# Base Rate Neglect: Why a Positive Test Often Isn’t What It Seems

Numbers · 6 min read · updated 2026-10-29

> A 99% accurate test says you have a rare disease. Are you 99% likely to have it? Rarely. Base rate neglect, worked out with whole people instead of formulas.

## The short answer

Base rate neglect is judging how likely something is from the evidence in front of you while ignoring how common it was to begin with. A 99% accurate test for a condition that affects 1 in 1,000 people produces about ten false alarms for every real case, so a positive result means roughly a 9% chance, not 99%. To work it out, picture a crowd of people and count the true and false positives.

Here’s a question that most people get wrong, including a lot of doctors.

A disease affects 1 person in 1,000. A test for it catches every case, and wrongly flags 5% of healthy people. You take the test and it comes back positive. What’s the chance you have the disease?

In 1978, Ward Casscells and colleagues put a version of this question to staff and students at Harvard Medical School. Almost half answered 95%, the number the test’s accuracy seems to suggest. The right answer is about 2%. Only about one in five got it.

And the problem isn’t reasoning ability. It’s that the question is posed in a way that makes the wrong answer feel obvious.

## Count people, not percentages

Let’s redo it with a crowd. Imagine 1,000 people take the test.

- **One** of them really has the disease, and the test catches it.
- The other **999** are healthy. The test wrongly flags 5% of them: about **50** people.

So about 51 people test positive, and exactly one of them has the disease. If you’re in the positive group, your chance of being the real case is 1 in 51, around 2%.

The 5% false positive rate sounds small. But 5% of a huge healthy group is a lot more people than 100% of a tiny sick one. That’s the mechanism, and the toy above lets you see it: slide the rarity and the false alarms, and watch how many of the highlighted people are real.

## The three numbers you need

Almost any test question needs three pieces of information, and the usual mistake is to use only one.

1. **The base rate:** how common is the condition among the people being tested? (1 in 1,000.)
2. **The hit rate, or sensitivity:** how many real cases does the test catch? (All of them, here.)
3. **The false alarm rate:** how many healthy people does it wrongly flag? (5%.)

The headline figure, “99% accurate”, usually describes only the second or third number. It doesn’t tell you the first. And the first is what decides how much a positive result means.

**The rule of thumb**

When a condition is rare, most positive results are false alarms, however good the test sounds.

## Decoding the jargon on the lab report

Test results come in words that sound alike and mean different things. Here’s the translation:

| The term | What it answers | Our example |
|---|---|---|
| Sensitivity | Of people who have it, how many does the test catch? | 100% |
| Specificity | Of people who don’t, how many does it correctly clear? | 95% |
| Positive predictive value | Of people who test positive, how many really have it? | about 2% |
| Prevalence (the base rate) | How common is it in the group tested? | 1 in 1,000 |

The first two are properties of the test. The third is a property of the test *and* the group, which is why it can swing from 2% to 90% without the test changing at all. When someone says a test is “99% accurate”, they’re nearly always quoting one of the first two. The number you need on a hospital letter is the third.

## It’s the same mistake in the other direction

Kahneman and Tversky gave this *base rate neglect* its name and showed it’s a general feature of how people judge. Told that Steve is shy, tidy and loves detail, people say he’s more likely to be a librarian than a farmer, forgetting that there are many more farmers than librarians. The description fits a librarian better, and that’s all they weigh.

The base rate pulls the other way when the condition is common. If half the people being tested have a condition, a positive result is strong evidence. If one in a thousand does, it’s weak. The test is identical in both cases. The meaning of its result is not.

*(An interactive illustration belongs here. Try it on the page: https://cognosc.com/blog/base-rate-neglect)*
## Where it has done real damage

**Breast screening.** David Eddy asked physicians in 1982 about a mammogram result. About 1 in 100 women in the group has breast cancer. The mammogram catches around 80% of cancers and wrongly flags about 10% of healthy women. A woman has a positive result: what’s the chance she has cancer? Most physicians he asked said around 75%. The answer is about 8%. Of the 1,000 women, 8 with cancer test positive, and about 99 without it do too. That’s a very different conversation to have with a frightened patient.

**The Sally Clark case.** In 1999, a British solicitor named Sally Clark was convicted of murdering her two infant sons, who had both died suddenly. At trial, an expert told the jury the chance of two cot deaths in one family like hers was 1 in 73 million. That figure was wrong for a technical reason, since it assumed the two deaths were independent. But there was a deeper error, which is the one in this article. Even a very small chance of a coincidence has to be set against the tiny chance of a mother murdering two children. Both are rare. The question is which is rarer. The Royal Statistical Society protested, and in 2003 her convictions were overturned.

**Everyday screening.** Airport scanners, fraud alerts on your card, drug tests, plagiarism checkers and the tools that claim to detect AI writing all share the same structure: an imperfect test aimed at something rare. If 1 in 1,000 passengers is a threat and the scanner wrongly flags 1 in 100, about ten innocent people are stopped for each real threat. That’s not a failing of the technology. It’s arithmetic.

## Why “natural frequencies” fix it

If the mistake is so common, you might think people are bad at probability. The more hopeful finding is that they aren’t, as long as the numbers are given the right way.

Gerd Gigerenzer and Ulrich Hoffrage found in 1995 that when the same problem is stated with counts of people, “10 in 1,000 have it, 8 of those test positive, 95 of the others do too”, accuracy jumps from around one in six to nearly half. Later work found the same effect for doctors. Percentages hide the structure of the problem. People hide it in a crowd and the answer falls out.

That’s why the method is always the same: **start with 1,000 people.** Split them into those who have the thing and those who don’t. Apply the test to each group. Count how many positives of each kind you get. The answer is the share of positives that are real.

*(An interactive illustration belongs here. Try it on the page: https://cognosc.com/blog/base-rate-neglect)*
## How to use it in real life

Whenever you hear that someone has tested positive for something, or that a system flagged them, ask:

- **How common is this to start with?** For a rare thing, assume a positive is probably a false alarm until it’s confirmed.
- **What are the false alarm rates?** Not just “how accurate” but “how often does it flag people who don’t have it”.
- **What happens next?** In good medicine a positive screening result leads to a more specific test, because the first test was never meant to be the verdict.
- **Who was tested?** The same test is far more informative in people with symptoms or a family history than in the general public, because the base rate in that group is higher.

None of this means distrusting tests. A positive result does change the odds, often a lot: in the 1,000-person example it moved the chance from 0.1% to 2%, a twentyfold jump. It just doesn’t take the odds to near-certainty, and the shortfall depends on how rare the thing was.

## The takeaway

When something rare is flagged by an imperfect test, count the false alarms before you believe the flag. It sounds like a minor statistical point. Applied to health, justice and security, it’s often the difference between a calm “let’s check” and a life-changing mistake.

## What to remember

- A test’s accuracy and the chance a positive result is correct are two different numbers.
- When a condition is rare, false alarms from the many healthy people outnumber real cases.
- Think in whole people: 1,000 people, how many have it, how many test positive of each group.
- Ask three things of any test: how common is it, how many cases does it catch, how often does it cry wolf?
- Doctors get this wrong too, and the fix is the format of the numbers, not the intelligence of the person.

## Questions people ask

### What is base rate neglect?

Ignoring how common something is when judging how likely it is given new evidence. People focus on how well the evidence fits and forget the starting odds, which can make a rare thing look likely just because a test, a description or a clue points at it.

### Why does a positive test for a rare disease often mean you’re probably fine?

Because there are so many more healthy people than sick ones. Even a small false positive rate, applied to thousands of healthy people, produces more false alarms than the real cases the test catches. Whether a positive is trustworthy depends on how common the condition is.

### What’s the difference between sensitivity, specificity and positive predictive value?

Sensitivity is how many real cases a test catches. Specificity is how many healthy people it correctly clears. Positive predictive value is the chance that a positive result is real, and it depends on both and on how common the condition is, which is why it’s the one people get wrong.

### What are natural frequencies?

A way of stating probabilities as counts of people, like “10 out of 1,000 have it, 8 of them test positive”, instead of percentages. Gigerenzer and Hoffrage showed that people, including doctors, reason far more accurately with them.

### Does base rate neglect apply outside medicine?

Everywhere there’s a filter that’s imperfect and a target that’s rare: spam filters, airport security, drug tests, fraud alerts, plagiarism and AI-writing detectors, and DNA evidence. Whenever a rare thing is flagged by an imperfect test, count the false alarms.

## Sources and further reading

- Casscells, W., Schoenberger, A. and Graboys, T. B. (1978). Interpretation by physicians of clinical laboratory results. New England Journal of Medicine, 299(18), 999–1001.
- Eddy, D. M. (1982). Probabilistic reasoning in clinical medicine: Problems and opportunities. In Kahneman, Slovic and Tversky (eds), Judgment under Uncertainty. Cambridge University Press.
- Gigerenzer, G. and Hoffrage, U. (1995). How to improve Bayesian reasoning without instruction: Frequency formats. Psychological Review, 102(4), 684–704.
- Gigerenzer, G., Gaissmaier, W., Kurz-Milcke, E., Schwartz, L. M. and Woloshin, S. (2007). Helping doctors and patients make sense of health statistics. Psychological Science in the Public Interest, 8(2), 53–96.
- Royal Statistical Society (2001). Statement regarding statistical issues in the Sally Clark case.
- Tversky, A. and Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases. Science, 185(4157), 1124–1131.

---
Canonical page: https://cognosc.com/blog/base-rate-neglect
From Cognosc (https://cognosc.com): free tests, articles and an AI tutor that has you reason things through.