LOGIC & PUZZLES / EVIDENCE
Base rates:
“Out of how many?”
A statistic about accuracy is incomplete without knowing how common the condition or event was before the test.
THE MAIN IDEA
Start with the group being counted.
A base rate is the starting frequency of an event or condition in a particular population. When interpreting a screening test, for instance, knowing only that it is “ninety-nine percent accurate” leaves several questions open. Accurate might refer to sensitivity, specificity, overall agreement, or another measure. These have different definitions, and none alone tells you the chance that a person with a positive result actually has a condition. The underlying prevalence matters because a rare condition can produce many more false positives than true positives even when a test performs well. Imagine, purely as arithmetic, a population of 10,000 people where a condition affects 1 percent. Suppose an illustrative test detects 99 percent of affected people and falsely flags 1 percent of unaffected people. That gives about 99 true positives and about 99 false positives, so only about half of positive results would be true in this idealized example. The arithmetic is not a description of any real medical test, and real performance can differ by population, use, and setting. Representing the case as natural frequencies—how many people in each group—often makes the relationship easier to understand than multiplying abstract percentages. Draw four boxes: condition and no condition, each split into positive and negative result. Then write the total people in each group and apply the stated rates. That makes missing data visible and helps spot when a word like “accurate” is being used without a definition. Base-rate neglect is not just a medical-testing mistake. It also appears when judging a rare event from a vivid anecdote, evaluating an email warning, or estimating whether a surprising pattern occurred by chance. Context does not override direct evidence; it tells you what starting point to combine with that evidence. Real health decisions require the test's actual published performance, the correct reference population, repeat-test policy, and guidance from qualified healthcare professionals. Do not use a classroom example to interpret a personal result. A careful question—“out of how many people like the group being discussed?”—can uncover whether a statistic is decision-relevant at all, and can help you ask for the missing denominator before reaching a confident conclusion.
TRY THIS
Convert a hypothetical percentage into people.
Choose round numbers for a made-up population and clearly label every assumed percentage. Count affected and unaffected groups first, then positive and negative outcomes inside each group. Calculate the share among positive results; never apply this demonstration to a personal health test.