Base Rate Fallacy

The systematic human tendency to underweight or ignore the prior probability of a hypothesis when vivid, specific evidence is available, so a posterior estimate collapses toward the likelihood instead of tracking Bayes' rule.

Version

v4 · 2026-09-28 · History · 2 corrections

Domain-specific #

83

Domain group

Social Sciences

Origin domain

Psychology & Behavioral Sciences

Subdomain

Judgment and Decision Making → Psychology & Behavioral Sciences

Core Idea

The base rate fallacy is the systematic tendency to underweight or ignore prior probability when vivid, specific evidence is available, producing posteriors that track the evidence far more than Bayes' rule warrants. Canonically, a 99%-accurate positive test for a 1-in-1000 disease yields under 2% true-positive probability, yet people report near 95%. The proximate mechanism is representativeness — judging probability by how well evidence resembles the hypothesis — which does not encode the prior, so the base rate drops out.

How would you explain it like I'm…

Forgetting How Rare It Is

Imagine a sickness that almost nobody has. A test says you have it, but the test sometimes says 'yes' by mistake to healthy people. Because there are so many healthy people, most of the 'yes' answers are actually mistakes. The base rate fallacy is when people forget how rare the sickness is and trust the test's 'yes' much too much.

Ignoring How Common Things Are

The base rate fallacy is a common thinking mistake: people focus on a specific, exciting clue and forget how common or rare something is to begin with. For example, say a disease affects 1 in 1,000 people, and a test almost always catches it but also says 'positive' for 5 out of every 100 healthy people. If you test positive, you'd guess you're 95% likely to have it, but the real chance is under 2%, because the healthy people are so many that their mistaken positives far outnumber the real ones. Even many doctors get this wrong. It helps a lot to think in counts: out of 1,000 people, 1 is sick and tests positive, but about 50 healthy people also test positive.

Neglect of Prior Probability

The base rate fallacy is the systematic tendency to underweight or ignore prior probability (the base rate) when a vivid, specific piece of evidence is available, so that probability judgments follow the evidence far more than Bayes' rule allows. Classic example: a disease affects 1 in 1,000 people; a test detects it 99% of the time but has a 5% false-positive rate. After a positive result, the true chance of disease is below 2%, yet most people, including trained clinicians, guess about 95%. Kahneman and Tversky explained this through representativeness: people judge probability by how well evidence fits a hypothesis, and fit says nothing about how common the hypothesis is. Vividness makes it worse, since a positive test is concrete while a population statistic is abstract. Gigerenzer showed that presenting the same information as natural frequencies ('1 in 1,000 has it and tests positive; of the 999 without it, about 50 also test positive') greatly reduces the error, because it makes the population structure visible.

 

The base rate fallacy is the systematic tendency to underweight or ignore prior probability when vivid, case-specific evidence is available, so that posterior estimates track the evidence far more than Bayes' rule warrants. In the canonical case, a condition with prevalence 1 in 1,000, a test with 99% sensitivity and a 5% false-positive rate yields a true-positive probability below 2% after a positive result, yet most people, trained clinicians included, report intuitions near 95%; the specific evidence crowds out the prior almost entirely. The proximate mechanism, following Kahneman and Tversky, is representativeness: the probability of a hypothesis is judged by how well the evidence resembles the hypothesis category, and because similarity does not encode prior probability, the base rate drops out. Differential vividness compounds this, since a positive result or detailed case description is concrete and imageable while a prevalence statistic is abstract. Gigerenzer showed that recasting the same information as natural frequencies (1 of 1,000 has the disease and tests positive; about 50 of the 999 without it also test positive) dramatically reduces the fallacy, because the frequency format makes the prior population structure explicit in a way percentages do not.

Scope of Application

Lives in one domain — human judgment and decision-making — the contexts below being application settings of a single substrate, a reasoner forming a posterior.

• Medical screening — the mammography/HIV-test problem: a positive rare-condition test over-read by patients and clinicians.
• Criminal forensics — DNA random-match probabilities presented without population context.
• Hiring and admissions — a vivid interview signal weighted without candidate base rates.
• Security and counter-terror screening — rare-event base rates swamped by false positives.
• Intelligence analysis — vivid intelligence overriding cool priors.

Clarity

Naming the fallacy separates two quantities intuition collapses: the likelihood (how characteristic the evidence is if the hypothesis holds) and the posterior (how probable the hypothesis is given evidence and prior). The felt answer to "the test is positive, how worried should I be?" reports the likelihood while seeming to report the posterior. The label converts a fuzzy worry into a prior-likelihood-posterior triple and points at a representational repair.

Manages Complexity

Over-confident inferences recur across screening, forensics, hiring, and intelligence, each seeming to need its own account. The fallacy compresses them to one diagnosis: vivid evidence crowds out the prior because representativeness substitutes resemblance for probability. An analyst tracks the same three quantities everywhere, predicts the danger zone in advance, and applies one uniform fix — recast the figures as natural frequencies that expose the prior population structure.

Abstract Reasoning

The fallacy supports a diagnostic move (from an over-confident estimate back to the dropped prior, with the self-probe "would my estimate change if I wrote the prior down first?"), an interventionist move (recast as natural frequencies and predict the estimate drops sharply, while exhortation will not), boundary-drawing (the danger zone is a rare base rate meeting a vivid signal; distinct from sibling representativeness errors), and a predictive move (anticipate who over-states, and by how much).

Knowledge Transfer

Within judgment and decision-making the diagnosis transfers as mechanism, intact, across screening, forensics, hiring, and intelligence — one substrate, a human reasoner, with the setting swapped, since the same representativeness machinery is at work. Beyond the human reasoner the fallacy does not travel: the mathematics it violates is substrate-general and recurs everywhere, but there it is the rule correctly applied, not a fallacy. What travels is the parent — Bayesian Updating — and any use of the fallacy outside a reasoning agent is borrowing the name. Carry Bayes, not the fallacy. dag_edges:

Relationships to Other Abstractions

Current abstraction Base Rate Fallacy Domain-specific

Parents (3) — outgoing typed relations

• Base Rate Fallacy is a kind of Bias Prime

  Base-rate fallacy is bias specialized to a posterior estimate that is systematically displaced toward vivid likelihood evidence because its prior is underweighted.

• Base Rate Fallacy includes Representativeness Heuristic Domain-specific

  Base-rate fallacy contains representativeness because resemblance or vivid likelihood evidence supplies the proxy estimate that displaces the prior.

• Base Rate Fallacy presupposes Bayesian Updating Prime

  The base-rate fallacy presupposes Bayesian updating because it is defined by a posterior that fails to integrate the prior with the likelihood.

Hierarchy paths (7) — routes to 6 parentless roots

• Base Rate Fallacy → is a kind of Bias

Show alternative paths (6)

Neighborhood in Abstraction Space

Base Rate Fallacy sits in a sparse region of the domain-specific corpus (65^th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Unclustered & Miscellaneous (2769 abstractions)

Nearest neighbors

• Conservatism Bias — 0.89
• Conjunction Fallacy — 0.87
• Cherry Picking — 0.85
• Ad Nauseam (Argument from Repetition) — 0.83
• Hasty generalization — 0.82

Computed from structural-signature embeddings · 2026-10-25