Bayes' rule: updating a probability after evidence
Condition in the direction the question asks
Bayes' rule reverses a conditional probability. If the question is about the chance of a condition after an observation, write P(condition | observation), not the test sensitivity P(observation | condition). The update is P(A | B) = P(B | A)P(A) / P(B).
The denominator counts every route to the observation
The observation can arise among people with the condition and among people without it. Therefore P(B) is the total probability of the observation: P(B | A)P(A) + P(B | not A)P(not A) when the two groups partition the population.
A frequency table is often the safest calculation: begin with a convenient hypothetical population, split it by the prior probability, then apply the appropriate conditional rate in each group.
Base rates are information, not a distraction
A sensitive test can still have a modest positive predictive value when the condition is uncommon, because even a small false-positive rate is applied to many more unaffected people.
Conversely, a negative result is interpreted through both the false-negative rate and the prior probability. The posterior is an update within a specified model; it does not establish whether the sensitivity, specificity, or prior applies to a new population.
Suppose P(A) = 0.30, P(B | A) = 0.60, and P(B | not A) = 0.10. What is P(A | B)?
After observing B, a student has correctly calculated P(A | B) = 0.40. A second observation C is conditionally independent of B conditional on whether A occurs: P(C | A, B) = P(C | A) and P(C | not A, B) = P(C | not A). Suppose P(C | A) = 0.75 and P(C | not A) = 0.25. Which expression equals P(A | B, C)?