Psychology concept
The finding that people who do worst on a task tend to overestimate how well they did. How big it is, and how much of it is a statistical illusion, is still argued over.
Reviewed 2026-10-07
The Dunning–Kruger effect is the finding that people who perform worst on a specific task tend to overestimate how well they did.
Original finding (1999): people in the bottom quarter scored around the 12th percentile but guessed they were around the 62nd.
Now: the pattern shows up again in new data, but several studies argue much of it comes from how the data are graphed and analyzed. The effect may be real but smaller than its fame suggests.
The original study
Justin Kruger and David Dunning published Unskilled and unaware of it in 1999. Across four studies, students took tests of humor, grammar and logic and then estimated how they had done. Those in the bottom quarter scored around the 12th percentile on average but estimated their performance around the 62nd. Original paper (abstract) ↗
How it's measured
Researchers compare an objective test score with the person's own estimate, often as a guessed percentile ("I did better than 60% of people"). The gap between the two is the miscalibration. The original studies were snapshots of specific tasks, not tracking how people learn over time.
What the evidence says now
SettledSelf-estimates and actual scores often don't line up, and low scorers tend to overestimate on average.DebatedHow much of that pattern is a statistical artifact (noise, floors and ceilings, regression to the mean) versus real overconfidence.UnknownHow far it applies outside test-like tasks, and across cultures.
2002 DefenseKruger and Dunning acknowledged that regression effects contribute to the pattern, but argued they don't explain all of it. Source ↗
2016 ChallengeSimulations with random numbers showed that common ways of plotting self-assessment data can create the pattern by themselves. Source ↗
2020 ChallengeIn 929 community participants, a test designed to separate real effect from artifact found no significant effect for intelligence; the authors suggested the effect may be smaller than reported. Source ↗
2022 ChallengeA statistical model showed bounded scores (you can't score below 0% or above 100%) can produce the pattern without any psychological explanation. Source ↗
2023SupportA response and replication reported a statistically significant effect in intelligence data, though small. A separate critique argued the 2020 result depended on how answers were recoded. Source ↗
2024 RefinesA simulation-based analysis estimated a sizable overestimate would affect only a small share of people. Source ↗
2025SupportUsing a British birth cohort, researchers reproduced the pattern and found it wasn't specific to one gender, though men and women erred in different directions. Source ↗
2026 ChallengeIn a low-stakes science test, at least 90% of people judged themselves imperfectly but reasonably well; the authors argue random guessing explains nearly all big overestimates. Source ↗
Common misunderstandings
"It means stupid people think they're smart." The studies are about misjudging performance on a specific task, not general intelligence.
The "Mount Stupid" curve. The popular chart of confidence rising, crashing and recovering as you learn isn't from the original research, which didn't follow anyone over time.
"Experts always underestimate themselves." High scorers in the original studies underestimated their relative standing, which isn't the same thing.
Before checking answers on any 10-question quiz, write down the score you expect. Do it for three quizzes. Is your gap the same each time?
Done when: you've compared three estimates with three real scores.
The pattern keeps appearing in data, including a 2025 birth-cohort study. But several analyses since 2016 show that much of it can come from statistics alone, and one 2024 analysis estimates a big overestimate affects only a small share of people. Fair summary: probably real, probably smaller than its fame.
By comparing a person's test score with their own estimate of it, often as a guessed percentile. The original logic test also asked people to estimate their score.
No. That curve describes learning over time; the original studies measured one test at one moment.
Not from a feeling. You'd need to compare your estimates with real results in a specific skill. Practice tests with honest scoring are the closest everyday version.
Some studies look at real settings. A 2025 study found tournament chess players overestimated their skill on average, most among lower-rated players. But there's no general answer for everyday life yet.
short, careful, and warns about the famous graph
Advanced; see exactly what was measured.
The statistical-artifact argument, with real exam data.
Readable, but argues one side. Pair it with the 2025 study above.
· added 2026-10-07 current
Fernandes and colleagues report a Dunning–Kruger pattern both when people worked alone and with AI, and a larger gap in the AI group. That differs from their own 2024 result, where the pattern disappeared with AI.
Connects to:
· added 2026-10-07 current
In a science-literacy assessment, at least 90% self-assessed imperfectly but reasonably well; the authors argue random guessing explains nearly all big overestimates.
Connects to:
· added 2026-10-07 current
Tournament players overestimated their skill on average, with the largest overestimates among lower-rated players. That is notable because chess ratings give constant, public feedback.
Connects to:
· added 2026-10-07 current
Using several ability measures from a large British cohort, the pattern held in secondary data and was not specific to one gender.
Connects to:
Part of Psychology: How psychology knows things