people. Yet they placed themselves, on average, at about the 62nd percentile, above the middle. Participants in the top quarter, by contrast, slightly underestimated their level.
The authors proposed an explanation: to judge your work well, you need the same knowledge that the work itself requires. Someone who cannot see the error in a problem will not see it in their own answer either. What let the strong participants down was something else, the authors argued: they assumed the task was just as easy for everyone else.
It sounds convincing. And for a long time this version was considered the main one.
Why scientists disagree
The problem lies in how the famous graph is built. Participants are split into four groups by their actual result, from the worst quarter to the best. For each group, two points are drawn: how they actually did and how they rated themselves. It turns out that a similar picture arises almost automatically, even when people know nothing at all about their own level.
A simple example. A class takes a test, and before the grades are handed out, each student is asked to guess their place among classmates. Suppose nobody understands anything and everyone answers at random, most of them slightly optimistic: “somewhere above the middle.” Now look at the people who actually ended up at the very bottom. Their random guess is almost inevitably higher than their real place, simply because there is nowhere lower to go. For those who ended up at the very top, it is the reverse: any guess except the highest will fall below their real place. The result is “weak ones overestimate, strong ones underestimate,” although there is no special blindness here. In statistics this is called regression to the mean. Joachim Krueger and Ross Mueller showed that these two factors, regression and the “I’m better than average” tendency, predict a similar picture.
Later, Edward Nuhfer and colleagues built the same graph from random numbers and got the familiar shape. And Gilles Gignac and Marcin Zajenkowski tested the effect on intelligence test data with stricter methods and concluded that it is mostly explained by a statistical artifact. That is one study on one type of test, but the conclusion is notable.
There are arguments on the other side too. In later studies by Dunning and colleagues, overestimation among people with low results persisted in real-life settings: in exams, in debates. It did not disappear even when people were paid for accurate self-assessment.
An honest bottom line looks like this. People with low results overestimate themselves on average, and this is a stable observation. But how much of it is a special “blindness of the incompetent,” and how much is ordinary statistics plus a general mild optimism, remains an open question.
See how chance draws the same graph
You can check this classroom example with your own hands.
Step 1. We create 400 fictional participants. Each gets a random test score and a random self-assessment, slightly above the middle. The self-assessment has no connection to the score: participants are just guessing.
Step 2. We split the participants into four groups by actual result, as in the 1999 study.
Step 3. We draw two lines. The dashed line shows how each group actually did. The green line shows how the group rated itself.
Step 4. Move the slider. It sets how closely the participants’ self-assessment is tied to their actual result: from “guessing at random” to “knowing themselves perfectly.”
What you see. Even when the self-assessment is chosen at random, you still get that very “effect”: the bottom group greatly overestimates itself, and the top group underestimates. The further you move the slider toward “know themselves,” the closer the lines get to each other.
What this means. The simulation does not prove that the effect does not exist. It shows why one such graph is not enough to claim that weak participants have a special “blindness.” That is why researchers argue: part of the picture may be real, and part may be statistics.
Where it is useful to know this
A new field. You watched a few videos about electrical repair, and the job looks simple. In a new subject we have little information about what we do not know. This does not mean you have “reached Mount Stupid.” It means the feeling of confidence in a new subject is a poor reflection of your real level, and it is worth checking against something external.
Work. Rating your skills in a job interview, estimating your own deadlines, deciding whether you need a specialist. Where there is no quick feedback, it is easy to misjudge yourself in both directions: to overestimate and to underestimate.
Arguments. Online, the Dunning–Kruger effect has long since become a label: it is a convenient way to explain why the person you are arguing with is confident and wrong. The irony is that the popular version of the effect is itself a simplification that few people have checked. A label says nothing about a particular person. Confirmation bias works in a similar way: it is much easier to notice other people’s biases than your own.
How to judge your knowledge more accurately
The Dunning–Kruger effect is one of the biases on the cognitive bias map. There it is marked as a bias whose interpretation is still debated. Another bias that has held up well under testing is the anchoring effect.
Frequently asked questions
Is it “Dunning-Kruger” or “Dunning–Kruger”?
Both versions appear. Strictly speaking, a dash joins the surnames of the two authors: “the Dunning–Kruger effect.” Online, a plain hyphen is more common.
So has the Dunning–Kruger effect been disproved?
No. What has been disproved is mostly the popular picture. Overestimation among people with low results is observed in different studies. The argument is about how much of it is a feature of thinking and how much is a result of statistics.
Is it the same as impostor syndrome?
No. Impostor syndrome is a persistent feeling that your successes are undeserved. In the 1999 study, strong participants underestimated themselves slightly and for a different reason: they assumed the task had been easy for everyone else too.
How can I tell whether I am overestimating myself?
From the inside, by feel, almost never. The only thing that helps is comparison with an external yardstick: a result, a test, feedback. If you regularly write down your estimate before the result and compare afterward, the pattern becomes visible after a while.
