Trap/Cognitive Bias/No. 0705

Overconfidence Effect

You promise the client Friday, leave no buffer, and are surprised when one ordinary snag pushes delivery into next week.

Also called Overconfidence Bias

a trap: easy to walk into

Evidence: Well established
Read: 6 min
Links: 23 connections
Useful when: Deciding under uncertainty · Forecasting · Risk and safety · Running projects

01You've seen this when…

in life
After a close call, you still rank your driving in the top quarter. In your carpool, everyone gives themselves the same rank.
at work
You expect to finish twelve customer interviews this week. You schedule around that number, even though you’ve never managed more than eight.
out in the world
A transit proposal gives a narrow range for future ridership. At the hearing, speakers treat the forecast as a settled fact.

02The idea

You can feel sure about a prediction just as you can about your ability or standing, even when you haven’t earned that certainty. Overconfidence is the gap between that confidence and a suitable check on reality.

The name covers three different mistakes:

Overestimation puts your performance too high, and overplacement puts your rank too high. With overestimation, you expect eight correct answers but get five. You predict more sales than you can deliver. With overplacement, you believe you’re among the best negotiators in your department, but comparable results put you near the middle. Illusory superiority belongs here.
Overprecision makes your uncertainty too small. You give a narrow range for a repair bill when the possible outcomes are much wider. The central estimate might be reasonable while the stated confidence exceeds what the evidence supports.

These forms need not travel together. Someone can underestimate their score while overestimating their rank, especially on an easy task where nearly everyone does well. Someone else can accurately judge their ability but be far too certain about a forecast.

The benchmark matters. A single failed prediction is too little evidence to establish overconfidence. A forecast with a 90% chance of success is allowed to fail sometimes. The problem appears when confidence repeatedly exceeds accuracy, or when self-assessment conflicts with a meaningful comparison.

03Why it happens

A clear story hides missing information. Once you can explain how the launch will succeed, the explanation feels complete. Unseen dependencies and unfamiliar failure modes don’t announce themselves. This resembles the illusion of validity: a convincing interpretation feels more reliable than its evidence warrants.
Your own case gets more attention than the comparison. You know your preparation, intentions and recent successes. You know less about everyone else’s. Ranking yourself requires both sides, but the second often receives only a guess.
Uncertainty is hard to represent. Producing one plausible answer is easier than working out all the ways it could be wrong. A forecast can therefore sound precise even when the person making it has considered very few alternatives.
Feedback doesn’t always correct the error. Some outcomes arrive months later. Others have ambiguous causes. Without a record, you can remember the confident successes and explain away the misses. A decision journal makes that harder.

There is also a social complication: assurance can draw attention and help someone win a promotion or close a sale. Confident speech may reflect an incentive to sound certain, whatever the speaker’s private level of confidence. Delivery alone is an unreliable guide to someone’s private confidence.

04A worked example

In a 1981 study, psychologist Ola Svenson asked American and Swedish students to compare their driving skill and safety with others in their comparison group. In the American sample, 93% placed themselves in the more skillful half.

What it looks like A group of drivers with considerable faith in their own skill. Each person may remember careful decisions, successful journeys and mistakes made by other drivers.

What’s actually going on Only half of a group can occupy its top half, so the reported rankings collectively overstate the drivers’ standing. This is evidence of overplacement at the group level. It leaves open whether every participant is overconfident. The study uses self-ratings and includes no driving test, so it cannot identify each person’s actual skill or the size of each person’s error.

What would have helped Defining skill before judging it, then using comparable evidence. Avoiding collisions measures something different from the separate skills of handling hazards and obeying traffic rules. A standardized assessment would provide a firmer comparison than memories of other people’s worst moments. For everyday decisions, the immediate lesson is simpler: keep your safety margin when your only reason to reduce it is feeling safer than nearby drivers.

05How to spot it

Confidence rises without new evidence. Repeating the plan makes it feel safer, although nothing relevant has changed.
The uncertainty range excludes ordinary trouble. A modest delay, price change or dependency failure would already push the result outside it.
Rankings lack a comparison. Someone claims exceptional ability without comparable results from the relevant group.
Forecasts disappear after the outcome. Nobody checks how often predictions at a given confidence level came true.
Every miss gets a special explanation. Success proves skill; failure is always an exception.

06What to do instead

Identify which confidence you’re checking. Is the concern your expected result, your rank or the width of your uncertainty? Each needs different evidence. A performance test may leave a narrow forecast range unchanged.
Borrow a relevant track record. Use reference-class forecasting: examine comparable repairs, launches or applications before adjusting for your case. Explain why any adjustment is justified.
Record predictions before results arrive. Write down the outcome, deadline and probability. Review a batch to put your most memorable hit in context. Calibration means that events assigned roughly 70% probability happen roughly 70% of the time.
Give alternatives room to exist. Run a pre-mortem, seek contrary evidence, or ask an informed person to identify what your forecast leaves out. Finding a failure route doesn’t make it likely, but it exposes assumptions to check.
Match commitment to uncertainty. Use a pilot, reserve time or money, and avoid commitments that require your best-case estimate to be right. When consequences are severe, keep protective measures in place regardless of your confidence.

07When it isn’t overconfidence

High confidence can be deserved. Experienced people with repeated, clear feedback may judge familiar situations accurately. Judge their confidence by how well it fits their record. Bold delivery alone is an unreliable test.

The direction of error also changes with the task. Research on the hard-easy effect finds that people often overestimate absolute performance on difficult tasks and underestimate it on easy ones. Relative rankings can move in the opposite direction. Task selection and measurement therefore affect the result.

Overconfidence differs from optimism bias, which concerns favorable expectations about outcomes, and from the Dunning-Kruger effect, which concerns the relationship between performance and self-assessment. Humility can be inaccurate too. Replacing every confident estimate with a pessimistic one can simply exchange one error for another.

08Roots

In the 1970s, Baruch Fischhoff worked with Paul Slovic and Sarah Lichtenstein to give people general-knowledge questions with two possible answers. Choosing an answer was only part of the task. Participants also marked how certain they were. Even answers treated as certain contained errors. The researchers could now compare a feeling of certainty with something countable.

Lichtenstein worked with Fischhoff and Lawrence Phillips on a 1982 review that brought together probability-calibration research. The work connected everyday judgment with a practical forecasting problem: what should a stated probability mean across repeated predictions?

The label later spread far beyond probability questions. In 2008, Don Moore and Paul Healy argued that researchers were often mixing three different phenomena: inflated performance estimates, inflated relative rankings and excessive precision. Separating them helped explain why apparently contradictory findings could coexist.

An adaptive explanation remains a hypothesis. Dominic Johnson and James Fowler’s 2011 model showed conditions under which overconfidence could help competitors obtain resources. That establishes a possibility within a model. Whether human overconfidence evolved for that purpose or benefits you in the decision at hand requires separate evidence.

09How solid is this?

All three forms have substantial experimental support, but they do not reliably occur together. Their size and direction depend on task difficulty, comparison groups, feedback and measurement; universal claims that everyone is overconfident are overstated.

11Origin and sources

Earlier probability-calibration experiments include Fischhoff, Slovic and Lichtenstein (1977). Sarah Lichtenstein, Baruch Fischhoff and Lawrence Phillips synthesized this research in 1982; Don Moore and Paul Healy clarified the three forms in 2008.

[1]Moore, D. A., & Healy, P. J. (2008). The trouble with overconfidence. Psychological Review, 115(2), 502–517.
[2]Lichtenstein, S., Fischhoff, B., & Phillips, L. D. (1982). Calibration of probabilities: The state of the art to 1980. In D. Kahneman, P. Slovic, & A. Tversky (Eds.), Judgment under uncertainty: Heuristics and biases (pp. 306–334). Cambridge University Press.
[3]Fischhoff, B., Slovic, P., & Lichtenstein, S. (1977). Knowing with certainty: The appropriateness of extreme confidence. Journal of Experimental Psychology: Human Perception and Performance, 3(4), 552–564.
[4]Svenson, O. (1981). Are we all less risky and more skillful than our fellow drivers? Acta Psychologica, 47(2), 143–148.
[5]Johnson, D. D. P., & Fowler, J. H. (2011). The evolution of overconfidence. Nature, 477(7364), 317–320.

Suggest an edit· Updated 2026-10-02