Glossary
Each term links back to the module that introduces it — the definition is a reminder, the module is the explanation. If a term here was never defined before a module used it, that’s a defect in the course, not in you.
- Base rate
The background frequency of a thing in its reference class. Ignoring it — base-rate neglect — is the classic Bayesian blunder.
introduced in Module III · Priors & base rates → - Beta distribution
The workhorse curve for belief about a proportion. Start flat; add heads to one shape parameter and tails to the other, and it sharpens on its own.
introduced in Module V · Updating in sequence → - Calibration
The match between your stated confidences and your hit rate: of all the things you call “80 % likely,” about 80 % should turn out true.
introduced in Module I · Beliefs as bets → - Credible interval
A range holding a stated share of your belief — “90 % credible interval 0.56–0.77” means you give 9 : 1 odds the true value is inside.
introduced in Module V · Updating in sequence → - Degree of belief
A probability read as personal confidence that something is true, rather than as a long-run frequency. The number you would use to price a bet.
introduced in Module I · Beliefs as bets → - False positive
An alarm without a fire: the test says yes, reality says no. When a condition is rare, false positives outnumber true ones.
introduced in Module II · Bayes by counting → - Likelihood
How probable the observed evidence is under a hypothesis: P(E | H). Not the same as P(H | E) — confusing the two is the transposition error.
introduced in Module IV · Weighing evidence → - Likelihood ratio
The strength of a clue: how much more probable the evidence is if the hypothesis is true than if it is false. Multiply your prior odds by it.
introduced in Module IV · Weighing evidence → - Naive Bayes
A classifier that multiplies one likelihood ratio per feature as if the features were independent. Wrong in theory, indecently effective in practice.
introduced in Module VI · Bayes in the wild → - Natural frequencies
Probabilities restated as counts of people — “9 of 98 positives” instead of “9.2 %.” The format in which humans get Bayes right.
introduced in Module II · Bayes by counting → - Odds
A probability expressed as a ratio of for to against: 75 % is 3 : 1. Odds are the native dialect of Bayesian updating — multiplying them is easy.
introduced in Module I · Beliefs as bets → - Odds-form Bayes
Posterior odds = prior odds × likelihood ratio. The whole theorem in one multiplication.
introduced in Module IV · Weighing evidence → - Posterior
What you believe after evidence has done its work: the updated probability, prior × evidence, renormalized.
introduced in Module II · Bayes by counting → - Prevalence
How common a condition is in the population before any testing — the base rate, worn by a disease.
introduced in Module II · Bayes by counting → - Prior
What you believed before the new evidence arrived — the base rate, or your last posterior wearing new clothes.
introduced in Module II · Bayes by counting → - Sensitivity
The share of true cases a test catches. A 90 % sensitive test misses 1 sick person in 10.
introduced in Module II · Bayes by counting → - Sequential updating
Feeding evidence in one piece at a time, each posterior becoming the next prior. Order does not matter; the product is the same.
introduced in Module V · Updating in sequence → - Specificity
The share of non-cases a test correctly clears. A 91 % specific test falsely flags 9 healthy people in 100.
introduced in Module II · Bayes by counting →