Online, people often say, "People with low skills are the most confident." They call this the Dunning-Kruger effect. In the real study, students took tests and guessed their own scores. Students with low scores thought they did better than they really did.

But they still rated themselves lower than the top students did. Everyone finds it hard to judge their own skills.

The Dunning-Kruger effect is often described as "the less you know, the more confident you are." However, the original 1999 study says something different. Researchers gave college students tests in humor, logical reasoning and grammar, and asked them to guess how they ranked. The lowest-scoring group was actually around the 12th percentile, but they guessed they were around the 62nd. Top students slightly underestimated themselves. Low scorers still rated themselves below top scorers.

Also, the popular curve showing a "peak of stupidity" does not appear in the paper. Some researchers argue that statistics can create this pattern, but a large 2021 study suggested that low performers really are worse at judging whether their answers are correct.

Online, the Dunning-Kruger effect is usually invoked to mock someone's overconfidence. The original 1999 study made a more modest claim. Cornell researchers tested students on humor, logical reasoning and grammar and asked them to estimate their percentile rank. Students in the bottom quarter, who actually averaged around the 12th percentile, placed themselves near the 62nd, while top performers slightly underestimated themselves.

The authors argued that the skills needed to perform well are the same skills needed to recognize one's own errors. Two popular versions get the data wrong. The roller-coaster curve with a "peak of stupidity" never appeared in the paper. And low performers did not rate themselves above top performers.

Their self-assessments rose with ability, only less steeply than their actual scores. Critics have demonstrated that regression to the mean and the better-than-average effect can reproduce much of the pattern even with random numbers, and one reanalysis concluded that the true effect is small.

A 2021 replication pushed back. In each of its two studies, with about 4,000 participants apiece, low performers were less sensitive to whether their answers were correct, as the original interpretation predicted. People at every level of skill misjudge their own competence to some degree.