The Dunning-Kruger Chart Is Drawn Wrong Almost Everywhere

If you have seen a chart labeled 'Dunning-Kruger effect' on social media, it probably showed a cartoon called 'Mount Stupid': a steep peak of overconfidence at 'know nothing', followed by a valley of despair, followed by a slope of enlightenment. This is not what David Dunning and Justin Kruger found, and it appears nowhere in their 1999 paper.

The real graph has two lines. One is actual test performance, broken into quartiles. The other is the participants' estimate of their own performance. Across three studies — humor judgment, English grammar, and logical reasoning — both lines slope up in parallel, but the gap between them shrinks as skill rises. Low performers overestimate themselves more than high performers underestimate themselves. That is the entire finding.

There is no peak, no valley. The cartoon curve began circulating on the internet around 2010, took on a life of its own, and is now more famous than the result.

The paper has also been challenged on statistical grounds. In 2020, Gilles Gignac and Marcin Zajenkowski argued in the journal Intelligence that most of the observed pattern could be reproduced by regression to the mean acting on any two imperfectly correlated measures. That does not make Dunning-Kruger fake. It means the effect size is smaller than popular retellings imply, and the interpretation is more contested than the memes suggest.

Dunning himself has spent twenty-five years correcting misquotations. The sound bite most often attached to him — that the unskilled do not know they are unskilled — was never a sentence in the paper. The finding is quieter and narrower, and the internet did not take the quiet version.