Base Rate Fallacy

The base rate fallacy, also called base rate neglect, is the tendency to give too little weight to the base rate, the background frequency of an outcome in a population, and too much weight to information about the particular case. Daniel Kahneman and Amos Tversky described the mechanism in 1973: intuitive predictions follow the representativeness heuristic, so "people predict the outcome that appears most representative of the evidence," and those predictions are "insensitive to the reliability of the evidence or to the prior probability of the outcome." The phrase is the title of Maya Bar-Hillel's 1980 paper "The base-rate fallacy in probability judgments." The literature often hyphenates it as "base-rate fallacy"; the two spellings name the same error.

Why the base rate matters so much. Any judgment of "how likely is this?" combines two pieces of information: how common the outcome is to begin with, and how strongly the specific evidence points to it. When the outcome is rare, even fairly strong evidence often leaves the probability low, because most of the cases that produce the evidence are not the rare outcome. The fallacy is treating the evidence as if it answered the question by itself.

Representativeness is the usual route in. Kahneman and Tversky's 1973 paper found that people predict "the outcome that appears most representative of the evidence" and "erroneously predict rare events and extreme values if these happen to be representative." A company with a charismatic founder and a fast-growing market looks like the companies that became famous, so it feels likely to succeed, even though the base rate of success among young companies is the more informative starting point. The error is not using description at all; it is letting the description replace the frequency.

What it is not. It is distinct from availability bias, which misjudges frequency itself because examples that come to mind easily seem common; the base rate fallacy can occur even when the true frequency is known and stated. It differs from the gambler's fallacy, which is an error about sequences of independent events, and from survivorship bias, which comes from looking only at a filtered sample of winners. Survivorship bias can make the base rate look better than it is, and the base rate fallacy then ignores even that distorted figure.

In money decisions. Card-fraud warnings and other screening tools produce signals about rare events, so a large share of their alerts can be false alarms without the tool being badly built. In investing, the base rate is the question of how often strategies, managers or companies with this profile have done well in the past, and a strong recent record or a compelling story is the case-specific evidence that tempts people to skip it. Asking "out of a hundred like this, how many?" before asking "how good does this one look?" is the practical habit that counters it.

The fallacy is easiest to see with a rare event and an imperfect signal, and the cure is to convert percentages into counts of a large group.

Take an example with invented figures for a card issuer's fraud screen. Suppose 1 in 1,000 transactions is fraudulent, the screen flags 99 percent of fraudulent transactions, and it also flags 2 percent of legitimate ones. Out of 100,000 transactions, 100 are fraudulent and the screen flags 99 of them. The other 99,900 are legitimate, and 2 percent of those, 1,998, are flagged too. That is 99 + 1,998 = 2,097 flagged transactions, of which only 99 are fraud: 99 ÷ 2,097 is about 4.7 percent. A screen that catches 99 percent of fraud still produces flags that are wrong about 95 percent of the time, because fraud is rare. A cardholder who reasons "the screen is 99 percent accurate, so this is almost certainly fraud" has committed the base rate fallacy. The same arithmetic applies to any rare outcome, whether a bankruptcy warning sign or a stock screen that flags the next big winner.

Pros

• Knowing the fallacy gives a concrete question to ask of any striking piece of evidence: how common is the outcome to begin with?
• Turning percentages into counts of a large group makes the correct answer easy to compute without formal statistics.
• It explains why alerts and screens can be useful even when most of their flags are false alarms.

Cons

• Good base rates are often hard to find, and a poorly chosen reference group can mislead as badly as ignoring the base rate.
• Over-correcting is possible: strong, reliable case evidence should move the estimate, and treating every case as average throws that away.
• It is hard to notice in the moment, because the case details feel more relevant than a statistic about other people.