In short
The base-rate fallacy is the tendency to judge how likely something is from the specific evidence in front of you, such as a witness's account or a test result, while ignoring how common that thing is in the first place. It shows up in reading medical tests, weighing eyewitness identifications, screening job applicants, interpreting security alerts and any situation where a fairly reliable signal is pointed at a rare event.
How the base-rate fallacy works
The fallacy is a failure of weighting: the evidence about the particular case is allowed to stand in for the whole answer, and the information about how common the case is to begin with drops out. In the cab problem the witness is right 80 percent of the time, so 80 percent feels like the answer, but Blue cabs are rare. Picture 100 cab accidents. About 85 involve a Green cab and 15 a Blue one. The witness calls 80 percent of the 15 Blue cabs Blue, which is 12, and miscalls 20 percent of the 85 Green cabs Blue, which is 17. Of the 29 times the witness says Blue, only 12 are right. That is 41 percent.
Kahneman and Tversky traced the error to the representativeness heuristic: people judge probability by how well the evidence resembles the outcome. Bar-Hillel's own account in 1980 was about relevance. Information about one cab feels more relevant than a city-wide percentage, so it dominates; she showed the dominance can be weakened without touching the mathematics. Her intercom version reverses the colors and asks about a Green cab: 85 percent of the cabs are Blue, and an intercom, heard through the window, is fitted in 80 percent of Green cabs and 20 percent of Blue ones. The arithmetic mirrors the cab problem, so the answer is again 41 percent, and the median answer moved to 48. The clue had the same diagnostic value in both versions; what changed was how specific to the case it felt.
Presentation matters as much as the problem. Gigerenzer and Hoffrage showed in 1995 that stating the same information as natural frequencies, such as 10 out of every 1,000 women, rather than as percentages, raised correct Bayesian answers several-fold, because frequencies let people count the way the paragraph above counts. Koehler argued in 1996 that the literature had oversold the fallacy: people do use base rates when the base rate is causally connected to the outcome, when they learn it from experience, and when the task is framed in frequencies. Base rates are often underweighted rather than ignored, and how much depends on how the problem is put.
The original study
Daniel Kahneman and Amos Tversky introduced the cab problem in a 1972 research bulletin, and the fuller treatment appeared in their chapter "Evidential impact of base rates" in the 1982 collection Judgment under Uncertainty. In their version, 85 percent of the cabs are Green, 15 percent Blue, the witness identifies the cab as Blue and is correct 80 percent of the time; the Bayesian answer is 41 percent and the median and modal response was 80 percent. Bar-Hillel’s 1980 printing swaps the two colors, so the cab the witness names is Green there; the proportions, the witness’s 80 percent accuracy and the Bayesian 41 percent are the same either way, and this page keeps the colors the 1982 chapter uses.
Their 1973 paper "On the psychology of prediction" is the companion study. Participants read personality sketches drawn, they were told, from a group of 70 engineers and 30 lawyers, or 30 engineers and 70 lawyers, and estimated the probability that each person was an engineer; the estimates barely moved between the two base rates. Given no description at all, people used the base rate correctly; given a description that said nothing diagnostic, they answered about 50 percent in both conditions, as if a useless description had canceled a useful base rate.
Maya Bar-Hillel's 1980 paper in Acta Psychologica, "The base-rate fallacy in probability judgments," gave the effect its name and its most careful mapping. She posed about 45 problems to roughly 1,500 people, most of them applicants sitting university entrance examinations in Jerusalem. Of 52 people who answered the cab problem, 36 percent gave exactly 80 percent, 80 percent was also the median, and only about 10 percent came close to 41. Across variations in wording and order the modal answer tracked the witness's reliability, given by as many as 70 percent of participants.
Read the diagram and evidence as text
A schematic mixed fleet sits behind a prominent witness report. The report feeds an 80% judgment diamond, separated from the 41% Bayesian-answer dot on a common 0–100% scale. Cab multiplicity is illustrative and does not encode the base-rate proportions.
The cab problem: what reached the answer
Bars are shares of the 52 participants
The witness's reliability: Right 80% of the time: the evidence about this cab
The base rate: Blue cabs are rare in the city
The common answer: 36% answered exactly 80%; 80% was also the median
The Bayesian answer: 41%: only about 10% came close
The witness's reliability → the signal becomes the answer → The common answer (error transition)
The witness's reliability → weighed against the base rate → The Bayesian answer
| Condition | Value and unit | Evidence quote |
|---|---|---|
| Participants who answered the cab problem | 52 participants | the distribution of estimates that 52 Ss gave to this problem |
| Witness's reliability in the cab problem, and the median answer | 80 percent | Eighty percent was also the median estimate. [Display conversion: 80 is the numeric transcription of Eighty.] |
| Participants answering exactly 80 percent | 36 percent | Thirty-six percent [Display conversion: 36 is the numeric transcription of Thirty-six.] |
| Bayesian answer to the cab problem | 41 percent | 41% |
| Participants coming close to the Bayesian answer, approximately | 10 percent | 10% |
The two medical problems in the demo are the classic clinical demonstrations. Casscells, Schoenberger and Graboys asked 60 students, house officers and physicians at Harvard Medical School in 1978 about a test for a disease with a prevalence of 1 in 1,000 and a 5 percent false-positive rate: 27 answered 95 percent and 11 gave the answer of about 2 percent. Eddy reported in 1982 that about 95 of 100 physicians put the chance of cancer after a positive mammogram between 70 and 80 percent, against 7.7 percent by Bayes' rule.
Everyday examples
When a reliable signal is pointed at something rare, most alarms are false. The rows below are illustrations, not findings from the studies cited on this page.
| Situation | What the bias does | What helps |
|---|---|---|
| A positive result on a screening test for a rare condition | The test's accuracy is read as the chance of having the condition, though most positives in a rare condition are false | Ask how many of 1,000 people like you have the condition, and how many of the rest would also test positive |
| A confident eyewitness picks a suspect from a lineup | The witness's confidence stands in for the probability; the number of plausible suspects is never counted | Treat confidence as one signal and ask what else points to the same person |
| A security system flags an employee's login as suspicious | An alert that is 99 percent accurate on millions of ordinary logins produces mostly false alarms, each treated as a breach | Compute the expected number of false alarms per day before setting the response policy |
| A hiring manager reads a glowing reference | The enthusiasm is taken at face value without asking how often references are glowing for average candidates | Ask the referee to rank the candidate against others they have recommended |
| A news story about a crime committed by a member of a small group | The vivid case drives the estimate of how common the behavior is | Find the rate per population before judging the group |
How to counter it
The counters are arithmetic habits, and the research says they work better than instruction in probability theory.
- Convert to natural frequencies. Take 1,000 people, work out how many have the condition, then how many in each group would produce the signal, and compare the two counts. Gigerenzer and Hoffrage found that this format alone raised correct answers substantially.
- Write the base rate down before you look at the evidence. Once the vivid detail is in view it is hard to give the prior its weight.
- Separate two questions that feel like one: how often the signal appears when the thing is true, and how often the thing is true when the signal appears. The witness's 80 percent answers only the first.
- Count the false alarms. For any alert, test or screen, multiply the number of people without the condition by the false-positive rate and compare it with the number of true positives.
- Notice when the evidence feels specific to the case. Bar-Hillel's work suggests that specificity is what makes a clue dominate, so a clue that feels tailored to this person deserves extra scrutiny, not less.
None of this makes the witness or the test worthless. A positive mammogram in Eddy's problem raises the probability of cancer from 1 percent to about 8 percent. The error is not in trusting the evidence; it is in forgetting what it is being added to.
What it is not
The base-rate fallacy is one of a family of errors that Kahneman and Tversky traced to judging by resemblance, and its neighbors break different rules.
- It is not the conjunction fallacy, in which a specific scenario is judged more probable than the general case that contains it. The conjunction fallacy ignores the relationship between a set and its subset; the base-rate fallacy ignores how large the set is to begin with.
- It is not the anecdotal fallacy, which prefers one concrete story to a statistical summary already in hand. That one reasons from a single case; the base-rate fallacy reasons about a single case while neglecting the population it came from.
- It is not the availability heuristic, which fills the gap when no frequency has been given. The base-rate fallacy is the failure to use a frequency you have been given.
Tests and tools that use this
Related biases
Base-Rate Fallacy FAQ
What is the base-rate fallacy?
It is the tendency to estimate how likely something is from the specific evidence about a case, such as a test result or a witness, while underweighting how common that thing is in the population. In the cab problem a witness who is right 80 percent of the time identifies a Blue cab in a city where only 15 percent of cabs are Blue; most people answer 80 percent, and Bayes' rule gives 41 percent.
Why is the answer to the taxi problem 41 percent?
Imagine 100 cab accidents. About 85 involve Green cabs and 15 involve Blue ones. The witness correctly calls 12 of the 15 Blue cabs Blue, and wrongly calls 17 of the 85 Green cabs Blue. So the witness says Blue 29 times, and is right in 12 of them. Twelve out of 29 is 41 percent. The witness's reliability is real; it is just being applied to a population in which Blue cabs are rare.
Is the base-rate fallacy real or overstated?
Both claims have support. The fallacy is easy to produce with problems like the cab problem, and clinicians in the Casscells and Eddy studies showed it with medical tests. Koehler argued in 1996 that people often do use base rates, especially when the base rate is causally linked to the outcome, learned from experience, or presented as frequencies. The current reading is that base rates are typically underweighted rather than ignored, and how much depends on presentation.
How did the demo on this page measure the fallacy?
You moved three sliders, one for the cab problem and two for medical screening problems, and the page compared each answer with the Bayesian result: 41, about 2 and about 8 percent. An answer above the Bayesian figure counts as overweighting the evidence, an answer at or below it does not. The outcome is based on the majority of the three. Three problems can show the pattern but cannot measure how you reason in general. Only the outcome label leaves your browser.
How does the base-rate fallacy affect medical tests?
A test with a low false-positive rate still produces mostly false positives when the condition is rare, because there are so many more healthy people to be wrong about. In Eddy's 1982 example, a positive mammogram with a 1 percent base rate, 79 percent sensitivity and a 9.6 percent false-positive rate implies a 7.7 percent probability of cancer, while most physicians he asked said 70 to 80 percent. This is why screening results are usually followed by a confirmatory test.
How can I avoid the base-rate fallacy?
Work in natural frequencies. Start with 1,000 people, count how many have the condition, then count how many in each group would produce the evidence, and compare the two counts rather than the percentages. Write down the base rate before looking at the specific evidence, and keep separate the question of how often the evidence appears when the thing is true from how often the thing is true when the evidence appears.
References
- Bar-Hillel, M. (1980). The base-rate fallacy in probability judgments. Acta Psychologica, 44(3), 211-233. doi.org
- Tversky, A., & Kahneman, D. (1982). Evidential impact of base rates. In D. Kahneman, P. Slovic, & A. Tversky (Eds.), Judgment under uncertainty: Heuristics and biases (pp. 153-160). Cambridge University Press. doi.org
- Kahneman, D., & Tversky, A. (1973). On the psychology of prediction. Psychological Review, 80(4), 237-251. doi.org
- Casscells, W., Schoenberger, A., & Graboys, T. B. (1978). Interpretation by physicians of clinical laboratory results. New England Journal of Medicine, 299(18), 999-1001. doi.org
- Eddy, D. M. (1982). Probabilistic reasoning in clinical medicine: Problems and opportunities. In D. Kahneman, P. Slovic, & A. Tversky (Eds.), Judgment under uncertainty: Heuristics and biases (pp. 249-267). Cambridge University Press. doi.org
- Gigerenzer, G., & Hoffrage, U. (1995). How to improve Bayesian reasoning without instruction: Frequency formats. Psychological Review, 102(4), 684-704. doi.org
- Koehler, J. J. (1996). The base rate fallacy reconsidered: Descriptive, normative, and methodological challenges. Behavioral and Brain Sciences, 19(1), 1-17. doi.org
Cite this source
Psychology.com. (2026, October 23). Base-Rate Fallacy. Psychology.com. https://psychology.com/biases/base-rate-fallacy