Overconfidence Bias: Examples & How to Calibrate

What is overconfidence bias?

Overconfidence bias is a gap between how sure you feel and how often you’re actually right. When that gap is consistent and in the “too sure” direction, you have a bias, not just a bad day.

Don Moore and Paul Healy’s 2008 paper “The Trouble with Overconfidence” showed that the word covers three different things, which often get mixed up:

1. Overestimation: believing your performance, ability or control is higher than it is. “I’ll finish this in a week.”
2. Overplacement: believing you’re better than others. “I’m a better driver than most people on this road.”
3. Overprecision: being too certain that your beliefs are accurate. Your ranges are too narrow and your “definitely” comes too fast.

Overprecision is the most persistent of the three. The other two can flip: on very hard tasks, people often underestimate themselves, and on tasks they find hard, they may rate themselves below average. Overprecision, being too sure of your estimates, shows up almost everywhere it has been tested.

Why does overconfidence persist when it costs so much? A few reasons work together:

• Feedback is slow or missing. Most of your predictions are never checked. Nobody tells you that your “definitely” was wrong four times this year.
• Memory is kind to you. You remember hits more vividly than misses, and you reinterpret misses as “basically right.”
• Confidence is rewarded socially. In meetings, interviews and pitches, the confident answer often wins the room, even when the hedged one was more accurate.
• You see your plan, not the world. When you estimate, you picture the steps you’ll take. You don’t picture the delays, the illness, the supplier who goes quiet. That inside view leaves out most of what goes wrong.

None of these are character flaws. They are features of how information reaches you, which is why the fix is a system, not a pep talk.

Where it comes from

Overconfidence research grew out of the study of calibration: how well people’s stated confidence matches their hit rate.

Calibration and the too-narrow interval

In a classic method developed in the late 1960s and early 1970s (Marc Alpert and Howard Raiffa’s work is the best-known example), people are asked to give ranges they are, say, 98% sure contain the true answer to trivia questions: the length of the Nile, a city’s population. If people were well calibrated, the true answer would fall outside the range about 2% of the time. In study after study, it falls outside far more often, often a large share of the time. People feel 98% sure while their ranges behave like coin flips with extra steps.

Sarah Lichtenstein, Baruch Fischhoff and Lawrence Phillips reviewed this calibration work in the early 1980s and found the same broad pattern across many studies: people tend to be overconfident, especially on hard questions.

Better than average

In a well-known 1981 study, Ola Svenson asked groups of students in the US and Sweden to compare their driving skill and safety with the other people in the room. Most rated themselves above the median, which can’t be true for most of a group. Similar “better than average” findings have shown up for many traits people care about.

Forecasting and expert judgment

Philip Tetlock spent years collecting predictions from political and economic experts, reported in Expert Political Judgment (2005). Many experts were poorly calibrated, and the most confident, single-big-idea thinkers often did worse. His later forecasting tournaments, described in Superforecasting (2015, with Dan Gardner), found that a small group of forecasters could be trained to be well calibrated, partly by keeping score.

Some professionals are known for good calibration. Weather forecasters are the usual example: when they say 70% chance of rain, it rains on roughly 70% of those days. They get fast, frequent, honest feedback. That turns out to be the key ingredient.

Money

In studies of brokerage account data, Brad Barber and Terrance Odean found that households that traded the most earned lower net returns, consistent with overconfidence in their ability to pick winners. In a 2001 study, they found men traded more than women and the extra trading cost them returns.

How to beat overconfidence bias (how to calibrate)

You beat overconfidence by turning “I’m sure” into a number, then keeping score. Calibration is a trainable skill. Six steps:

1. Put a percentage on predictions. Not “this will work,” but “70% this ships by June.” Probabilistic thinking is the fix because a number can be checked; a feeling can’t.
2. Use ranges, and make them wider than feels comfortable. For any estimate (cost, time, revenue), give a low and a high you’d be 90% sure contain the answer. If your ranges keep missing, widen them.
3. Start from the base rate. Before thinking about why your project is special, ask how long similar projects usually take or how often similar ventures succeed. Then adjust. This counters overestimation and the planning fallacy.
4. Run a pre-mortem. Imagine it’s a year later and the plan failed. List the most likely reasons. This tends to surface risks your confidence was hiding.
5. Build a margin of safety. When stakes are high, plan as if you’re less right than you feel. A margin of safety turns overconfidence from a disaster into an inconvenience.
6. Keep score. Log predictions with their confidence in a decision journal. Every few months, group them by confidence level and check the hit rate. If your 90% calls come true 65% of the time, you’ve learned the most useful number about yourself you’ll ever find.

Step 6 is the one that changes behavior. Without a record, hindsight bias rewrites your memory so you “knew it all along,” and calibration never improves.

3 real-world examples

These are hypothetical examples of common patterns.

Business

Say a startup founder tells investors the product will launch in four months with 10,000 users by year end. Both are point estimates with no range, a sign of overprecision. A calibrated version: “We think launch is 60% likely by month four and 90% likely by month seven. Our year-end user range is 2,000 to 12,000, depending on whether the channel partnership works.” This sounds less exciting. It is also far more useful for planning cash.

Personal life

Say you’re sure you can renovate the bathroom yourself over two weekends. You’ve watched the videos, you’re handy, you’ve done smaller jobs. That’s overestimation (of your speed) plus overprecision (no room for surprises). A quick calibration check: how long did your last three DIY projects take compared with your plan? If they ran double, assume this one will too, and decide whether it’s still worth doing yourself.

Investing

Say an investor puts 40% of their savings into one stock because they’re “certain” about the company. Even if their analysis is good, certainty is rarely warranted in markets, where many outcomes depend on things no one can know. A calibrated investor might say “I think there’s a 65% chance this outperforms,” which naturally argues for a smaller position. The size of a bet should match the confidence you can justify, not the confidence you feel.

Common mistakes

• Treating overconfidence as arrogance. Quiet, modest people can be badly overprecise. It’s about how narrow your estimates are, not how loud you are.
• Fixing it with blanket self-doubt. Underconfidence costs you too: missed opportunities, slow decisions. The goal is calibration, not humility as a pose.
• Ignoring it in experts. Experts are often better on average and still overprecise, especially outside their circle of competence.
• Never checking. Without a scorecard, you can’t tell confidence from accuracy, and you’ll remember your hits more than your misses.
• Confusing it with the Dunning-Kruger effect. That’s one specific pattern within overconfidence; see the Dunning-Kruger effect page.

Overconfidence bias vs Dunning-Kruger effect

The Mentarium take

Overconfidence is the bias most worth fixing because it multiplies all the others. It’s on three of the four Bias Radar lists (Career, Money and Business), and it’s the reason anchoring, confirmation bias and the planning fallacy do so much damage: you don’t just make the error, you feel sure about it.

The cure is boring and it works: write the probability down. That’s the Decide step of the Mentarium Decision Loop (Frame → Map → Stress-test → Decide → Review): commit to a choice, and record your prediction and your confidence %. Then the Review step, on a set date, compares that number with what happened. Weather forecasters are calibrated not because they’re humbler than the rest of us, but because they get scored every day.

You don’t need daily scoring. Twenty logged predictions over a few months are enough to show whether your 80% means 80%. Most people who try it discover their certainty is worth less than they thought, and that discovery, made once with real data, does more than any amount of advice to be humble.