Where it sits
And the gap is widest on precisely the questions where being wrong costs most.
Tell it apart
| Aspect | Overconfidence effect | Dunning–Kruger effect |
|---|---|---|
| Is a | Family of three | One pattern |
| Concerns | Confidence vs accuracy | Skill vs self-assessment |
| Reverses on hard tasks | Yes, by type | Yes |
| Named | 2008 (the split) | 1999 |
Related, not interchangeable. Overconfidence names the measurement family; Dunning–Kruger names one pattern inside it — and mixing them is how 'better than average' evidence gets cited for claims about people overestimating their scores.
What it doesn't mean
Moore and Healy documented systematic REVERSALS. On difficult tasks people overestimate their own score but believe they are WORSE than others; on easy tasks they underestimate their score but believe they are better. A 2018 study across four populations found underconfidence in 22% of its 32 experimental conditions, concluding overconfidence is highly context dependent rather than universal.
They are separately measured and separately distributed. Moore and Healy's survey found roughly 64% of empirical studies examined overestimation, 31% overprecision, and only about 5% overplacement — yet overplacement, the 'better than average' version, is the one most often cited in popular writing.
What it really means
The single most useful thing anyone has done with this subject is split it into three.
Before Moore and Healy's 2008 paper, overconfidence was one word covering three different measurements that behave differently and sometimes point in opposite directions. Their titles for them are now standard:
Overestimation — thinking you did better than you did.
Overplacement — thinking you beat more people than you did. This is the famous one: 93% of American drivers rating themselves better than the median, in Svenson's 1981 study.
Overprecision — excessive certainty in your own estimates. Asked for a range you are 90% sure contains the true answer, people supply ranges that contain it less than half the time.
These are not three ways of saying the same thing, and the evidence for one is not evidence for another. Moore and Healy's own survey of the literature found roughly 64% of empirical studies examined overestimation, 31% overprecision, and only about 5% overplacement.
Which is a problem, because overplacement is the version popular writing cites most. The claim best supported by the fewest studies is the claim that travelled.
Then the finding that undoes the simple story entirely. People are not simply always overconfident. Moore and Healy documented systematic reversals, and they run opposite ways for different types:
On difficult tasks, people overestimate their own score but believe they are worse than others.
On easy tasks, they underestimate their own score but believe they are better.
So on a hard exam you will think you did well and that everyone else did better. On an easy one, the reverse. A 2018 study across four populations found underconfidence in 22% of its 32 experimental conditions and concluded the effect is highly context dependent.
The practical residue is narrow and worth keeping. Overprecision is the sturdy one — the tendency to give ranges that are too tight. It is also the most correctable, because unlike a belief about yourself, it can be checked against outcomes. Write the range down. Look later. Most people find theirs needs to be roughly twice as wide as it feels.
Where it comes from
English — beyond
Latin confidere — to trust fully
The vocabulary that matters is the 2008 three-way split, coined precisely to stop three different phenomena being reported as one.
- 1981
Svenson's driving study: 93% of American drivers rate themselves better than the median.
- 2008
Don A. Moore and Paul J. Healy publish 'The Trouble With Overconfidence', formalising three distinct types and showing they dissociate.
In real life
- Overestimation: you're sure you got 18 out of 20 in the quiz. You got 13.
- Overplacement: asked how their driving compares with other people's, most drivers put themselves in the better half.
- Overprecision: asked for a range you're 90% sure contains the height of the Eiffel Tower, you say between 280 and 310 metres. The tower is about 330 metres to its tip, so the range was too narrow to deserve that much certainty.
Questions
Q01What are the three types?
Overestimation — thinking you did better than you did. Overplacement — thinking you beat more people than you did. Overprecision — excessive certainty in your own estimates. They dissociate, and only the second is the famous 'better than average' finding.
Q02Which is the most robust?
Overprecision. Moore and Healy found it more persistent than the other two — and, oddly, its presence actually reduces the magnitude of both.
Q03What's the hard-easy effect?
Overconfidence appears reliably on hard tasks and reverses on easy ones — with the direction of reversal differing for overestimation and overplacement, which is exactly why treating them as one effect produces contradictions.
Sources
- Overconfidence effect — Wikipedia.
- Muthukrishna et al. (2018), 'Overconfidence is universal? ... four populations' — PLOS ONE.
- Moore, D. A. & Healy, P. J., 'The Trouble With Overconfidence' — Psychological Review 115(2), 502-517 (2008); author's copy hosted at Ohio State.
- over- (etymology) — Wiktionary.