Summary: New research finds that the widely cited Dunning-Kruger effect is largely a statistical artifact. After correcting for measurement noise and regression to the mean in large replication datasets, the apparent pattern reverses: overconfidence is common across ability levels and tends to be strongest among the highest performers.
Applying improved statistical modeling to extensive replication datasets, the research team accounted for volatility in test scores and randomness in self-assessments. Once those statistical distortions—introduced by sorting people strictly by observed test scores—are removed, the classic Dunning-Kruger pattern disappears. Instead, overconfidence appears as a general human tendency that increases with actual ability, suggesting that the original pattern was produced by mathematical bias rather than a unique psychological deficit among low performers.
Key Facts
- Statistical artifact identified: The familiar Dunning-Kruger curve—where lower performers seem to overestimate their abilities most—arises from the interaction of measurement noise, luck, and uncorrected regression to the mean when participants are grouped by raw test scores.
- Trend inversion after correction: When statistical models explicitly model randomness in both test outcomes and self-predictions, the data show that higher-ability individuals display the greatest overconfidence.
- Volatility distorts estimates: Low test scores often reflect unlucky draws or ambiguous items; sorting on these scores inflates estimated overconfidence for those individuals while producing an opposite bias for high scorers.
- Overconfidence is widespread: Overconfidence appears across ability levels rather than being confined to low performers, and its magnitude tends to scale positively with true competence after proper corrections.
- Signaling as a possible explanation: The authors argue that overconfidence can serve as a costly signal of underlying ability. Because high-ability people have more to gain from convincing others they are competent, they have stronger incentives to display confident behavior.
Source: University of Bath
Researchers at the University of Bath and the London School of Economics and Political Science (LSE) report that the conventional interpretation of the Dunning-Kruger effect—namely that the least competent are the most overconfident—may be reversed once statistical artifacts are addressed.
For decades, the Dunning-Kruger effect has been understood to mean that people who perform poorly are unaware of their limitations and therefore overestimate their performance, while high performers assess themselves more accurately. Hundreds of studies have replicated the canonical pattern using simple comparisons of predicted versus actual scores.
Because of that body of work, the Dunning-Kruger label has become a popular shorthand for blaming misguided confidence on ignorance. However, the new study published in Psychological Review argues that much of this pattern is produced by statistical biases introduced in how participants were sorted and analyzed.
When the researchers applied more robust statistical techniques that model randomness in both test performance and self-assessment, the apparent effect flipped: the most capable participants consistently showed the greatest overconfidence. This undermines the simple narrative that low ability causes exaggerated self-assessment and suggests instead that overconfidence is a general human tendency that interacts with incentives and signaling motives.
Professor Chris Dawson of the University of Bath’s School of Management explained that earlier methods were vulnerable to a statistical illusion. Sorting participants by observed test scores without accounting for chance variation produces a spurious pattern in which unlucky low scorers appear unrealistically confident and lucky high scorers appear more modest than they actually are.
Professor David de Meza from LSE’s Department of Management emphasized that test outcomes are inherently noisy. Factors like luck, ambiguous wording, or transient state effects can push an otherwise capable person’s score down, making their self-predicted performance look disproportionately high. The opposite bias affects high performers who may have benefited from fortunate conditions.
To overcome this, the team analyzed thousands of observations across large replication datasets using models that incorporate measurement error and regression to the mean in both objective scores and subjective predictions. Once those sources of noise were properly modeled, the Dunning-Kruger pattern did not survive; overconfidence was shown to be both widespread and positively associated with true ability.
The authors acknowledge that incompetence can sometimes produce poor self-awareness, but they highlight alternative mechanisms for overconfidence. Notably, they propose a signaling theory: because overstating one’s ability can influence others’ beliefs and decisions, high-ability individuals have greater reason to display confidence to signal their competence, making strategic overconfidence adaptive in social contexts.
Key Questions Answered:
A: The original approach grouped participants by observed test scores without accounting for random variation and measurement error. Low scores can result from bad luck or ambiguous items, and this causes regression-to-the-mean artifacts that make self-assessments appear excessively overconfident.
A: Models that control for noise in both test outcomes and predictions show that overconfidence persists across all ability levels, but its measured magnitude increases with true ability—contrary to the classic Dunning-Kruger interpretation.
A: Beyond limitations in self-knowledge, overconfidence can function as a social signal. High-ability individuals gain more from persuading others of their competence, so they have stronger incentives to behave confidently, making strategic overconfidence more common among the able.
Editorial Notes:
- This article was edited by a Neuroscience News editor.
- The journal paper was reviewed in full by editorial staff.
- Additional context and explanation were added by our editors.
About this psychology research news
Author: Lynn Li
Source: University of Bath
Contact: Lynn Li – University of Bath
Image: The image is credited to Neuroscience News
Original Research: Open access.
“Talking the Talk, Not Walking the Walk: The Coevolution of Overconfidence and Loss Aversion” by Chris Dawson and David de Meza. Psychological Review. DOI: 10.1037/rev0000644
Abstract
Talking the Talk, Not Walking the Walk: The Coevolution of Overconfidence and Loss Aversion
Evolutionary theory faces a puzzle in explaining why two apparently opposing behavioral biases—overconfidence, which encourages risk-taking, and loss aversion, which restrains it—coexist. Overconfidence prompts action, while loss aversion limits initiative.
A prominent evolutionary account, proposed by Trivers (1976), suggests self-deception evolved to better deceive others. That account leaves open questions about why sincere signals are trusted and why overconfidence is often tempered by loss aversion. The authors propose a signaling framework in which self-deception can make signals of ability more credible: decision errors caused by high self-belief are less costly for more capable individuals, and the benefits of being perceived as able grow with true ability, making innate overconfidence a potentially reliable indicator of competence.
Under this perspective, loss aversion complements overconfidence in an evolutionary equilibrium. Loss aversion reduces some decision costs associated with overconfidence but remains largely hidden, preserving overconfidence’s signaling value. The paper argues that these biases can be mutually reinforcing: the combined payoff to agents from having both traits may exceed what would be achieved if they were absent. From this integrated viewpoint, simple prescriptions to eliminate both overconfidence and loss aversion may be misguided.