Base-rate neglect

What it is

Base-rate neglect is the tendency to ignore how common something is in general (the base rate) when judging a particular case. We focus on the specific evidence in front of us, such as a description or a test result, and underweight the background odds.

An everyday example

A bank’s fraud alert flags Priya’s card payment. Suppose, for illustration, that 1 in 1,000 payments is fraudulent, the system catches 99% of fraud, and it wrongly flags 2% of honest payments.

Most people guess the flagged payment is very likely fraud. Count it out with 100,000 payments: about 100 are fraud, and 99 of those get flagged. Of the 99,900 honest ones, about 2,000 get flagged too. So only about 1 in 21 flagged payments is actually fraud. The low base rate dominates.

Why it happens

Specific evidence feels relevant and causal; base rates feel abstract and statistical. Kahneman and Tversky (1973) argued that people often judge by representativeness: how well a case matches a category, rather than how likely that category is to begin with.

Probabilities and percentages also make the math hard. Combining a 1-in-1,000 base rate with 99% and 2% in your head is genuinely difficult.

How to counter it

• Ask “out of how many?” Before judging a case, find out how common the outcome is in the relevant group.
• Turn percentages into counts. Imagine 1,000 or 100,000 people or events and count who ends up where.
• Separate the two questions. “How well does this evidence fit?” is different from “How likely was this to begin with?”
• Be extra careful with rare events. When something is rare, even accurate tests produce many false alarms.

What the research shows

In a classic study, people read brief personality sketches and estimated whether each person was an engineer or a lawyer. Some were told the group held 70% engineers, others 30%. Their estimates barely changed with this information; they relied on how “engineer-like” the description sounded (Kahneman & Tversky, 1973).

The effect is not fixed, though. When the same kind of problem is presented as natural frequencies (“10 out of 1,000”) instead of probabilities, many more people reason correctly without any training (Gigerenzer & Hoffrage, 1995). This suggests that part of the problem lies in how information is presented, not only in how we think.