Have you finished a tutorial thinking, “I understand,” only to get stuck when trying it yourself? Following an explanation and doing the task independently are different experiences. Sometimes we do not yet know what we are missing.
The Dunning–Kruger effect invites us to examine the gap between the ability we think we have and the performance we demonstrate. Reducing it to “people who know the least are always the most confident” oversimplifies the research.
What we see at first may be only part of the picture. This is conceptual artwork, not experimental evidence.
What is the Dunning–Kruger effect?
The name comes from research published by Justin Kruger and David Dunning in 1999. Participants with low test performance tended to substantially overestimate their performance. [1]
The researchers proposed that skills needed to perform correctly may also be needed to recognize errors. A lack of skill could make both doing a task and detecting mistakes difficult. This explanation involves metacognition: monitoring and evaluating our own thinking, and has been tested and challenged in later research. [1, 3, 4]
Imagine a new programmer whose code runs once. They assume it is finished, without knowing to test empty inputs, connection failures, or simultaneous users. They may not know what needs checking.
Everyday examples here are hypothetical illustrations, not evidence that a particular person exhibits this effect.
Overestimating is different from being the most confident
Suppose two people estimate their correct answers on ten practice questions.
| Person | Estimated correct | Actually correct | Overestimate |
|---|---|---|---|
| A | 6 | 3 | 3 |
| B | 9 | 8 | 1 |
A overestimates more, but B expects more correct answers. Lower performance can therefore accompany greater overestimation without the highest confidence. These invented numbers explain a distinction; they are not study data or proof of a mechanism.
What about the famous confidence mountain?
You may have seen a curve where confidence rises sharply, falls into a valley, then climbs with knowledge. That curve is not a result presented in the original 1999 paper. The original research compared performance and self-assessment across groups with different scores; it did not follow one person from beginner to expert. [1]
We should not treat the curve as a universal emotional journey. Feeling less confident alone does not establish growing expertise.
What is still debated?
Observing a gap and establishing its cause are different questions.
Ehrlinger and colleagues reported follow-up evidence in 2008 supporting the explanation involving difficulty recognizing one’s own errors, including in realistic settings and with incentives for accurate assessment. [2]
Gignac and Zajenkowski argued in 2020 that familiar patterns could arise from statistical factors and people’s tendency to rate themselves above average. Their intelligence study did not support some predictions using the methods they tested. [3]
One issue is regression toward the mean. When performance and self-assessment correlate imperfectly, people with extremely low or high performance can have estimates closer to the middle. This pattern alone cannot establish a psychological mechanism. [3]
In 2022, McIntosh and colleagues used a matrix reasoning task and found results challenging the account that low performers have a particular deficit in evaluating themselves. Low performers in that study were also less confident than high performers. [4]
These debates do not make inaccurate self-assessment disappear. They caution against using one explanation for every error, or one task to judge an entire person.
How can we apply this to learning?
These are practical applications of the article’s themes, not a program established by the cited studies to prevent every instance of the effect.
1. Turn “I understand” into something you can try
Close the tutorial. Solve a new problem, explain the steps, or make something independently. Ask, “What can I do without looking at the example?”
2. Record a prediction before seeing the result
Estimate how many practice questions you will answer correctly. Compare that estimate with the result. Repeating this creates a clearer record than relying on memory.
3. Ask for actionable feedback
Instead of “Is this good?”, ask “Which part of the reasoning is weakest?” or “What case have I not tested?” Specific examples and criteria give you something concrete to revise.
4. Ask what would change your mind
Write down what evidence would lead you to revise a conclusion. This makes correcting errors part of learning from the outset.
5. Describe the boundaries of your knowledge
“I can do this part, but have not tried that situation” communicates experience more clearly than “I know this topic.”
A practical learning framework, not experimental findings or a diagnostic test.
A short exercise: choose one topic. Write down what you think you can do, try without help, and record what the result tells you.
Apply it to yourself before labeling others
Confident speech or one mistake is insufficient to identify this effect. We need evidence about both someone’s performance and their assessment of it in the same area.
A person can be an expert in one domain and a beginner in another. The original paper limited its explanation to specific abilities, rather than assigning permanent categories to people. [1]
A useful question is “What evidence supports my belief that I understand?” Practice, clear criteria, and willingness to revise give confidence a firmer basis.
References
- Kruger, J., & Dunning, D. (1999). Unskilled and unaware of it: How difficulties in recognizing one’s own incompetence lead to inflated self-assessments. Journal of Personality and Social Psychology, 77(6), 1121–1134. Original paper · DOI
- Ehrlinger, J., Johnson, K., Banner, M., Dunning, D., & Kruger, J. (2008). Why the unskilled are unaware: Further explorations of (absent) self-insight among the incompetent. Organizational Behavior and Human Decision Processes, 105(1), 98–121. Article
- Gignac, G. E., & Zajenkowski, M. (2020). The Dunning–Kruger effect is (mostly) a statistical artefact: Valid approaches to testing the hypothesis with individual differences data. Intelligence, 80, 101449. DOI
- McIntosh, R. D., Moore, A. B., Liu, Y., & Della Sala, S. (2022). Skill and self-knowledge: empirical refutation of the dual-burden account of the Dunning–Kruger effect. Royal Society Open Science, 9, 191727. Abstract and full-text links