Dunning-Kruger effect: why the people who know doubt most

Ryan has been going to a coding course for three weeks and on Friday, over a beer, he explains to his friends that the whole field is overrated and anybody could pick it up in six months. Hannah has done the same work for ten years, runs a team, and when asked whether she understands it she says she is solid in one corner of it and the rest keeps slipping away from her.

Which of the two sounds more certain at that table is obvious within a sentence. To explain it, people reach almost automatically for a name that went mainstream within a few years: the Dunning-Kruger effect. Between what travels under that name on social media and what two psychologists actually measured in 1999, though, there is a considerable stretch of road.

What the 1999 study actually measured

Justin Kruger and David Dunning handed students at Cornell University several tasks. In one they judged which jokes were funny, and their ratings were set against ratings from professional comedians. In another they solved logic problems from law school admission tests. In a third they corrected grammar.

Then came the question that mattered. Everyone had to estimate not how many points they had scored, but how they stood against the other participants. How many people in the room they had beaten.

The result gets quoted to this day. Participants from the weakest quarter placed themselves on average above the middle of the group, though by points they belonged at the bottom. Those from the top quarter sold themselves short: they placed themselves high, but lower than they actually stood.

The authors explained it with a double curse. The knowledge you need to solve a task is roughly the knowledge you need to judge whether your solution is any good. Whoever lacks it has nothing left to notice the gap with. Among the strongest performers a different mechanism was at work: the task felt easy to them, so they assumed it had been easy for everyone else.

The last part of that paper tends to get forgotten. When the weaker participants went through a short training session in logic, not only did their performance improve, so did the estimate of it. They could judge themselves better because they finally had something to judge by.

The Mount Stupid that is not in the data

The picture most people call up under that name looks rather different. Experience runs along the horizontal axis, confidence along the vertical one. The curve shoots up at once into a sharp peak known as Mount Stupid, crashes into the Valley of Despair and from there climbs slowly toward the measured self-assurance of the expert.

No such picture appears in the 1999 paper. The original graph has four performance quarters along the horizontal axis and two lines: actual results rise steeply, while the self-estimate runs almost flat and sits a little above average in every group. No peak, no fall.

The difference between the two is substantive, not cosmetic. The internet curve claims that a beginner is more confident than an expert. The data say something far more ordinary: almost everybody puts themselves a notch above average, and among the best that guess simply lands closest to the truth.

A second flaw comes on top. The study was a snapshot, not a film. Nobody in it was followed from novice to master, so no curve through time can come out of it. The horizontal axis of the popular graph measures something the original research never measured.

Or is it only statistics?

Popular accounts reliably leave one thing out: the famous effect has had serious critics from the start, and their argument is uncomfortably simple.

It rests on regression to the mean. Every measurement carries a piece of randomness, so whoever lands right at the bottom is partly there through bad luck. Ask that person about a second, independent thing, namely their own estimate, and it comes out less extreme on average. Add the almost universal habit of rating oneself a notch above average, and the famous picture draws itself without anyone being blind to their own limits.

Ed Nuhfer and his colleagues showed in 2016 that a similar looking graph can be produced from randomly generated numbers. Gilles Gignac and Marcin Zajenkowski compared measured intellectual ability with people's own estimates of it in 2020 and concluded the effect is mostly a statistical artifact. They proposed different procedures for testing the hypothesis.

The other side is not staying quiet. Dunning and others object that the metacognitive part rests on data beyond one graph. People who lack a skill also judge other people's solutions worse, and the training above improved their estimates, which regression to the mean alone does not explain.

The dispute is not settled and this text will not settle it. Knowing about it is worth something, though. The names of Dunning and Kruger get played in arguments as a trump card for dismissing anyone inconvenient, and a trump card built on contested ground is not a strong one. It is more useful to know what survived even among the critics.

What is left of the famous effect

On one point both camps agree. A connection between how good we are at something and how good we feel we are does exist, but it is weak. Self-assessment is an instrument with a wide spread, and leaning on it alone is like measuring a room in paces.

The second point is plainer still. Anyone long in a field carries an ever more detailed list of what they cannot do. That is not a collapse in confidence but a widening horizon. Hannah can name ten things she is not up to, because she knows they exist. Ryan cannot name them, not out of arrogance, but because he has not run into them yet.

And then there is the mechanism from the top performers, which deserves more attention than it gets. Whatever is easy for you looks easy in general. Someone who reads a spreadsheet at a glance or calms an angry client with one sentence rarely counts that as a skill. They treat it as standard issue that everyone has, and shift their own level closer to the middle than it belongs.

The other side of the same coin

This is where the famous effect meets a phenomenon that looks like its mirror image. When a genuinely competent person reads their own uncertainty as evidence of incompetence, we call it impostor syndrome. The results are on the table; the inner verdict disagrees. The piece on what impostor syndrome is and where the doubts come from takes it apart in detail.

Doubt does not have one shape. In some people what dominates is the feeling of being a fraud, the quiet assumption that everyone around overrates them and that one day it will come out. Others are slowed down by discounting success: every good result gets explained away within seconds as a small thing that did not really count. And the third source tends to be luck and circumstance, where the credit for a job well done goes to timing, to the team or to a forgiving client.

A tempting shortcut then offers itself, the one you hear at every other conference: whoever doubts probably knows their craft, whoever is confident probably does not. It sounds pleasantly balanced and it is the same error as the original, merely reversed. A feeling is not a measuring device in either direction. There are people who lack confidence and also have plenty to work on, and it would be strange if there were not.

What separates healthy uncertainty from the impostor pattern is not the strength of the feeling but its relationship to evidence. Uncertainty before something new eases once the first results arrive. The impostor pattern overwrites results: good work is written off as preparation, chance or a low bar, so the doubt walks into the next situation exactly as strong as before. What that fear looks like from the inside is described in the piece on why you keep waiting to be found out.

How to calibrate yourself

Calibration is the one practical answer to both sides of the coin. The point is not to trust yourself more or less, but to narrow the gap between estimate and reality. A few things work better for that than thinking about yourself.

The first is a criterion instead of a feeling. Before you start something, write down in one sentence what a good result will look like. Once you hand the work in, you compare it with that sentence rather than with whatever mood you are in. It looks like a formality, but this is the step that separates judging the performance from judging yourself.

The second aid is feedback from someone who sees the output. You know your intention, other people see the result, and those are two different sets of data. The question "what would you do differently" brings in more than "was it any good?", because nearly everyone answers the second politely. And now honestly: when did you last get feedback that genuinely moved your estimate of yourself?

The third thing is to compare yourself with a specific skill, not with a whole person. The sentence "he is better" hands you nothing to work with, whereas "he gets the materials ready faster because he built himself a template" does. Where comparison helps and where it merely feeds the sense that you are not enough is unpacked in the piece on why everyone else looks more capable.

If you are wondering whether your doubts stay in a harmless range or already eat into your peace and your opportunities, a short impostor syndrome test will give you a rough picture. One honest note belongs with it: it measures the degree of self-doubt, not your abilities. Whether you are good at something is decided by your work and the feedback it brings, not by a questionnaire.

On your next larger task, try one small thing. Before you hand it over, write a single number on paper: how many people out of ten in your field would, in your view, have done it better. After the feedback arrives, come back to that number and add what actually happened. Three or four rounds will tell you more about your calibration than any graph with a peak and a valley.