The 90% test: why 'I'm sure' is right far less often than it feels

When people say they are 90% sure a range contains the answer, it usually contains the answer far less often than nine times in ten. In one experiment, 334 people gave 90% ranges for the years twentieth-century US presidents were first elected. Those ranges held the right year about 54% of the time.

The gap has a name: overprecision, being surer about your numbers than your knowledge allows. The quiz below lets you see your own version in ten minutes. The answers are at the bottom, each with its source.

How does the 90% test work?

For each question, write down a low number and a high number. Choose them so that you are 90% sure the true answer lies between them. Do not look anything up.

A 90% range is a bet. You should be willing to risk nine to win one that the answer is inside. If your ranges are honest, you will miss about one question in ten.

The quiz: ten questions, ten ranges

#	Question	My low	My high
1	In what year did the Wright brothers make the first powered airplane flights?
2	In what year was Isaac Newton’s Principia first published?
3	In what year did the United Nations officially come into existence?
4	In what year did the first iPhone go on sale in the United States?
5	How tall is the Eiffel Tower today, antennas included, in meters?
6	How tall is Mount Everest, in meters?
7	What is the average distance from the Earth to the Moon, in kilometers?
8	How fast does light travel in a vacuum, in kilometers per second?
9	How many member states does the United Nations have?
10	How wide is Rembrandt’s painting The Night Watch, in centimeters?

Write all ten ranges before you read on. The research comes next, so the answers stay out of sight for a while.

What has research found about overprecision?

Overconfidence comes in three forms, as Don Moore and Paul Healy laid out in 2008: thinking you did better than you did, thinking you did better than others, and being too sure your beliefs are precise. Their review concluded that “Overprecision appears to be more persistent than either of the other 2 types of overconfidence.” The 90% test measures that third form.

Ranges come out far too narrow. Jack Soll and Joshua Klayman (2004) found subjective ranges were systematically too narrow for what people actually knew, “sometimes only 40% as large as necessary to be well calibrated.” How much depended heavily on how the question was asked.

Asking for 90% barely changes the width. Karl Halvor Teigen and Magne Jørgensen (2005) ran five experiments. Ranges that people were told to make at 90% could come with a felt confidence of 50% or less, and interval width stayed roughly the same across very different instructions. Their paper’s title sums it up: “When 90% confidence intervals are 50% certain.”

Big samples show the same tilt. In a quiz of 50 true-or-false questions run by Andrew and Michael Mauboussin, more than 11,000 people took part. Their average confidence was 70%. Their average score was 60%.

Experts are not exempt, but method helps. Andrew Speirs-Bridge and colleagues (2010) asked epidemiologists, public health experts and ecologists for 80% ranges. Across their three studies, those ranges held the answer 68.1% of the time: better than older studies, still short of 80%. In two of the studies, a four-step way of asking reduced overconfidence more than the usual three-point version.

How do you score your quiz?

Check your ranges against the answers below and count the hits.

If every range really were a 90% range, you would score 8 or more about 93 times in 100. A score of 7 or lower would happen by bad luck only about 7 times in 100, and 6 or lower about once in 100. That is plain probability arithmetic, not a study result.

So a 6 is a strong hint that your ranges are too narrow. A 9 on ten questions proves little. Ten is a small sample: treat your score as a lead to follow, not a verdict on you.

How do you widen a range honestly?

“Honestly” matters. A range from zero to a billion never misses and tells nobody anything. A useful range can still miss. It just misses about as often as you said it would.

Four techniques. The first three were tested in the studies above; the fourth is a rule of thumb from The Calibrated Mind:

1. Ask for the two ends separately. First: “What number am I 95% sure the answer is above?” Then, as a separate act: “What number am I 95% sure it is below?” Two 95% edges make one 90% range. Teigen and Jørgensen found that separate “more than” and “less than” questions widen ranges.
2. Guess first, rate second. Write the low end, the high end and a best guess, and only then rate how confident you are that the answer is inside. That is the four-step method Speirs-Bridge’s team tested.
3. Spread your chances across the whole scale. Split the possible answers into bins, for example each decade of a century, and give every bin a probability, all adding up to 100%. In Uriel Haran, Don Moore and Carey Morewedge’s third experiment, this method held the right answer 76.91% of the time, against 54.34% for plain 90% ranges.
4. Run the bet test. Would you risk €9 to win €1 that the answer sits inside your range? If not, widen it until you would.

How does CALIBRATE use the 90% test?

CALIBRATE is a nine-step system for matching how sure you are to what your evidence can carry. It comes from The Calibrated Mind by Len P. van der Hof, available in English and Dutch. You do not need the book to use it.

1. Calibrate confidence. Turn a feeling of certainty into a number you can check later.
2. Assess assumptions. List the hidden premises under your judgment and test the weakest one.
3. Lean on probabilities. Replace words like “likely” with numbers and ranges.
4. Investigate alternatives. Build the strongest case against your own view.
5. Build updating rules. Decide in advance how much new evidence should move you.
6. Record and review. Keep a prediction journal and score it.
7. Act with appropriate uncertainty. Set a threshold for acting instead of waiting for certainty.
8. Teach and extend. Explain your numbers to someone else. The gaps show fast.
9. Evolve your model. Review on a rhythm, because calibration drifts.

The quiz is step 1, Calibrate confidence, in miniature. The book’s bet test turns odds into a floor on your confidence: risking one to win one claims at least 50%, three to win one at least 75%, nine to win one 90%. If you would not take nine to one on your range, you were never at 90%.

Here is step 1 on a hypothetical work estimate. Your team asks how long a website rebuild will take, and your first answer is “eight weeks.”

• Low end, asked on its own: “I’m 95% sure it takes more than six weeks.”
• High end, asked separately: “I’m 95% sure it takes less than sixteen weeks.”
• Bet test: would you risk €9 to win €1 on six to sixteen weeks? You would. Write it down with today’s date.

Six to sixteen weeks says far more than “eight weeks” did, and the width is the point: it is information. It tells the people planning around you how much weight the estimate can bear. Step 6, Record and review, then scores the range when the project ends. Calibration techniques for founders shows how to turn such scores into a weekly habit, and what calibration means explains the idea underneath.

Answers

1. 1903. The first flights took place on 17 December 1903 at Kitty Hawk, North Carolina (Smithsonian National Air and Space Museum).
2. 1687. The first edition of the Principia was printed in 1687 (The Royal Society).
3. 1945. The United Nations officially came into existence on 24 October 1945 (United Nations).
4. 2007. The iPhone went on sale in the US on 29 June 2007 (Apple Newsroom).
5. About 330 meters. The official site lists a current height of 1,083 feet, antennas included (Tour Eiffel).
6. 8,848.86 meters. The height Nepal and China announced jointly on 8 December 2020 (The Kathmandu Post).
7. 384,400 kilometers. The average Earth to Moon distance (NASA).
8. 299,792.458 kilometers per second. Exact by definition of the SI units (BIPM).
9. 193. The count since South Sudan joined in 2011 (United Nations).
10. 453.5 centimeters. The painting is 379.5 cm high and 453.5 cm wide (Rijksmuseum).

Try this today

Give the quiz to one colleague and compare scores. It takes ten minutes, and the conversation afterward is the useful part: whose ranges missed, and on which kind of question.

Then write your next three work estimates as 90% ranges using the two-question method. Put each resolution date in your calendar. When they resolve, count the hits. That small record will teach you more about your own ranges than any quiz.

Cite this:The 90% test: why 'I'm sure' is right far less often than it feels.Len P. van der Hof. https://lenvanderhof.com/en/blog/overconfidence-90-percent-test/ · Published 8 October 2026.