Base Rate Fallacy

The glitch, explained plainly.

Imagine there are 100 dogs in a park — 95 are friendly and 5 are mean. A dog walks up to you wearing a spiked collar and growling a little. Your brain screams 'Mean dog!' because of the scary details. But if you remembered that 95 out of 100 dogs here are friendly, you'd realize it's still way more likely the dog is just a friendly one having a bad moment. The Base Rate Fallacy is when you forget about the 95-out-of-100 part because the spiked collar story is more interesting.

The Base Rate Fallacy occurs when individuals presented with both general statistical information (how prevalent something is in a population) and specific individuating information (details about a particular case) systematically overweight the specific details and underweight or entirely ignore the base rate. This leads to probability estimates that violate Bayes' theorem, the normative framework for updating beliefs given new evidence. The fallacy is especially pernicious in diagnostic contexts — medical testing, criminal forensics, and hiring algorithms — where a low base rate of the target condition means that even highly accurate tests produce far more false positives than true positives. The bias is driven by the brain's preference for narrative and concrete detail over abstract statistical information, making vivid descriptions feel more informative than they actually are.

Where it shows up.

1. 01 A company's drug screening test is 98% accurate. When an employee tests positive, the HR director immediately moves to terminate them. A colleague points out that only 0.5% of employees actually use drugs, meaning most positive results at this company are false positives — but the HR director dismisses this, saying 'The test is 98% accurate, that speaks for itself.'
2. 02 A venture capitalist reads a glowing profile of a startup founder who dropped out of Stanford, codes in three languages, and previously worked at Google. She immediately rates the startup as highly likely to succeed, even though she knows that over 90% of startups fail regardless of the founder's impressive résumé.
3. 03 A security system at an airport correctly identifies prohibited items 99.5% of the time. When the system flags a passenger's bag, the security officer treats it as near-certain that a prohibited item is present. He doesn't consider that millions of bags pass through daily, the vast majority containing nothing prohibited, meaning most flags are false alarms.
4. 04 A jury is told that DNA found at a crime scene matches the defendant with a 1-in-a-million accuracy. The prosecutor argues this virtually proves guilt. However, in a city of 10 million people, roughly 10 people would match — making the DNA evidence far less conclusive than the jury assumes, since they never considered the size of the population that could have been tested.
5. 05 A teacher notices that a student who is exceptionally neat, wears glasses, and reads voraciously fits the profile of a future valedictorian. She recommends the student for an advanced honors track, without considering that only 1% of students become valedictorian, while many students who are neat, bespectacled, and bookish do not — the description simply matches a stereotype, not a statistical likelihood.

Where it shows up at work.

The same glitch looks different depending on the terrain. Finance, medicine, a relationship, a team — same mechanism, different costume.

In history.

• The false positive paradox in post-9/11 mass surveillance programs: analysts estimated that data-mining algorithms would generate tens of thousands to billions of false positives for every true terrorist identified, because the base rate of terrorism is extraordinarily low.
• The O.J. Simpson trial (1995): defense attorney Alan Dershowitz argued on television that only 0.1% of men who batter their partners go on to murder them, using base rate reasoning to counter prosecution claims — though critics noted this itself was a misapplication of conditional probability.
• David Eddy's 1982 study found that fewer than 5% of physicians correctly estimated the probability of breast cancer given a positive mammogram, with most confusing the test's sensitivity with the posterior probability — a textbook case of base rate neglect in clinical medicine.