Bayesian Disease Test Truth

This question presents a medical testing scenario where a rare disease and an imperfect diagnostic test interact. The candidate is asked to determine how likely it is that someone actually has the disease after receiving a positive result, given information about how common the disease is and how the test behaves on sick and healthy individuals. This kind of problem mirrors real-world issues in clinical decision-making, risk communication, and algorithmic classification, and often appears in introductory quant interviews and data science screening rounds to test comfort with uncertainty and evidence.

The solution leans on conditional probability and Bayesian inference: translating a verbal description into prior, likelihood, and posterior terms, then correctly normalizing over all ways a positive result can occur. Strong answers typically reframe the situation in terms of proportions in a large hypothetical population, compute the number of true and false positives, and then compare them. Interviewers watch for awareness of base-rate effects, the ability to avoid intuitive but incorrect "test accuracy equals disease probability" reasoning, and clear explanation of why rare-event settings can make even good tests produce many false alarms.

What it tests

Bayesian inference governs problems where we update our beliefs about an event (such as having a disease) after observing new evidence (like a test result). The core structure is that the probability of the event given the evidence depends not just on the accuracy of the evidence (test sensitivity and specificity), but crucially on the base rate (prevalence) of the event in the population. This is because even a highly accurate test can yield mostly false positives if the event is rare: the large pool of unaffected individuals can generate more false positives than there are true positives. The denominator in Bayes' formula aggregates all ways the evidence could arise, weighting each by how common its source is. This pattern holds because probability is fundamentally about proportions: we must compare the number of true positives to all positives, not just look at the test's properties in isolation.