The Dunning-Kruger chart, explained

In 1999, two psychologists at Cornell gave students tests of logic, grammar, and humor, then asked them how well they thought they'd done compared with their peers. The bottom quarter of test-takers placed themselves, on average, at the 62nd percentile (meaning they believed they'd outscored 62% of their peers), even though their actual scores put them at the 12th: a gap of fifty percentile points. The chart from that study became one of the most cited in all of psychology, with close to 7,000 citations to date, and its name entered everyday speech: the Dunning-Kruger effect.

Here is the part most people who use the phrase don't know: you can get the exact same chart out of a random-number generator. No students, no tests, no real incompetence anywhere in the pipeline: just arithmetic. The simulator below does it live.

At 0%, the self-estimate line comes out nearly flat, hovering around 50, because when guesses are completely independent of actual skill, their average always sits near the middle of the scale, no matter the quartile. The actual-score line, by construction, climbs from about 12 to about 88. The two lines cross somewhere in the middle: exactly the shape from the textbooks. Even at 0% insight, with zero real link between how good someone is and how good they think they are, the gap in the worst quartile comes out to nearly 38 points: three-quarters of the fifty from the real 1999 study. Drag the slider higher and the gap shrinks, not the other way around: pure noise accounts for more of the famous effect than any amount of partial self-knowledge would.

The 1999 test

The original study, by Justin Kruger and David Dunning and published in the Journal of Personality and Social Psychology, used four different tests: logical reasoning, grammar, humor. It found the same pattern every time: the bottom quarter wildly overestimated their standing, while the top quarter slightly underestimated theirs (actual scores above the 86th percentile, estimates around 70–75). The idea quickly became a cultural cliché, with cartoons, jokes, and office arguments, repeated so often it hardened into a folk axiom about human nature: fools don't know they're fools.

Why the chart draws itself from pure noise

The explanation came in 2016, from geologist and education researcher Edward Nuhfer and colleagues, in the journal Numeracy. The mechanism is simple once you see it: if you group people by their real score, and self-assessment has even a small real connection to reality, the average self-estimate inside each group slides toward the middle of the scale, simply because self-assessment also contains noise, and noise, by construction, averages out to a central value. Statistics has a name for this: regression to the mean. The graphing convention Dunning and Kruger used, copied since into thousands of presentations and papers, makes this mathematical artifact look identical to a real psychological effect.

That doesn't mean people have no idea where they stand. When Nuhfer analyzed real data from more than 17,000 students, he found something less dramatic and more useful: self-assessment accuracy tracks a person's numeracy and how specific the question is. When a test is concrete and feedback is clear, people rate themselves reasonably well. When the question is vague and the scale is an abstract percentile, the artifact above shows up, no matter who's answering.

The statisticians' fight

From here, the literature turns into an active methodological dispute, not a closed case. In 2020, psychologists Gilles Gignac and Marcin Zajenkowski reran the analysis using item-response-theory methods applied directly to IQ self-estimates, and found that self-assessment error stays roughly constant across the whole ability range, not just at the tail. Their conclusion, published in the journal Intelligence: the Dunning-Kruger effect is, for the most part, a statistical artifact. Three years later, philosopher Avram Hiller contested that result, arguing that Gignac and Zajenkowski's choice to recode estimates onto a linear IQ scale, rather than a normalized one, could account for their finding on its own. Another group, led by Curtis Dunkel, reran the numbers and found a real but small residual effect. Nobody in this dispute defends the popular version (fools have no clue), but nobody has agreed on whether anything real survives once the artifact is subtracted out, either.

The 2026 reversal

The most recent chapter flips the story entirely. In a paper published on 27 July 2026 in Psychological Review, economists Chris Dawson (University of Bath) and David de Meza (London School of Economics) went back through the large datasets used in the original studies and corrected for the regression to the mean discussed above. The result: once the artifact is removed, the pattern reverses, and the most able people turn out to be the most overconfident, not the least. The mechanism the authors propose comes down to strategy: genuinely able people have the most to gain by signaling hidden ability beyond what they can prove on the spot. It's a kind of bluff, and the authors argue this tendency to bluff coevolved with loss aversion. In other words, if a "classic" effect still exists somewhere, it may have had the wrong label on it for a quarter of a century.

An American effect

There's also a geographic limit rarely mentioned, about the ingredient behind the effect: people's tendency to rate themselves favorably in the first place. A meta-analysis by Steven Heine and Takeshi Hamamura, in Personality and Social Psychology Review, compared that tendency across 91 cross-cultural comparisons. "d" here measures how large the gap between groups is: 0 means no difference, and anything above 0.8 counts as large. Western samples show a clear self-serving bias (d = 0.87), a 0.84 average gap from East Asian samples, which show almost none (d = −0.01); Asian-American samples fall in between (d = 0.52). In Japan, people tend to rate themselves lower than their peers rate them, practically the opposite of the story this article opened with. A good part of the urge to think you're better than you are turns out to be a self-promotion norm specific to individualist Western culture, which gives the effect described above a cultural root alongside its statistical one.

Not in every domain

The effect doesn't even show up everywhere it should, if it were a general feature of cognition. A 2024 study in Scientific Reports tested the pattern on self-assessed creativity and found no strong support for it. "Competence" is a word that covers wildly different things: logical reasoning, grammar, creativity, driving, a sense of humor. How well any of them can be self-judged depends on how clear and how fast real-world feedback arrives, not on one rule that applies to every human skill.

Test your own calibration

A single test, given to a single person, can't prove anything about a "law" of self-knowledge, any more than a single dice roll proves a die is loaded. But you can feel, firsthand, how easily a guess slides one way or the other. Answer the six questions below, then, before you see the results, estimate how many you think you got right.

What actually holds up

Not everything to come out of this story is false or up for debate. Another finding of Nuhfer's, based on the real data from more than 17,000 students mentioned above, doesn't depend on who wins the statistical fight: self-assessment accuracy tracks how specific and frequent the feedback is, not just one vague estimate of "where you stand compared to everyone else." The practical upshot is simple: one number, taken in isolation, tells you far less than it seems to, and the best defense against your own blind spots is access to concrete, repeated feedback.

Method. The simulator above generates 240 pairs of numbers (actual score, self-estimate) live in your browser, using the mechanism described in the text, without loading data from any external database; regenerating the population repeats the whole process from scratch. The calibration quiz is an illustrative six-question general-knowledge probe, not a validated psychometric instrument; the "crowd" you're implicitly compared against is nothing more than the mechanism explained above, applied to your own answers. Figures about the cited studies come from the published papers and their academic and press coverage, checked in September 2026; where full-text access to a paywalled journal wasn't available, figures were taken from reliable secondary sources rather than recomputed from the original dataset.

• Kruger, J. & Dunning, D. (1999), "Unskilled and Unaware of It," Journal of Personality and Social Psychology, 77(6).
• Nuhfer, E. et al. (2016), "How Random Noise and a Graphical Convention Subverted Behavioral Scientists' Explanations of Self-Assessment Data," Numeracy, 9(1).
• Gignac, G. E. & Zajenkowski, M. (2020), "The Dunning-Kruger effect is (mostly) a statistical artefact," Intelligence, 80.
• Hiller, A. (2023), comment on Gignac & Zajenkowski, Intelligence.
• Dunkel, C. et al. (2023), reanalysis of IQ self-assessment, Intelligence.
• Dawson, C. & de Meza, D. (2026), "Talking the Talk, Not Walking the Walk," Psychological Review, 27 July 2026.
• Heine, S. J. & Hamamura, T. (2007), "In Search of East Asian Self-Enhancement," Personality and Social Psychology Review, 11(1).
• "No strong support for a Dunning-Kruger effect in creativity" (2024), Scientific Reports, 14.