Psychology says the least skilled people are the most confident in their own competence, precisely because the skills needed to do a task well are the same ones needed to know you're doing it badly.

The finding is one of the most frequently cited in modern psychology, and one of the most consistently misunderstood in popular discussion. It comes from a 1999 paper by two Cornell psychologists whose central claim was as simple as it was uncomfortable. The people worst at any given task, on their data, were also the people most confident they were good at it. And the reason for the pattern was not what most readers would guess.

It was not that the least skilled were stupid. It was not that they were arrogant. It was not that they were deluded about themselves in some general way that also affected their work. The pattern the researchers found was much more specific. The mental machinery you use to perform a task competently is, on their argument, the same mental machinery you would need to accurately notice when you are performing it badly. Which means that if you lack the machinery, you are automatically deprived of both. You cannot do the task well. And you cannot see that you cannot do it well. The two failures are not two separate problems. They are one problem with two visible faces.

What Kruger and Dunning actually measured

According to the original 1999 paper by Justin Kruger and David Dunning, published in the Journal of Personality and Social Psychology under the title “Unskilled and Unaware of It: How Difficulties in Recognizing One’s Own Incompetence Lead to Inflated Self-Assessments,” and hosted in full text by the University of Michigan’s Social Sciences and Society Initiative, the researchers ran four separate experiments testing participants on three domains. Sense of humour. Grammar. Logical reasoning. In each case they gave the participants the objective test, scored it, and separately asked each participant to estimate how they had done relative to everyone else.

The results were consistent across all three domains. Participants in the bottom quartile, meaning those who had actually scored in roughly the twelfth percentile of the sample, estimated that they had performed at roughly the sixty-second percentile. They believed themselves to be better than two-thirds of the people around them. They were, in fact, worse than seven-eighths. The gap between where they were and where they thought they were was enormous, and it was systematic.

Participants at the top of the distribution showed the opposite pattern. Top scorers estimated themselves as slightly worse than they actually were. They assumed the tasks had been easy for others too. Which turned out, on the data, to be wrong.

The middle of the distribution, on either finding, was largely accurate. The failure of self-assessment lived at the two extremes.

Kruger and Dunning tested the mechanism directly in their fourth study. They took a group of low performers on a logical reasoning task and gave them a short training session in the underlying skill. After the training, the same participants were asked to re-estimate their performance on the original test they had already taken. Their self-assessments came down. As their competence at the skill improved, their capacity to see how badly they had actually done improved along with it. The same improvement did not appear in a control group given a different, unrelated activity between the two assessments. Which supported the specific mechanism the paper had proposed. Metacognitive accuracy tracks skill. You have to have some of the underlying skill to see that you did not have enough of it.

The same mismatch plays out in adulthood itself, where the people most sure they have it figured out are often the ones improvising hardest, which is exactly what this video explores:

Why the finding has held up, and where it has been challenged

According to a 2008 follow-up paper by Joyce Ehrlinger, Kerri Johnson, Matthew Banner, David Dunning and Justin Kruger, published in Organizational Behavior and Human Decision Processes under the title “Why the Unskilled Are Unaware: Further Explorations of (Absent) Self-Insight Among the Incompetent”, the same team ran five additional studies attempting to address the criticisms that had built up around the original finding. Some critics had argued that the effect was a statistical artefact produced by regression to the mean. Some had argued that low performers simply had less information about their performance rather than being cognitively blind to it. Some had argued that offering people money to be accurate would eliminate the miscalibration.

The Ehrlinger team tested each of these alternative explanations directly. Real-world settings, meaning student self-assessments before and after actual exams, showed the same pattern as laboratory settings. Financial incentives for accuracy reduced the effect only slightly. Low performers who were shown the answers of high performers still failed to recognise their own errors. On the accumulated evidence of the follow-up studies, the metacognitive interpretation held up. The unskilled remained genuinely unaware in a way that pointing better information at them did not fix.

The story does not end there. Over the following two decades, several independent research groups have argued that the specific quantitative pattern Kruger and Dunning identified can, on some of their data, be partly reproduced by simple statistical models that do not require any metacognitive deficit at all. Regression to the mean and the mathematical constraints of self-assessment ranges can generate a curve that looks like Kruger-Dunning without any of the psychology. This is a genuine and unresolved debate in the field. The 1999 paper is one of the most cited in psychology. It is also one of the most-contested.

The most careful recent evidence, on balance, suggests that both explanations contribute. Some of the effect is statistical. Some of it is genuinely psychological. A 2024 study of introductory STEM students at Purdue and the University of Illinois, published in Frontiers in Education, tested this directly by tracking how students updated their metacognitive predictions after actually taking an exam. Low-performing students were both more overconfident before the exam and less able to correct their self-assessment afterwards, even with the direct feedback of the exam result in front of them. The pattern, on that evidence, is not simply a mathematical artefact. Something psychological is going on. It is just not always as clean as the popular retellings suggest.

What the whole body of research still supports, on the peer-reviewed record, is the core observation. People who are bad at a task are, on average, meaningfully worse at recognising their own performance than people who are good at it. The reasons are contested. The observation itself is not. And the specific implication is worth sitting with. It is not, from the inside, possible to reliably tell whether you are in the first group or the second. The same machinery that would let you know is the machinery you either have or do not have. If you have it, you know. If you do not have it, you also do not know that you do not have it.

Which is the specific reason the finding was uncomfortable when it was published, and remains uncomfortable now. There is no obvious way, from inside your own head, to check.

Kiran Athar is a writer, not a psychologist or a cognitive scientist. This piece draws on peer-reviewed research in the Journal of Personality and Social Psychology, Organizational Behavior and Human Decision Processes, and Frontiers in Education.