What Is the Dunning-Kruger Effect? Why Beginners Overrate Themselves

Dunning and Kruger tested Cornell students on logic, grammar and, in a nod to the robber, humour, then asked each to estimate how they had done compared with everyone else. Those who scored in the bottom quarter estimated on average that they were in the 60th percentile, above the middle. Their self-rating was not merely optimistic; it was almost unrelated to their actual score. The top quarter, meanwhile, slightly underrated themselves, guessing they were good but not quite as good as they were.

The explanation offered was a double burden. The skills needed to do a task well are the same ones needed to judge whether it has been done well. Someone who cannot spot a grammatical error in their own writing has no way of knowing it is there, so they feel fine about the writing. Give the weak performers training in the task and their self-estimates dropped towards reality: they had learned enough to see their mistakes.

The effect is usually drawn as a graph that has become an internet fixture: confidence soaring on the first day of learning, plunging into a valley as the learner discovers how much there is to know, and climbing slowly as real expertise builds. The peaks and valleys in that popular picture are an invention; nothing in the research has that shape. What the data show is simpler and duller: everyone's self-estimate sits nearer the middle than their real score, so the worst performers overrate themselves and the best underrate themselves.

The finding applies within a skill, not across people. The same person may be well calibrated about their cooking and badly calibrated about their driving. It is not a personality type, and it is not a description of other people. Nearly everyone has a domain in which they are the bottom quarter.

Since 2020 a number of researchers have argued that part of the pattern is a statistical artefact. If people's self-estimates were simply noisy guesses unrelated to their ability, the same graph would appear, because the lowest scorers have nowhere to guess but upward and the highest have nowhere to guess but downward; that is regression to the mean, not a psychological effect. Simulations with random self-estimates reproduce a good deal of the original curve.

The response from Dunning and others is that the artefact explains some of the gap but not all of it, and that later studies using better measures, and the fact that training changes the estimates, show a real component remains. The fair summary is that the effect exists but is smaller and less dramatic than its reputation, and that the original explanation, poor performers lacking the tools to evaluate themselves, is supported without being the whole story.

The uncomfortable implication is that the effect cannot be felt from inside. The beginner's confidence is a genuine confidence, produced by the absence of any signal that anything is wrong, and it feels identical to the confidence of an expert. That is why people in the grip of it are not helped by being told about the effect; they agree it exists and assume it applies to someone else. In one follow-up, participants told about the effect and asked to re-estimate barely moved.

What does move the estimates is feedback and comparison: seeing your work marked against a standard, watching someone skilled do the same task, or being forced to explain your reasoning in enough detail for the gaps to show. Expertise brings its own bias in the other direction, because experts find the task easy and assume others do too.

The effect has become a way of calling other people stupid, which is the one use the research does not support, since the people invoking it are subject to it as well. The useful applications are all pointed inward:

• •Treat strong confidence in a new field as a sign you have not yet found the hard parts
• •Seek marking, testing and comparison rather than self-assessment
• •Explain your reasoning to someone who knows more, and notice where it thins
• •Expect your estimates to be too high where you are weakest and slightly low where you are strongest
• •Remember the effect applies per skill, and you have skills in every quarter

Some groups are unusually good at knowing how good their own judgements are, and the best studied are weather forecasters. When forecasters in the United States say there is a seventy percent chance of rain, it has rained on close to seventy percent of those days, a match that most experts in other fields do not achieve. The reason is not special intelligence. Forecasters make many predictions, state them as numbers, and find out within a day whether they were right, over and over for years. Fast, frequent, unambiguous feedback is exactly what the Dunning-Kruger research says is normally missing.

You can test your own calibration with a simple exercise. For ten questions with numerical answers, such as the length of a river or the year a building was finished, give a range you are ninety percent sure contains the true answer. If you are well calibrated, about nine of your ten ranges should contain it. Most people who try this find that far fewer do, because they make their ranges too narrow, a sign of overconfidence even in people who think of themselves as cautious. Repeating the exercise and checking results is one of the few reliable ways to improve.

The Dunning-Kruger effect is often mixed up with a broader tendency called the better-than-average effect, in which most people rate themselves above the typical person on almost any desirable trait. In a well-known study published in 1981, the psychologist Ola Svenson asked students in the United States and Sweden to compare their driving skill with that of others in their group. Most placed themselves in the top half, including around nine in ten of the Americans, which cannot be true of any group. The two ideas differ: one describes nearly everyone, while Dunning-Kruger describes how the error changes with actual skill.

The impostor phenomenon is sometimes presented as the opposite of Dunning-Kruger, but it is a different thing. Described by the psychologists Pauline Clance and Suzanne Imes in 1978, from their work with high-achieving women, it is the persistent feeling of being a fraud whose success is down to luck and who is about to be found out, despite clear evidence of ability. It is not a medical diagnosis, and it is reported by people of all kinds. The strongest performers in Dunning and Kruger's studies did not feel like frauds; they knew they had done well and misjudged only their rank.

The Dunning-Kruger effect is usually told about beginners, but overconfidence is not limited to people with little knowledge. In a long project published in 2005, the psychologist Philip Tetlock collected tens of thousands of predictions from experts about political and economic events. Many did little better than simple rules of thumb, and those most confident in a single big idea tended to do worst.

Tetlock later ran forecasting tournaments and found a group of so-called superforecasters who consistently beat others, including professional analysts. What marked them out was not genius but habits: breaking questions into parts, updating views in small steps as evidence arrived, and keeping careful track of their own errors. Good calibration, it turns out, is a skill that can be practised.

The Dunning-Kruger effect is the tendency of people who are weak at a task to overrate their performance, partly because judging the work needs the same skill as doing it, and partly, later analysis suggests, because of how self-estimates behave statistically. It is real but modest, it is invisible from inside, and the remedy is external feedback rather than self-reflection.