Dunning-Kruger Effect: Why the Least Skilled Overrate Themselves Most

The pattern is a gap between skill and self-assessment; the explanation Kruger and Dunning proposed for it is that judging your own work takes the same skill as doing the work. A beginner at chess does not see the move that would have won, so the game feels closer than it was. A new writer does not notice the sentence that lost the reader, so the draft feels finished. Kruger and Dunning called this a double burden: the gaps in skill that produce a poor result are the same gaps that hide the result from the person who produced it.

On that account the effect is not about intelligence, and it is not a personality type. It is a property of feedback. In domains with fast, unambiguous feedback (a runner's stopwatch, a typing speed counter) self-estimates line up with performance. In domains where the feedback is slow, social or absent (judgment, humor, reasoning, management) the estimates drift upward, and they drift most for the people with the least to go on.

Three things follow from this. First, the bottom performers in the 1999 studies did not simply guess high; they were also worse at recognizing competence in others, grading other people's tests less accurately than the top performers did. Second, the top performers underestimated themselves, mainly because they assumed the task was as easy for everyone as it was for them. Third, in the fourth study, a short lesson in the logic the test required brought the lowest scorers' self-estimates down and made them better at spotting their own wrong answers.

How much of the pattern reflects a real deficit in self-insight is debated. On any single test, the lowest scorers will tend to have estimated higher than they scored, because measurement error and regression to the mean push in that direction, and Gignac and Zajenkowski argued in 2020 that a model with no metacognitive deficit at all reproduces most of the classic chart. Ehrlinger, Dunning and colleagues concluded in 2008, from studies that corrected for test error and paid for accurate self-assessment, that a deficit in self-insight remains. The observed pattern is not in question; what is contested is how much of it statistical artifacts generate or amplify, and how much is a genuine difficulty in judging one's own work.

Justin Kruger and David Dunning published four studies in 1999 in the Journal of Personality and Social Psychology under the title "Unskilled and unaware of it." Cornell undergraduates took tests of humor, logical reasoning and grammar, then estimated their own ability and their test score as a percentile among their peers.

In the logical reasoning study, participants answered 20 items drawn from a Law School Admission Test preparation guide. The quarter with the lowest scores sat, on average, at the 12th percentile of actual performance but placed their reasoning ability at the 68th percentile and their test score at the 62nd. The quarter with the highest scores sat at the 86th percentile and placed themselves at the 74th, a small underestimate. Across the four studies the bottom quartile overestimated their percentile by roughly 50 points.

The fourth study added a training condition. Half of the participants worked through a short lesson on logical reasoning after the test, then re-estimated their earlier performance. Those who had scored lowest and received training revised their self-estimates down and became better at telling which of their own answers were wrong. The lowest scorers who did not receive the training showed no such correction; their estimates of their own raw score moved up rather than down.

The chart most people recognize, a curve that rises to a peak of confidence and then falls into a valley before climbing again, does not appear in the paper. The original figures plot four quartiles of actual performance against perceived performance, and the line through them slopes gently upward. The chart in the demo above shows the two end quartiles from the logic test, not the whole figure.

The pattern shows up wherever a person has just enough knowledge to act and not enough to check the result. The rows below are illustrations, not findings from the studies cited on this page.

In the 1999 training study, a short lesson in the skill the test required brought the lowest scorers' self-estimates closer to their results, so the first counter is to learn the domain well enough to judge it. In the meantime a few habits make the gap easier to see.

• Get feedback that does not depend on your own judgment: a score, a time, a reviewer, a test with an answer key. The demo on this page works because the quiz grades itself.
• Before a task, write down your predicted result. Afterward, compare. A written prediction cannot be quietly revised once the outcome is known.
• Ask an expert what beginners typically get wrong. The errors you cannot see are usually the errors everyone at your level makes.
• Treat fluency as a warning sign. The feeling that a subject is simple arrives before competence, not after it.
• After learning the basics, go back over your own answers. In the 1999 training study, the lowest scorers who worked through a short lesson became better at telling which of their own answers were wrong; the study did not test whether grading other people improves with training.

None of these require humility as a character trait. They are procedures rather than attitudes, and they are open to the confident and the anxious alike. People who underestimate themselves need the same external feedback, in the other direction.

The Dunning-Kruger effect is often used as a label for anyone confident and wrong, which stretches it past what the research supports. The effect describes a specific relationship between skill and self-assessment in one domain at one time. It does not say that confident people are usually incompetent, and it does not say that doubt is evidence of skill.

• It is not the overconfidence effect, which is the general tendency for stated confidence to exceed accuracy at every skill level. Dunning-Kruger is about how that gap changes with skill.
• It is not illusory superiority, the finding that most people rate themselves above average on desirable traits. Illusory superiority is a population average; Dunning-Kruger describes who contributes most to it.
• It is not the illusion of validity, the confidence that comes from a coherent story about a prediction regardless of its accuracy. That one does not depend on skill at all.
• It is not imposter syndrome (see the Imposter Syndrome Test), which runs the other way: a capable person who cannot internalize evidence of their competence. The top performers in the original study did underestimate their percentile, but Kruger and Dunning attributed that to assuming the test was as easy for others as for them; the study did not measure imposter feelings.