Probability & Statistics
Bayesian Reasoning
Master the framework that doctors, intelligence analysts, and data scientists use to update beliefs rationally when new evidence arrives. These exercises build your intuition for Bayes' theorem through scenarios involving criminal investigations, medical diagnosis, A/B testing, and geopolitical analysis. You will learn when evidence should dramatically change your mind versus when it should barely shift your confidence.
Context
Why this exercise
Bayesian reasoning is the discipline of updating beliefs in proportion to evidence. Where naive intuition treats new evidence as a verdict ('the test was positive, so the disease is present'), Bayesian reasoning treats new evidence as one input that combines with prior probability to produce a posterior. This framework underlies modern statistical inference, machine learning, intelligence analysis, and medical diagnosis. The exercise drills the specific moves: identifying the prior, the likelihood, and the posterior; using natural-frequency representations to make Bayes' theorem intuitive; and recognizing the classic failures (base-rate neglect, prosecutor's fallacy, confusion of conditional probabilities) that produce confidently wrong answers.
Before you start
Bayes' theorem was published by Reverend Thomas Bayes posthumously in 1763 and developed into a general framework for inference by Pierre-Simon Laplace. The theorem itself is a one-line piece of arithmetic — P(H|E) = P(E|H) × P(H) / P(E) — but its interpretive power is enormous. It tells you exactly how to combine a prior belief (P(H), what you thought before seeing the evidence) with the likelihood of the evidence under the hypothesis (P(E|H)) and the overall probability of the evidence (P(E)) to produce a posterior belief (P(H|E)). When the prior is low and the evidence is moderately reliable, the posterior remains lower than intuition expects — which is why a 90%-accurate test for a 1-in-1000 disease still produces more false positives than true positives, and why the posterior probability of having the disease after a positive test is far below 90%.
Gerd Gigerenzer's research at the Max Planck Institute showed that the difficulty most people have with Bayesian reasoning is not mathematical but representational. The same problem expressed in probability format ('the test has a 90% sensitivity and a 5% false positive rate; the disease has a 1% prevalence; what is the probability that a positive-testing person has the disease?') is solved correctly by about 15% of physicians, while the same problem expressed in natural frequencies ('out of 1,000 people, 10 have the disease and 9 of them test positive; of the 990 without the disease, 50 test positive; if you test positive, what is the probability you have the disease?') is solved correctly by about 65% of the same physicians. The procedural fix for any Bayesian problem is to translate it into natural frequencies before reasoning about it.
The prosecutor's fallacy is the most consequential failure of Bayesian reasoning in real-world settings. It is the confusion of P(evidence | innocent) — usually a small number, like the random-match probability of a DNA profile — with P(innocent | evidence) — which depends on the prior probability that the defendant committed the crime and can be much larger. The 1989 Sally Clark case in England, in which a mother was convicted of murdering two children based on a misapplied 1-in-73-million statistical claim, is the textbook example. As you work the scenarios, practice writing out the prior, likelihood, and posterior explicitly, translating probability claims into natural frequencies when possible, and noticing when an argument is implicitly equating P(A|B) with P(B|A). For background on the underlying inference framework, see Types of Reasoning.