What is the Dunning-Kruger effect, and does it hold up?

The Dunning-Kruger effect is the pattern that people with the least skill in a task tend to overestimate their performance the most. The pattern itself shows up again in newer data. What is contested is why it appears. The original explanation, that incompetence removes the ability to notice incompetence, has been challenged by researchers who point to statistical artifacts, rational estimation and direct tests of metacognition. [1] [2]

Kruger and Dunning reported four studies in which participants took tests of humor, grammar and logic. Those scoring in the bottom quartile grossly overestimated their performance: their scores put them around the 12th percentile, but they estimated themselves to be in the 62nd. The authors proposed a dual burden, in which unskilled people reach poor conclusions and also lack the metacognitive ability to recognize this. They also reported that training participants, which raised their skill, helped them see the limits of their earlier abilities. [2]

A pre-registered Registered Report using a matrix reasoning task found the basic relationship again. Estimation error was negatively related to skill, with correlations of about -0.57 for relative estimates and -0.43 for absolute estimates. So the descriptive pattern is not simply an accident of one old dataset. [1]

Dunning and Kruger's position is the dual-burden account. Their evidence was the link between poor performance and weak metacognitive skill, and the improvement in self-assessment after training. [2]

A second position holds that the effect is largely a statistical artifact. One paper derives it from a simple statistical model with boundary constraints on scores, with no psychological explanation needed. The authors of the Clearer Thinking analysis add that measurement error and floor and ceiling effects can create a fake effect, and that few study designs can rule this out. [3] [4]

A third position says a real effect may exist but reflect rational behavior. In their simulations, people who assume they are average when they lack evidence, then adjust to the evidence they have, produce data matching the classic pattern. The same authors say they cannot rule out the original explanation. [4]

A fourth line of evidence comes from direct measurement. The Registered Report found that poor performers had less information to judge their answers, but their metacognitive efficiency was no worse, and they were less confident than good performers, not more. The authors concluded that the dual-burden account is refuted and that the classic effect is a regression artifact, though compatible with a noise-plus-bias account. The search results do not show how many specialists hold each view, but the critical work is recent and published in peer-reviewed or well-documented form. [1]

One study by Nuhfer's team had students rate each answer on a 25-question science literacy test as "nailed it," "not sure" or "no idea." Among students in the bottom quarter, only 16.5 percent significantly overestimated their ability and 3.9 percent significantly underestimated it, so nearly 80 percent judged themselves fairly well. The article describing this also says most people think they are better than average, a separate bias. [5]