You’ve probably come across the idea, possibly even as a meme. The person who is bad at something doesn’t know they’re bad at it. The person who is good at something doesn’t know they’re good at it. Confidence is inversely correlated with competence. Put a graph of this in front of most people, and they’ll nod in recognition. It matches something they feel they’ve seen in offices, relationships, public life, and themselves.
The idea is usually called the Dunning-Kruger effect, and it comes from a 1999 paper by two psychologists at Cornell, David Dunning and Justin Kruger. The paper has been cited over ten thousand times. It’s been referenced in countless articles about overconfident politicians, dismissive colleagues and clueless bosses. It’s also, in the pop-psychology version that dominates the internet, substantially wrong — or at least not what Dunning and Kruger actually found. The real picture is subtler, more interesting, and worth understanding if you want to reason honestly about your own confidence in your abilities.
What Dunning and Kruger actually found
The original 1999 study had four experiments. Participants were tested on their skill at grammar, logic, or humour (rated against a panel of professional comedians), and asked to rate their own performance.
The core finding was that people in the bottom quartile — those who performed worst on the actual tests — tended to rate their performance around the 60th percentile. That is, they thought they were above average, when in fact they were well below it. People in the top quartile, meanwhile, tended to rate their performance around the 70th or 75th percentile — that is, above average, which they were, but below their actual ranking.
Dunning and Kruger’s interpretation was elegant. To accurately assess your skill in a domain, you need a certain amount of skill in that domain. The same knowledge that lets you do a task well also lets you recognise when you or others are doing it poorly. People with very low skill, therefore, face a double curse: they do the task poorly, and they lack the meta-cognitive knowledge that would let them notice they’re doing it poorly. They think they’re fine because they don’t have the tools to see that they’re not.
This is the real claim of the original paper. It’s carefully argued, and it’s narrower than the internet version.
What the internet version gets wrong
The popular version of Dunning-Kruger is usually rendered as a specific graph — a line that rises steeply from low to moderate competence (confidence high at zero knowledge, dropping sharply as you learn enough to realise you know nothing, then rising again as you become expert). This graph appears in countless presentations, articles and memes. It has a name: “the Dunning-Kruger curve”.
Here’s the thing. Dunning and Kruger never drew this graph. It does not appear in their paper. Their actual findings are much simpler: in the bottom quartile, people overestimate their ability; in the top quartile, they slightly underestimate. The rest of the “curve” is an artistic addition by popularisers, not a finding.
The internet meme also usually implies something Dunning-Kruger doesn’t say: that incompetent people feel more confident than competent people. This isn’t true. In the original data, incompetent people rated their performance at around the 60th percentile. Competent people rated theirs at around the 75th. The competent are, in absolute terms, more confident than the incompetent. What’s asymmetric is the gap between their self-rating and their actual ranking — the incompetent are far from accurate, the competent are close to it.
The popular meme has flipped a real finding into a punchier and somewhat inaccurate story.
The statistical critique
A more serious critique has emerged in the last decade, led by the statisticians Gilles Gignac and Marcin Zajenkowski, and extended by Patrick McIntosh and colleagues in a 2019 paper. Their argument, put simply: most of what Dunning and Kruger observed can be explained by a statistical phenomenon called regression to the mean, plus the natural tendency for most people to rate themselves as slightly above average.
Here’s why. If you take any two imperfectly-correlated measures — say, test performance and self-rating — and plot one against the other, you get a specific pattern. People at the extremes of one measure tend to be less extreme on the other, just because of noise in both. The worst performers on a test, by chance, weren’t all uniformly bad; some of them got unlucky questions. The best performers, by chance, weren’t all uniformly excellent; some got lucky. When both groups self-rate, they’ll tend to rate themselves somewhere closer to the middle than their actual performance — which, on the graph, looks exactly like the Dunning-Kruger effect.
Combine this with the above-average effect documented decades earlier by the Swedish psychologist Ola Svenson — who famously found that 93 per cent of American drivers and 69 per cent of Swedish drivers rated themselves in the top 50 per cent for driving skill — and you can generate something that looks like Dunning-Kruger without anyone actually being systematically overconfident in the specific way Dunning and Kruger claimed.
The critics’ conclusion: some of what Dunning-Kruger describes is real, but a significant portion of the apparent effect is statistical artefact that would show up even if everyone had perfect self-insight about how variable their performance was.
Where the effect still seems to survive
That said, the critics don’t go so far as to say there’s nothing there. The more careful recent work suggests that a genuine effect exists at the extremes of skill and in specific domains. In some domains — particularly those requiring complex judgement, where the person has to recognise good reasoning to produce it — people with very low skill do appear to have genuinely distorted self-assessments beyond what regression to the mean can explain.
Perhaps the clearest example is in driving. Decades of research have found that among drivers who fail driving tests, a strikingly high proportion continue to rate themselves as above-average drivers even after the failure. Unlike many test performances, failing a driving test should give you clear information about your competence. And yet, most failing drivers don’t update. Here, the “I don’t know what I don’t know” story has real force.
A second example is in writing. People with poor grammatical skills often rate their writing as fine — partly because evaluating writing requires the same skills that producing it does. A person who can’t recognise a dangling modifier in someone else’s writing is unlikely to notice it in their own.
So the honest picture is this: there’s a real effect, it shows up most clearly in specific domains where competence and meta-cognition draw on similar skills, and it’s smaller and more context-dependent than the viral meme suggests. It’s not a universal law of human psychology. It’s a tendency that appears in certain conditions.
What changes if you take the careful version
If you took away from “Dunning-Kruger” the idea that everyone around you who disagrees with you is probably too incompetent to know they’re wrong, you’ve taken away the wrong thing. The real lesson, calibrated to the actual research, is humbler and harder.
You can’t trust your own confidence. Neither can the people around you trust theirs. Both you and they are assessing your skills using the same tools you’re trying to assess. In domains where you’re genuinely competent, you probably slightly underestimate yourself. In domains where you’re incompetent, you probably significantly overestimate yourself. The second error is larger than the first.
This has a practical corollary that most of the pop-psychology treatments miss. The solution isn’t to become more humble about everything — that just produces paralysed people who can’t act. The solution is to develop mechanisms for checking your self-assessments against external feedback: test results, performance reviews, honest friends, comparison to people who do the task for a living. The feedback loops that correct for the limits of your own self-perception.
In domains where such feedback is available — driving, writing, playing an instrument, cooking, anything with clear external standards — you have the materials to improve. In domains where feedback is scarce — many interpersonal skills, much of management, most of parenting, significant parts of professional life — you’re running somewhat blind. The honest admission that you can’t trust your own self-assessment in such domains is probably more useful than any specific insight the original research provides.
The broader point about self-assessment
Related research from Justin Kruger himself (in later work) and from David Dunning’s ongoing programme has extended the original finding in a direction worth mentioning. In a series of studies, they found that people’s ability to identify good performance in others is limited by their own performance. Bad cooks can’t tell the difference between mediocre and excellent restaurants. Bad writers can’t recognise great writing. Bad chess players can’t tell which chess moves are brilliant. This is a much less catchy finding than the viral Dunning-Kruger story, but it’s probably more important for ordinary life.
It means that the voices around you who rate your work as good or poor are filtered through their own competence. A friend who doesn’t write well telling you your essay is great is giving you information about themselves as well as about your essay. A boss who doesn’t manage well telling you your work is excellent is giving you a mixed signal. Calibrating feedback against the skill of the person giving it turns out to be an unusual but important meta-skill.
The question that remains
The story of Dunning-Kruger is, in its own way, a miniature case study in what happens to scientific findings when they enter culture. A careful, bounded finding gets compressed into a punchy narrative. The narrative goes viral. The narrative gets drawn as a graph the researchers never drew. Millions of people now believe something that the original study never quite said.
The useful thing to take away isn’t the meme, and isn’t the pure critique either. It’s the quieter, more honest version: your self-assessment in any domain is probably slightly off, it’s more off in domains where you’re less skilled, and the fix is external feedback rather than introspection.
The question to carry is a harder one than the viral story asks. It isn’t are other people around me lacking in self-awareness? — which is easy and usually self-serving. It’s:
Which of your confidences are anchored in external checks you actually trust, and which are ones you’ve never really tested?
Key research referenced: David Dunning and Justin Kruger’s original 1999 paper (Journal of Personality and Social Psychology); Gilles Gignac and Marcin Zajenkowski’s statistical critique (2020); Patrick McIntosh and colleagues’ 2019 paper extending the critique; Ola Svenson’s above-average driver research (1981); Dunning’s ongoing research programme on self-assessment.