Base-rate neglect

About

Base-rate neglect is the failure to consult prior probability when weighing particular evidence. The vivid case stands alone; the population it came from is never called.

Taxonomy - Probabilistic reasoning bias, prior neglect. The error is treating diagnostic evidence as if it could be read without the relevant class frequency.

Informal Example - "The test is positive, so the rare condition is probably present," without considering how rare the condition is.

Minimal Symbolic Form

E supports H ⊢ HighP(H | E)

while ignoring:

P(H)

The evidence is read in isolation; the base rate goes unconsulted.

Probabilistic Analysis

Invalid Update Pattern:

P(H | E) determined by P(E | H) alone

Posterior probability cannot be determined by sensitivity-like evidence alone.

Missing Necessary Condition

Bayesian updating requires the prior:

P(H | E) = P(E | H)P(H) / P(E)

Neglecting P(H) breaks the inference.

Diagnostic Model

Biased Model:

E → H

Corrective Model:

Prior(H) + Likelihood(E|H) + FalsePositive(E|¬H) → Posterior(H|E)

Diagnostic evidence must be interpreted relative to both prevalence and error rates.

Bayesian Formalization

The relevant comparison is:

P(H | E) = P(E | H)P(H) / [P(E | H)P(H) + P(E | ¬H)P(¬H)]

A low base rate can keep the posterior low even when evidence seems strongly diagnostic.

Epistemic Representation

𝔼(E) ∧ Ignore(P(H)) ⇏ □(H)

Evidence for H does not justify belief in H until the prior and alternatives are included.

Countermodels

Rare Hypothesis, Common False Positive

LowP(H) ∧ P(E | ¬H) significant

Most positive cases can still be false positives.

Balanced Update

Prior(H) included → Lower posterior than intuitive estimate

The correction is not anti-evidence; it is complete evidence.

Historical Background

Base-rate neglect has no traditional Latin name. The Neo-Latin neglectus frequentiae prioris is descriptive.

Its deeper history belongs to probability theory and inductive logic: evidence must be assessed against prior frequency and alternative explanations.

Earlier logical treatments of signs and testimony sometimes approach this problem, but the explicit formal account belongs to Bayesian probability rather than ancient fallacy lists.