;colony/science  / Mathematics, Made Visual  / How should evidence change your mind?
Mathematics, Made Visual

How should evidence change your mind?

A test that is right 99 times out of 100 can still be wrong about you most of the time. The arithmetic is brutal and exact.

Plate 86 — When a test can lie Bayes' theorem · base-rate neglect
Make the disease rare and a 99%-accurate test still flags mostly false alarms.
Predict firstWhen you make the disease rarer with the sliders, what happens to the false alarms?
truly sick & flagged (10)false alarm (10)all clear
PLATE 86 · WHEN A TEST CAN LIE
How common the disease is 1%
10 of 1000 people actually have it.
Test accuracy 99%
Catches the sick and clears the healthy this often.
Positive → really sick
50%
Flagged in total
20/ 1000
Say only 10 in 1000 people are really sick. A very good test still gets a few healthy people wrong — and there are so many healthy people that those mistakes pile up. So out of everyone the test flags, only about 50 in 100 are truly sick. A red flag is a reason to check again, not to panic.
Try with the plate
  • Drag the sliders and watch false alarms swamp the real cases.
  • Make the disease rarer and read how low a positive result drops.

Evidence should update your prior belief, not replace it. Using Bayes' theorem, you start with how common something is, then let the new result shift that figure. When a condition is rare, even an accurate test produces mostly false alarms, so a positive result is a nudge to look closer, not the final word.

The short answer

Imagine a disease that only 1 in 1000 people actually have, and a test that is right almost every time. Test all 1000 people: the test correctly flags the 1 sick person, but it also slips up on about 10 of the 999 healthy ones. So 11 people get a scary red result, and only 1 of them is truly sick. A positive test is a nudge to look closer, not the final word. Drag the sliders in the simulator and watch the false alarms swamp the real cases.

The common mix-up

Most people think a 99% accurate test that comes back positive means a 99% chance of being sick. In fact for a disease affecting 1 in 1000, only about 1 in 11 positives is truly ill, because false alarms drawn from the huge healthy crowd swamp the few real cases.

What's actually happening

Here is a question that fools most people, including many doctors. A disease affects 1 in 1000. A test for it is 99% accurate, meaning it correctly catches the sick and correctly clears the healthy 99% of the time. You take the test. It comes back positive. How worried should you be? The gut answer is 99% worried. The real answer is about 9%.

The trap is forgetting how few people are actually sick. Run the numbers on 1000 people. Only 1 is genuinely sick, and the test flags that person. But there are 999 healthy people, and a 1% error rate on 999 means the test falsely flags about 10 of them too. So 11 people walk away with a positive result, and only 1 is truly ill. Your real chance of being sick, given that red flag, is 1 out of 11, roughly 9%. The test did not lie about its accuracy. The rarity of the disease did the damage, because a tiny slice of a huge healthy crowd still outnumbers the genuinely sick.

The fix is Bayes’ theorem, named after the Reverend Thomas Bayes, whose work was published in 1763 after his death. It says evidence should not replace your prior belief, it should update it. Start with how common the thing is, then let the test shift that number, and the answer falls out honestly. This is why doctors confirm a rare diagnosis with a second, independent test, why a single airport scanner alarm rarely means a real threat, and why a strong-looking result on a rare claim deserves a calm second look rather than a panic. Good thinking is not about how convincing the new evidence feels. It is about where you were standing before it arrived.

Remember this

Evidence should update your prior belief, not replace it, because on a rare condition even an accurate test is mostly false alarms.

Try it at home Run the 1000-people test
  1. 1Draw a grid of 1000 squares. Colour 1 of them as the truly sick person (a 1-in-1000 disease).
  2. 2Now mark the false alarms: about 1% of the other 999, so roughly 10 healthy squares, also get flagged by the test.
  3. 3Count the flagged squares: 11 in total, only 1 truly sick. A positive result means about a 1-in-11 real chance. You have just done Bayes by hand.

Common questions

How can a 99% accurate test be wrong about me most of the time?

If a disease affects only 1 in 1000, testing 1000 people flags the 1 sick person but also about 10 healthy ones. So 11 get a positive result and only 1 is truly ill, a real chance of roughly 9%.

What is base-rate neglect?

It is the common mistake of ignoring how rare something is and trusting the test alone. A tiny slice of a huge healthy crowd still outnumbers the genuinely sick, so forgetting the base rate badly overstates your risk.

Why do doctors confirm a rare diagnosis with a second test?

Because a single positive on a rare condition is mostly false alarms. An independent second test shifts the odds again and separates the true cases from the false ones, which is Bayes updating in practice.

Built & checked by Nilesh Singh · how this is made · last updated June 2026