Overconfidence Bias
Learn it → Apply it → Save as Skill
Overconfidence bias is the tendency to be more certain of your own judgments, predictions, and abilities than the facts actually support. Psychologists Don Moore and Paul Healy reviewed decades of conflicting findings in their 2008 paper "The Trouble With Overconfidence," published in Psychological Review (volume 115, issue 2, pages 502 to 517), and showed that what often gets treated as one bias is really three separate patterns that do not always travel together: expecting to perform better than you actually will, believing you are better than other people, and being more certain a specific estimate is correct than the evidence warrants. A person can be accurately humble about one of these and badly miscalibrated on another at the same time.
Dr. Wallace's Favorite Mental Models turns one well-chosen way of thinking at a time into something you can actually use. Each entry gives you a fast overview, an interactive walkthrough, varied real and practical examples, and an AI-ready Markdown skill you can keep.
Instead of only reading about a mental model, you can build your own AI skills library over time: a personal set of portable reasoning tools that help ChatGPT, Claude, and other AI assistants apply the thinking you value.
Overconfidence shows up anywhere a person has to estimate, predict, or judge: a delivery date, a driving skill, a court case, a stock forecast, a medical diagnosis. It has three separate faces. Overestimation is thinking you will perform better than you actually will. Overplacement is believing you are better than other people, the familiar "better than average" pattern. Overprecision is stating a narrower range of certainty than the evidence supports, such as saying you are 90% sure when the honest number is closer to 50%. The three do not always move together, which is why someone can be realistic about their own skill and still give dangerously narrow estimates.
Important: both figures are the documented results of published research, not invented illustrations. A group cannot mathematically all be above its own median, and a person who says "100% certain" every time should, by definition, be right every time if the label meant what it claimed.
Use the two axes below, whether the pattern is tied to how skilled someone is or shows up regardless of skill, and whether it is about judging yourself or judging outside information, to see how overconfidence bias relates to three commonly confused neighbors. Select a dot or a chip to read how each one compares.
Each stage builds on the one before it. Use the numbered buttons or Next to move through the walkthrough.
Open an example to see overconfidence bias in a specific context. The first five are documented research; the rest are common hypothetical scenarios used to teach the method, clearly labeled as such.
Ola Svenson asked American and Swedish drivers to rate their own driving skill and safety compared with other drivers. In the American sample, 93% placed themselves above the median, a result that is mathematically impossible for the group as a whole since only half of any group can be above its own median. The study is one of the most widely cited demonstrations of overplacement, the "better than average" form of overconfidence.
Don Moore and Paul Healy's 2008 paper "The Trouble With Overconfidence," published in Psychological Review, reviewed decades of research that seemed to contradict itself: some studies found people were overconfident, others found them underconfident. Their explanation was that researchers had been measuring three different things, overestimation of your own performance, overplacement relative to others, and overprecision in stated ranges, and treating them as if they were one phenomenon that should always move in the same direction.
A long line of calibration research, associated with psychologists including Baruch Fischhoff, Sarah Lichtenstein, and Paul Slovic, has repeatedly asked people to answer factual questions and then state how confident they are. When people say they are 100% certain of an answer, they are typically right only about 80% of the time. When asked to give a range they are 90% sure contains the true answer, the true answer often falls inside that range only around half the time, a textbook case of overprecision.
Stuart Oskamp gave clinical psychologists an increasing amount of case material about a real patient and asked them to make diagnostic judgments at each stage. As the psychologists received more information, their stated confidence rose from about 33% to about 53%, but their diagnostic accuracy stayed below 30% throughout. More information made the professionals feel more certain without making their judgments more correct.
Itzhak Ben-David, John Graham, and Campbell Harvey surveyed chief financial officers quarterly from 2001 to 2011, collecting more than 13,300 forecasts in which each CFO gave an 80% confidence range for the coming year's stock market return. The true return fell inside that stated range only 36.6% of the time, far below the 80% the range was supposed to guarantee, showing severe overprecision among senior financial professionals forecasting a market they study for a living.
An engineering team is asked how long a new integration will take and answers "two weeks" with real confidence, even though the work touches a third-party system nobody on the team has used before. The estimate is stated as if it were a near-certainty rather than a guess with wide error bars.
A founder pitching a seed round states with total confidence that the company's technical lead is insurmountable, dismissing the possibility of a fast-moving competitor entering the space, without being able to point to specific evidence ruling that scenario out.
A lawyer facing a case with several genuinely uncertain legal questions tells a client the odds of winning are "95%," a number that sounds precise and reassuring but is not actually anchored to any track record of similar cases or a real accounting of how each open question could go.
A physician forms a working diagnosis early in an appointment and begins interpreting new symptoms as confirming it, feeling increasingly certain as the appointment goes on, even though the added details were never weighed against an alternative explanation.
This is an illustrative, invented scenario designed to teach the method. The figures are the product manager's own working notes, not documented research statistics.
Expand each item for the research and reasoning behind the model.
Don Moore and Paul Healy's 2008 paper "The Trouble With Overconfidence," published in Psychological Review (volume 115, issue 2, pages 502 to 517), reviewed decades of findings that often seemed to contradict each other, some showing strong overconfidence, others showing underconfidence on the same kinds of tasks. Their contribution was showing that researchers had been bundling together three distinct patterns, overestimation of your own performance, overplacement relative to others, and overprecision in stated ranges, that do not necessarily rise and fall together, which explained much of the apparent contradiction in the earlier literature.
No. Ola Svenson's 1981 finding that 93% of American drivers rated their own skill above the median has been followed by similar "better than average" findings across many other domains, including job performance, ethical behavior, and sense of humor. The pattern is strong enough to have its own nickname, the Lake Wobegon effect, after the fictional town where "all the children are above average."
Not reliably. Itzhak Ben-David, John Graham, and Campbell Harvey's survey of chief financial officers across more than 13,300 forecasts from 2001 to 2011 found their 80% confidence ranges for stock market returns captured the true result only 36.6% of the time. Stuart Oskamp's 1965 study of clinical psychologists found a similar pattern in a different profession: confidence rose from about 33% to about 53% as psychologists received more case information, while diagnostic accuracy stayed below 30% the entire time.
No. Calibration researchers describe a "hard-easy effect": people tend toward overconfidence on hard or unfamiliar questions, where the honest answer is "I'm not sure," but can flip toward underconfidence on easy or very familiar questions, where they second-guess answers they actually know well. Treating overconfidence as a universal, one-direction rule misses this reversal.
Not entirely. Cross-cultural research, including meta-analytic work associated with psychologist Steven Heine and colleagues, has found the self-enhancing "better than average" pattern, overplacement specifically, tends to be weaker or sometimes reversed in some East Asian cultural samples compared with North American ones. Overprecision in formal calibration tasks, stating ranges that are too narrow, appears to show up more consistently across cultural groups, which is one more reason Moore and Healy's three-part split matters.
The Dunning-Kruger Effect is specifically about people with low ability in a particular skill also lacking the metacognitive skill to recognize how low that ability is, and the gap between confidence and ability shrinks as actual skill improves. Overconfidence bias is broader and does not require low skill at all; overprecision in particular shows up in genuine experts, including the CFOs in the Ben-David, Graham, and Harvey study, who have deep, relevant experience and are still badly miscalibrated on specific numeric ranges.
Dunning-Kruger is about low skill producing low awareness of that low skill. Overconfidence, especially overprecision, shows up in genuine experts too, so assuming high skill rules it out is itself a mistake.
Overconfidence is a calibration problem, not a character flaw. Sincerely confident people are still often wrong at close to the rate the research predicts, which is exactly what makes it hard to self-detect.
Someone can be accurately humble about their own average skill, low overplacement, while still giving dangerously narrow confidence ranges, high overprecision. Checking only one type can miss the other.
Chief financial officers forecasting markets they study professionally were right about their own 80% confidence ranges only 36.6% of the time. Experience did not fix the miscalibration on its own.
General humility does not change a specific numeric range. The fix that actually works is checking a track record against stated confidence and widening the particular estimate that keeps missing.
Assuming the bias always points toward overconfidence misses that people often swing toward underconfidence on genuinely easy or familiar questions, second-guessing answers they actually know well.
One wrong forecast does not prove overconfidence, and one right forecast does not prove good calibration. The pattern only becomes visible across many predictions compared against their stated confidence.
someone with a long, checked track record of 90% predictions that actually land correctly about 90% of the time is calibrated, not biased. The model targets systematic miscalibration, not confidence itself.
on hard or unfamiliar tasks, people often swing toward overconfidence, but on easy or very familiar tasks, research on the hard-easy effect finds people can swing toward underconfidence instead.
the model describes a pattern across many judgments, not evidence about the accuracy of any single forecast in isolation.
research finds the self-enhancing "better than average" pattern is weaker or reversed in some cultural samples, while overprecision in formal calibration tasks looks more consistent across groups.
This browser tool structures your thinking. It does not send your information anywhere or perform AI analysis.
Save as Skill converts the model into portable Markdown instructions that teach an AI how to spot miscalibrated certainty, ask the right follow-up questions, and avoid the common mistakes above.
Save the skills you find useful and organize them into your own reusable library. The Markdown is intentionally model-neutral, so it can be used wherever an AI system accepts uploaded files, project knowledge, custom instructions, or reusable text context. Platform capabilities differ, so this is a portable skill file rather than a claim of one-click installation.
Describe a specific estimate, prediction, or confident statement you are working through. This builds a separate, personalized prompt that asks an AI to apply overconfidence bias directly to your situation, distinct from the prompt builder above.
No. It describes a specific, checkable gap between stated certainty and actual accuracy, not a claim that judgment is generally untrustworthy. The useful move is checking particular estimates against a track record, not adopting blanket self-doubt.
Stuart Oskamp's 1965 study found clinical psychologists' confidence rose steadily as they received more case details, while their diagnostic accuracy stayed flat. Extra information can create a feeling of mastery over a case without actually resolving the genuine uncertainty involved, which inflates stated confidence faster than it inflates real accuracy.
Before stating a confidence level or range, pull up your last several similar estimates and check how often the true answer actually landed where you said it would. If a "90% confident" label has only been right half the time, treat that gap, not the current estimate alone, as the thing worth fixing.