Overconfidence Bias: A 7-Question Calibration Check Before You Commit

A few years ago I started a small habit. Whenever I made a call at the studio, I wrote the prediction in a notebook with a number beside it. "We will win the tender: 80%." "The client signs before Christmas: 90%." At the end of the year I counted. Of the predictions I had marked 90%, roughly six in ten came true. I had been sure, and sure turned out to be different from right.

That gap has a name: overconfidence bias, the tendency for our confidence to run ahead of our accuracy.

In 1982 Sarah Lichtenstein, Baruch Fischhoff and Lawrence Phillips reviewed the research on calibration up to 1980. Being well calibrated means the things you call 70% likely happen about 70% of the time. Across many studies of general-knowledge questions, people's stated confidence was higher than their hit rate, and the gap was widest on hard questions. On very easy questions it could shrink or even reverse.

Wait a minute. If people are sometimes underconfident, is overconfidence even one thing? Don Moore and Paul Healy asked exactly that in 2008. They separated three different beliefs that research had lumped together. Overestimation is thinking you will do better than you actually do. Overplacement is wrongly believing you are better than other people. Being better is not the error. Misjudging where you stand is. Overprecision is being too certain that your estimate is right, drawing your range of possibilities too narrow. On difficult tasks, they found, people tend to overestimate their own performance yet believe they are worse than others. On easy tasks it flips. Overprecision appeared to be the most persistent of the three.

That nuance matters for a real decision. "Am I overconfident?" is too vague to answer. "Which of these three am I doing right now?" is a question you can actually work with.

There is also good news, with a caveat. In 1980 Asher Koriat, Sarah Lichtenstein and Baruch Fischhoff gave people two-choice general-knowledge questions. After choosing an answer, some wrote one reason it might be wrong, and their confidence came closer to their accuracy. Writing a reason in favour changed nothing, perhaps because we already think of those. Writing one of each made only a small difference, too small to be reliable. These were trivia questions, not life decisions, so treat arguing against yourself as a promising habit rather than a proven cure.

Overconfidence sits close to a few neighbours, so it helps to draw the lines. Trusting an expert's intuition is about when someone else's judgment has earned its weight. Overconfidence is about your own confidence level, before you commit. The work of mapping what you genuinely know is the Circle of Competence. Calibration asks a sharper follow-up: inside or outside that circle, how sure should you be?

Seven questions to run before you commit to something you feel certain about:

1. How sure am I, as a number? Write a percentage. "Pretty confident" hides more than it says.
2. Sure of what, exactly? Name the claim. Is it about how well I will do, how I compare with others, or how precise my estimate is?
3. What range would honestly surprise me? Give a low and a high you would be genuinely shocked to fall outside. If the range feels comfortable, widen it.
4. What is the strongest reason I might be wrong? Write it down on its own, before any reasons in favour. This is the step closest to the research.
5. How often does this work out for people in my position? Find the base rate, then ask what makes you truly different from it.
6. What is the smallest test I can run before committing fully? A pilot, a trial month, a conversation. Something reversible.
7. What prediction will I record, and when will I check it? Put the number and the date in a decision journal. Over months, the record shows whether your 80% really means 80%.

Here is the tender, run through it. How sure: 80%. Sure of what: that our proposal would beat the other agencies, a claim about where I stood, which is exactly where overplacement hides. The range: I could not honestly rule out coming third. The strongest reason I might be wrong: two competitors had worked with this client before, and we had never delivered at this scale. The base rate: I had no record of tenders like this one, so I started from a rough assumption. Eight agencies had bid. If we were all roughly equal, that meant one chance in eight, and nothing I knew made us six times likelier than an average bidder. The smallest test: a short call to ask what the client valued most, which I had skipped. The prediction I recorded: 35%, a judgment call rather than a calculation, checked on the announcement date. We came second. I was disappointed, and also, for once, not surprised.

But hold on. Confidence is not the enemy. Some decisions need conviction to carry them through the hard middle, and a well-calibrated 90% is a fine reason to act. The point is not to doubt everything. It is to let the number you feel and the number you would bet become the same number. For a wider sweep of related traps, the cognitive bias checklist is a good companion.

This is where ClearMind fits, quietly. It will not tell you how confident you should be. It works as a mirror: its Circle of Competence walkthrough asks what you genuinely know, where your knowledge ends and who knows what you don't, then reflects back the patterns and blind spots in your answers. The worksheet above is yours to run alongside it. The judgment, and the decision, stay with you.

Confucius said that to know what you know, and to know what you do not know, is true knowledge. Calibration is that old idea with a pencil and a number attached.