The Dunning-Kruger Effect, Explained

The Dunning-Kruger effect is the finding that people who score lowest on a test of a skill tend to overestimate their performance the most. In the original 1999 studies, people in the bottom quarter on tests of humor, grammar and logic scored around the 12th percentile on average but guessed they were around the 62nd (Kruger & Dunning, 1999). Kruger and Dunning argued that the skills you need to do well are often the same skills you need to notice you’re doing badly (McIntosh et al., 2022). Later studies have reproduced the basic pattern (Krueger & Mueller, 2002; Jansen et al., 2021). Researchers still disagree about whether it reveals a real blind spot in low performers or comes mostly from the way the data are analyzed (McIntosh et al., 2022).

What Kruger and Dunning found in 1999

Justin Kruger and David Dunning tested undergraduates in psychology courses at Cornell University on humor, logical reasoning and English grammar, then asked them to judge how they had done (Magnus & Peresetsky, 2022). Across four studies, they found:

• The bottom quarter grossly overestimated their scores and ability. Overall they performed around the 12th percentile and placed themselves around the 62nd (Kruger & Dunning, 1999). On the logic test alone, bottom-quarter participants averaged around the 12th percentile and ranked themselves at the 68th (McIntosh et al., 2022).
• The top quarter underestimated where they stood, though less severely (same source).
• Several analyses tied the overestimation to weaker “metacognitive” skill, the ability to tell right answers from wrong ones. The authors called this a dual burden: poor performers make mistakes and also can’t see them (Kruger & Dunning, 1999).
• When the researchers improved participants’ skills, those participants got better at recognizing the limits of their ability (same source).

A 2008 follow-up, with Dunning and Kruger among its authors, reported five more studies. Poor performers still lacked insight in real-world settings and when given incentives to be accurate. A meta-analysis in the paper concluded that their optimistic estimates came from not seeing their own errors, not from misjudging other people (Ehrlinger et al., 2008).

What it does and doesn’t mean

In popular culture, the effect has turned into the idea that stupid people are too stupid to know they’re stupid (McIntosh et al., 2022). The research is narrower than that.

• The 1999 studies measured performance on particular tests in particular domains (Kruger & Dunning, 1999). Being in the “bottom quarter” meant scoring lowest among the other participants on that test, so anyone can land there for some skill.
• Low scorers didn’t think they were the best. A guess of the 62nd percentile is above the middle and well short of the top (Kruger & Dunning, 1999).
• Overrating yourself is common at every level. Kruger and Dunning started from the observation that people tend to hold overly favorable views of their abilities in many areas (same source). People routinely rate themselves above the 50th percentile on average, even though the true average is 50 by definition (McIntosh et al., 2022).

The debate over what causes it

The main criticism is statistical. When the same test score is used both to sort people into groups and to calculate how wrong their estimates were, luck on the test pushes the errors in opposite directions: unlucky test-takers look overconfident and lucky ones look underconfident. This is regression to the mean. Add the general tendency to rate yourself above average, and you get the classic graph (McIntosh et al., 2022).

A 2026 paper gives a clear example from another research group. In a study of 2,043 hospital visitors who took a Covid-19 quiz, people rated their own knowledge close to 7 out of 10 on average whatever their score, and their ratings were unrelated to how much they knew. Subtracting quiz scores from those ratings still produced the classic Dunning-Kruger pattern (McIntosh et al., 2026).

Here is how the debate has unfolded in peer-reviewed journals:

Year	Study	What it found or argued
2002	Krueger & Mueller	Replicated the pattern but found no evidence that metacognitive skill explained it. The lopsided errors disappeared when regression or the better-than-average effect was statistically removed (PubMed)
2002	Kruger & Dunning (reply)	Argued the critics used unreliable tests and unsuitable measures, and that regression plus the better-than-average effect can’t explain their experimental results (PubMed)
2006	Burson, Larrick & Klayman	Across 12 tasks, best and worst performers were about equally inaccurate on moderately hard tasks, and the best were less accurate on harder ones. They proposed a “noise-plus-bias” explanation (PubMed)
2016	Nuhfer et al.	Random-number simulations showed that common ways of graphing self-assessment data, including charts grouped by quartile, create artifacts that invite misinterpretation (Numeracy)
2021	Jansen, Rafferty & Griffiths	A replication with about 4,000 people in each of two studies supported the view that low performers are less able to tell whether their answers are correct in grammar and logical reasoning (PubMed)
2022	Magnus & Peresetsky	A model built on the fact that scores can’t go below 0 or above 100 fit data from 665 students predicting their grade on a statistics exam almost perfectly with no psychological explanation (PMC)
2022	McIntosh et al.	In a preregistered study of 151 people on a reasoning task, poor performers were worse at telling right from wrong answers but were less confident, not more. Their test scores, not their metacognition, drove the pattern (PMC)
2024	Lebuda et al.	In two creativity studies (425 and 317 people), the original quartile analysis showed the effect, but newer statistical tests did not consistently support it (PMC)

Where the evidence stands

The basic pattern is not in doubt: when you compare people’s estimates with their scores, the lowest scorers overestimate the most (Krueger & Mueller, 2002; Lebuda et al., 2024). The cause is still contested.

Both sides agree on one point: low performers are worse at telling which of their answers are right (Jansen et al., 2021; McIntosh et al., 2022). They disagree on whether that is what makes them overestimate. Jansen’s study is far larger and used the original kinds of tests, and it supports Kruger and Dunning’s idea of a metacognitive deficit. McIntosh’s study is newer and preregistered and measured confidence answer by answer, but it used a different task and a much smaller sample, and it found that metacognition didn’t drive the pattern. In a 2026 paper, McIntosh’s group summarized earlier work as showing that analyses which do detect real miscalibration among poor performers find biases that are negligible in practical terms (McIntosh et al., 2026). Most of the newer studies covered here lean toward a largely statistical explanation, but the largest replication still finds some real effect, so the question isn’t settled.

What it means for judging your own skills

A few practical points follow from this research.

• Expect your overall self-rating to be off. Global estimates of ability are noisy for nearly everyone (McIntosh et al., 2022). On hard tasks, even top performers misjudge where they stand (Burson et al., 2006).
• Rate your confidence answer by answer. McIntosh’s group suggests asking how confident you are in each answer instead of guessing your total score. When measured that way, low performers in their studies were appropriately less confident (McIntosh et al., 2026).
• Get an outside score. Feedback and testing show where you actually are, and in 1999, building participants’ skill helped them see their limits (Kruger & Dunning, 1999). Our guide on how to learn something new covers ways to test yourself.
• Look for reasons you might be wrong. People overestimate the accuracy of their judgments partly because they don’t consider why those judgments might be wrong (Nickerson, 1998). Our guide to confirmation bias explains the habit.
• Don’t use it as an insult. McIntosh’s group warns against using the effect for dismissive caricatures of people who disagree (McIntosh et al., 2026). Labeling someone “Dunning-Kruger” also doesn’t answer their argument, which is the problem with an ad hominem.