The Dunning-Kruger Curve You Can Draw From Noise

Edward Nuhfer's group ran a quiet experiment in 2016: they generated columns of random numbers, treated them as quiz scores and self-ratings, and applied the analysis Justin Kruger and David Dunning used in 1999. The famous chart came out anyway. Bottom-quartile "performers" looked overconfident, top-quartile ones humble. Only noise was being measured.

The issue is the slicing. Kruger and Dunning sorted people by their actual scores, then plotted self-assessment against that ranking. When two variables aren't perfectly correlated, the extremes of one will always look closer to the middle on the other. Statisticians call it regression to the mean. It happens whether anyone is metacognitively confused.

Gignac and Zajenkowski tested this directly in 2020 using nonlinear regression on IQ data, the kind of valid test the original design didn't include. The asymmetric "unskilled and unaware" pattern shrank to a modest, roughly linear miscalibration. Most people overrate themselves a bit. Low performers overrate themselves more, but not in the dramatic, qualitatively-different way the meme implies.

Then the floor and ceiling. Jan Magnus and Anatoly Peresetsky, in 2022, fit a censored tobit model to 665 economics students at Moscow's Higher School of Economics. A student who scored 5 cannot rate themselves below 0, so noise skews up; a 95-scorer faces the opposite squeeze. The model fit the data almost perfectly without any cognitive deficit. "There is an effect," the authors wrote, "but it does not reflect human nature."

Miscalibration is real. The better-than-average effect is real. What the critics keep showing is that the specific shape everyone learned, where incompetence breeds confidence in a way competence doesn't breed humility, is mostly an artifact of how the data was sliced. The lesson the meme teaches, that other people are too dumb to know they're dumb, may be the cleanest example of the bias it claims to describe.