New study overturns Dunning-Kruger, revealing why the overconfident can be right

A new analysis from researchers at the University of Bath and the London School of Economics (LSE) challenges the popular interpretation of the Dunning–Kruger effect—the idea that the least competent people are the most overconfident. For years, studies based on multiple-choice test performance have shown that low scorers tend to overestimate how well they did, while higher scorers make smaller errors.

But the new work, published in Psychological Review, argues that the headline phenomenon may be an artifact created by how test data are statistically conditioned. In many experiments, researchers sort participants by observed test scores and then compare those scores to participants’ predictions. The trouble is that observed scores are noisy: luck, item difficulty, and wording can shift results even when underlying ability is the same.

Using data from thousands of participants across large replication datasets, the team applies improved statistical techniques designed to account for randomness in both true performance and self-estimated performance. Rather than treating “being low on the test” as a stable marker of incompetence, the model separates genuine ability from the fluctuations that come from chance.

When those corrections are applied, the pattern reverses. High-ability individuals appear to be the most overconfident, while the naïve version of Dunning–Kruger largely disappears. In other words, the familiar story “stupid people don’t know they’re stupid” may reflect measurement bias more than a universal cognitive deficit.

The researchers also outline why overconfidence persists even after statistical clarification. Their results suggest overconfidence can function as a signal for hidden ability—something that people with greater capability have more incentive to communicate. Another contributor involves loss-related preferences, which can shape how people interpret and act on uncertain feedback.

“Our research demonstrates that the original methodology fell victim to a statistical illusion,” the authors write, emphasizing that the apparent psychology may be driven by hidden analytical traps. Correct the trap, and overconfidence becomes a more general trait rather than a hallmark of low skill.

Taken together, the findings reshape how researchers should interpret decades of confidence-versus-performance data. They also raise a caution for public understanding: some “viral” effects may be partly statistical mirages.

In the paper, Talking the Talk, Not Walking the Walk: The Coevolution of Overconfidence and Loss Aversion, the authors provide a framework that treats test scores and beliefs as variables jointly shaped by uncertainty—turning a widely repeated claim into a more nuanced, and potentially more interesting, picture of human judgment.