What Is the Dunning-Kruger Effect? What the Research Actually Shows

Are incompetent people really more confident? Discover the scientific truth behind the Dunning-Kruger effect. Read the full guide and try the RIOT IQ test!

Dr. Russell T. WarneChief Scientist

What Is the Dunning-Kruger Effect? What the Research Actually Shows

Few psychology findings have made the leap from academic paper to cultural shorthand as completely as the Dunning-Kruger effect. Named after psychologists David Dunning and Justin Kruger, it has become the go-to explanation for why confidently wrong people exist — invoked in political commentary, workplace discussions, and social media arguments with remarkable frequency. The trouble is that what most people understand the Dunning-Kruger effect to mean is a simplified version of the original finding, and the scientific status of even that original finding has become considerably more complicated over the past decade than popular coverage suggests.

This article explains what Dunning and Kruger actually found, what the effect does and doesn't have to do with IQ, what the replication debate has settled and what it hasn't, and what the most recent research says about how people estimate their own cognitive ability.

What the Original 1999 Study Found

Kruger and Dunning's original paper, published in the Journal of Personality and Social Psychology in 1999 and titled "Unskilled and Unaware of It," tested participants on logical reasoning, grammar, and sense of humor. The central finding was this: those who performed in the bottom quartile rated their skills far above average. People scoring at the 12th percentile estimated themselves to be performing at the 62nd percentile on average — a gap of 50 percentile points between actual and perceived performance.

The proposed mechanism was metacognitive: the same skills needed to perform well on a task are the skills needed to recognize whether you're performing well. A person who lacks competence in logical reasoning also lacks the reasoning apparatus to accurately evaluate their own reasoning. Incompetent individuals tend to overestimate their own skill levels, fail to recognize genuine skill in others, and fail to recognize the extent of their inadequacy — but importantly, once trained to improve their skills, they can recognize and acknowledge their previous lack of competence. This last point is crucial: the effect isn't permanent. Competence builds metacognition.

The study was accompanied by a symmetric but less-discussed finding about high performers: highly expert people underrate their skills socially because they overestimate the knowledge level of their peers. If experts assume their competence is more common than it is, they will underestimate how they stand out relative to others. The popular version of the Dunning-Kruger effect focuses almost entirely on incompetent overconfidence, but the original paper gave equal weight to expert underestimation.

What It Has to Do With IQ

The relationship between the Dunning-Kruger effect and IQ is more specific — and more interesting — than most popular treatments acknowledge.

The effect is not a claim that less intelligent people in general are overconfident in general. The Dunning-Kruger effect is sometimes misunderstood as claiming that people with low intelligence are generally overconfident, instead of denoting specific overconfidence of people unskilled at particular areas. The distinction matters considerably: the original finding is domain-specific, not a blanket statement about intelligence and self-awareness.

That said, a growing body of research has applied the Dunning-Kruger framework specifically to intelligence self-estimation — asking how well people can judge their own IQ relative to their actual measured ability — and the findings are nuanced. One study found that people overestimated themselves by as much as 30 IQ points on average — a striking finding that reflects both genuine miscalibration and the statistical complexity of how self-estimates map onto psychometric scores.

A 2022 preregistered study by Hofer and colleagues at the University of Graz tested Dunning-Kruger effects specifically for self-estimates of general, verbal, numerical, and spatial intelligence in 281 participants. Standard analyses indicated Dunning-Kruger effects in general, verbal, and spatial intelligence — but improved statistical methods only yielded support for a Dunning-Kruger effect in verbal intelligence specifically: people with lower verbal intelligence tended to have less accurate self-knowledge about it. For other intelligence domains, the effect was weaker or absent once statistical artifacts were accounted for.

This domain-specificity is consistent with the broader Dunning-Kruger literature. Verbal ability is an area where people receive relatively frequent social feedback — through conversation, education, and writing — and yet still show systematic miscalibration among low performers. Spatial intelligence and numerical reasoning, which receive less social feedback, show less consistent patterns of self-estimation error. The relationship between ability and self-estimate accuracy appears to be partly driven by how much daily-life feedback a domain generates.

The Replication Controversy

The Dunning-Kruger effect is one of the most frequently cited findings in all of psychology, and it is also one of the most actively contested in recent methodological literature. Understanding what the controversy is actually about is important for reading the effect correctly.

The primary methodological critique targets a statistical artifact called regression to the mean. The concern is this: in the original design, actual performance scores were used both to rank participants and to determine whether their self-assessments were accurate. This "double-dipping" — using the same variable for two purposes — can artificially generate the appearance of overconfidence among low scorers even when no genuine metacognitive bias exists, because measurement error in a test score will produce apparent miscalibration as a statistical artifact. McIntosh and Della Sala argue that self-estimates are often uncertain and only weakly related to actual performance, leaving ample room for regression effects to skew results and potentially create the illusion of overconfidence among low performers.

Dunning and Kruger have responded directly to this critique, noting that the effect survives even when regression to the mean effects are tested directly across several studies with reduced noise. In 2021, Jensen and colleagues conducted studies with over 3,500 respondents each and used item-response theory models to test which computational model best anticipated the pattern of misjudgment — the model that worked best included the assumption that poor performers were genuinely exhibiting systematic metacognitive error, not just statistical noise.

A 2025 reanalysis using composite measures of cognitive ability from longitudinal household survey data confirmed the overestimation pattern among low-ability individuals but attributed much of the effect to regression to the mean and measurement error, reducing the unique metacognitive component. The current state of the literature is not that the Dunning-Kruger effect is "fake" — the overconfidence pattern among low performers exists and replicates. The debate is about how much of it reflects genuine metacognitive deficit versus statistical artifact, and the honest answer is that both contribute, in proportions that vary by study design, domain, and population.

What Recent Research Adds

Two recent developments extend the Dunning-Kruger framework in directions that weren't anticipated by the original 1999 paper.

The first is the medical education literature. A 2024 study of first-semester medical students found that 35.5% of students overestimated their academic performance, with a significant negative correlation between self-assessment accuracy and actual scores (ρ = -0.590). Students who performed worst were most likely to overestimate — the classic pattern. The same study noted that poor performers overestimated their results significantly less when they were rewarded for the accuracy of their self-assessment, suggesting that the miscalibration is not fixed and responds to incentive structures. This has direct policy implications for high-stakes training environments: mandatory calibration exercises, 360-degree feedback systems, and metacognitive training programs have all shown measurable effects in reducing overconfidence among novices.

The second is an unexpected AI-related reversal. A 2025 study found that use of large language models like ChatGPT leads users to consistently overestimate their cognitive performance regardless of actual ability or AI literacy — and identified a reversal of the typical Dunning-Kruger pattern: higher AI literacy was linked to greater overconfidence, not less. This is the opposite of the usual effect, where higher ability is associated with more accurate self-assessment. The mechanism proposed is that AI tool use reduces the metacognitive reflection that normally occurs during effortful cognitive work — when AI handles the work, the user doesn't generate the internal feedback that calibrates self-assessment. The implication for how we think about metacognition in an AI-assisted environment is substantial and unresolved.

What the Effect Says About Metacognition

Whether or not the Dunning-Kruger effect is as large and as purely metacognitive as the original paper suggested, the underlying phenomenon it points toward is real and well-supported independently of any single study: people's self-assessments of their own cognitive abilities are only moderately accurate.

Self-estimates of intelligence and other abilities correlate only moderately with corresponding objective performance criteria. The overall pattern — that people overestimate their ability on average, with the most severe overestimation among the lowest performers — is consistent across dozens of studies, even if the magnitude varies considerably depending on domain, methodology, and population. The practical significance of this is direct: people who have never been measured on a specific cognitive ability will tend to generate a self-estimate that is biased upward, particularly if they are genuinely below average in that domain.

This is one of the strongest psychometric arguments for objective measurement. Not because self-knowledge is worthless — correlational accuracy between self-estimates and measured ability is generally positive, meaning people are not completely in the dark about their own abilities — but because self-knowledge is systematically biased in ways that objective measurement corrects. People who self-assessed their intelligence on average rated themselves five IQ points lower after completing an intelligence test, compared to those who self-assessed before taking the test — meaning exposure to actual performance data recalibrates self-assessment downward toward accuracy.

The Takeaway

The Dunning-Kruger effect describes a real and replicated phenomenon: low performers consistently overestimate their relative standing, in part because the metacognitive skills needed for accurate self-assessment are the same skills the domain demands and that low performers lack. The effect is domain-specific, not a global statement about intelligence and self-awareness. Its magnitude is actively debated in the methodological literature, with regression to the mean accounting for some but not all of the observed pattern. The most recent research shows it is sensitive to incentives, training, and — in a concerning new direction — AI tool use may be producing a novel form of overconfidence that reverses the usual ability-accuracy relationship.

What the Dunning-Kruger literature most clearly establishes is something I consider a compelling argument for objective cognitive assessment: self-knowledge about cognitive ability is systematically biased, and that bias is strongest precisely where it is most consequential — among individuals whose ability falls below what they believe it to be. An objective measurement doesn't just satisfy curiosity. It corrects a bias that most people are carrying without knowing it.

If you want to replace a self-estimate with an actual measurement — and see your cognitive profile across multiple domains with a known margin of error — the RIOT is built to give you exactly that.

References

Psychology Today. (2026). Dunning-Kruger Effect. https://www.psychologytoday.com/us/basics/dunning-kruger-effect
Britannica. (2026). Dunning-Kruger Effect — Definition, Examples & Facts. https://www.britannica.com/science/Dunning-Kruger-effect
Wikipedia. Dunning-Kruger Effect. https://en.wikipedia.org/wiki/Dunning%E2%80%93Kruger_effect
British Psychological Society. (2025). The Dunning-Kruger Effect and Its Discontents. https://www.bps.org.uk/psychologist/dunning-kruger-effect-and-its-discontents
PubMed Central / Journal of Intelligence. (2022). Less-Intelligent and Unaware? Accuracy and Dunning-Kruger Effects for Self-Estimates of Different Aspects of Intelligence — Hofer et al. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8883889/
PubMed Central / BMC Medical Education. (2024). Prevalence of Dunning-Kruger effect in first semester medical students. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11515314/
TechXplore. (2025). AI use makes us overestimate our cognitive performance, study reveals. https://techxplore.com/news/2025-10-ai-overestimate-cognitive-reveals.html
Grokipedia. (2026). Dunning-Kruger Effect — Replication and Statistical Critique. https://grokipedia.com/page/Dunning%E2%80%93Kruger_effect