Share or save this article
AUDIO
Listen to this article
日本語
Checking your sign-in status…
An audio edition based on this article.
People who score poorly on a test say, “I did pretty well.” High scorers, meanwhile, assess their performance cautiously. This account led to a famous explanation: the less capable people are, the less able they are to recognize their own ignorance. Yet treating an extreme graph circulating online as a law of human character is risky. The question is **how directly low scores are linked to misjudging those scores**.
What the original experiments measured
In 1999, Justin Kruger and David Dunning reported four studies involving humor, grammar, logic, and related tasks. They compared participants’ actual performance with their assessments of how well they had done. The authors interpreted the results as showing that lower-performing participants overestimated their abilities more. They argued that difficulty recognizing errors also affected self-assessment.
The famous finding was that participants in the bottom quarter ranked around the 12th percentile on average, yet estimated their standing at around the 62nd percentile. A percentile indicates a position within the whole group, counting from the bottom; it does not mean 62% of answers were correct. These figures came from particular experimental results. They do not show that everyone with low ability necessarily thinks the same way.
The research rests on an intriguing line of reasoning. The knowledge needed to judge what is correct overlaps, to some extent, with the knowledge needed to check one’s own answers. Someone who does not know grammar, for example, may both answer grammar questions incorrectly and struggle to recognize those answers as wrong. It is therefore quite plausible for both performance and self-assessment to be inaccurate.
How measurement shapes the graph
Here is the crucial point. Suppose few people understand their actual ability perfectly, and everyone’s self-assessment contains some error. Rank people by their actual scores, from lowest to highest, and select only the bottom group. Comparing their self-assessments with their scores will tend to make the self-assessments look higher. The group was selected for low results, including chance failures, while self-assessment is a measurement with different errors.
In statistics, “regression to the mean” plays a role in this phenomenon. When a group selected for extreme values is measured again using a method with different errors, the results tend to move toward the middle. Consider a simple illustration: someone whose test performance is around 50 happens to score 30 and enters the bottom group. If they rate their ability at around 50, this looks like an “overestimate of 20 points.” That alone, however, does not establish that they lacked the ability to assess their own competence.
The same process works in reverse for the top group. People who happen to score well look modest if they rate themselves below that score. Of course, this does not fully explain every observed result. The point is that **how much of the apparent gap reflects a psychological deficit and how much arises from the measurement procedure** must be examined separately.
| What the graph appears to show | Possible explanation | Further checks needed | |---|---|---| | Low scorers report relatively high self-assessments | Difficulty recognizing errors | Does the same pattern appear on other tasks? | | Low scorers assess themselves near the average | Measurement error and regression to the mean | Does the analysis account for the error structure? | | High scorers appear modest | Humility, different comparison groups, statistical regression | Can they distinguish their number of correct answers from their rank? |
Counterevidence: “Almost entirely a statistical artifact”
In 2020, Gignac and Zajenkowski used self-assessments and objective measures of intelligence to argue that conventional tests of the Dunning–Kruger effect have statistical problems. In an analysis combining multiple datasets, they accounted for regression to the mean and the tendency to rate oneself above average. They reported that the results did not support the hypothesis that people with lower ability are especially inaccurate in their self-assessments.
It would also be premature to conclude that “the 1999 research was entirely wrong.” The 2020 study mainly examined self-assessments of intelligence, which are not directly equivalent to the humor and grammar tasks used in 1999. People may indeed sometimes lack the ability to recognize their own incorrect answers. The dispute concerns more than whether such psychological processes exist. It also concerns **whether the familiar curve itself can be called proof of those processes**.
This distinction also matters when someone makes a highly confident statement. Confidence alone does not reveal whether the person lacks knowledge, is performing a role, or is assessing something different. Using the effect as a label to single someone out—equating confidence with the Dunning–Kruger effect—goes beyond the scope of the research.
What does measuring your own confidence require?
FORWARD BASIS takes the view that asking **what self-assessment data are being compared with** is more valuable than popularizing the effect as a rule for mocking people with lower ability. Are people predicting their rank or their percentage of correct answers? Did they take the test once or several times? Did they see the answer key before assessing themselves? Is the comparison group the same? When these details change, the meaning of the numbers changes too.
Self-assessments can be inaccurate for several reasons. Sometimes people lack the knowledge to recognize their mistakes. Other discrepancies may arise from luck with the questions, error, or confusion between scales. The 1999 research popularized the idea of measuring confidence and ability separately. Subsequent research taught us to examine the measurement methods when interpreting the graph. The conclusion is not that “people with lower ability are always arrogant.” It is that **any discussion of the gap between someone’s confidence and actual ability must also explain how that confidence was measured**.
SOURCE
Sources
These are the primary sources we used to check the facts. The article identifies our own analysis.
Recommended articles
Why Do We Assume Everyone Thinks Like Us? Experiments on the False-Consensus Effect
Drawing on four studies from 1977 and replication research, this article examines the tendency to assume most people share one’s choices—and the limits of that tendency.
Revenue Down 12%, Operating Profit Up 49%: Why Console Unit Sales Alone Cannot Explain Nintendo’s Business
Nintendo’s interim financial results on November 5, 2026, showed that profit rose even as console unit sales fell. However, software’s share of sales was not the only factor. Understanding the profit increase requires distinguishing its drivers, including a one-time factor: US tariff refunds.
Does Costco’s 10% Revenue Growth Mean Customers Bought 10% More? Separating Gasoline, Exchange Rates, and Gross Profit
In fiscal 2026, Costco increased net sales from merchandise and other sales by about 10% from the previous year. That looks like solid performance for a membership warehouse retailer. But revenue reflects more than the volume customers bought. It also reflects gasoline prices, exchange rates used to convert overseas revenue into dollars, and the impact of adding stores. When revenue grows, what should we conclude has increased? We break down the company’s official US financial results.[1][2]
The Dunning–Kruger Effect
What the Graphs Really Show