HPC › Decisions › Glossary
Bayes' Theorem is a mathematical rule for revising the probability of a hypothesis in light of new evidence. It states that the posterior probability equals the prior probability multiplied by the likelihood of the evidence, then normalised. Published posthumously in 1763, it is the formal foundation of rational belief updating.
The underlying mathematics serves as the foundation for Bayesian inference, a broader family of statistical methods that treat all uncertain quantities as probability distributions.
How it works
The theorem is expressed as P(H|E) = P(E|H) × P(H) / P(E). 1 P(H|E) is the posterior probability of hypothesis H given evidence E; P(H) is the prior probability encoding what is known before evidence arrives; P(E|H) is the likelihood of observing E if H is true; and P(E) is the total probability of the evidence, the normalising constant that scales the result. The prior and the likelihood ratio are multiplied together; the posterior is the revised belief that emerges.
Despite the theorem's mathematical clarity, human reasoners systematically neglect base rates. When evaluating a positive diagnostic test, most people anchor on the test's stated accuracy and disregard the background prevalence of the condition, producing posterior estimates that overstate the true probability by a factor of ten or more. 2 This is base-rate neglect, a predictable error driven by the representativeness heuristic: the mind judges how closely evidence resembles a prototype rather than computing how it should shift a prior.
One reliable correction is to reframe problems using natural frequencies rather than conditional probabilities. Presenting a medical scenario as '10 in 1,000 patients have the condition; of those 10, nine will test positive; of the 990 without the condition, 49 will also test positive' makes the Bayesian structure transparent without formal calculation. In controlled experiments, approximately 46% of participants reached the correct answer using this format, compared with roughly 16% using standard probability statements. 3 The representation format, not formal training, is the most actionable lever.
Example
A screening programme tests for a condition affecting 1 in 1,000 people. The test identifies 90% of cases correctly and returns a false positive for 5% of healthy individuals. Intuition after a positive result suggests a probability close to 90%. Applying the theorem yields approximately 1.8%: because the condition is rare, most positive tests come from the 5% false-positive rate applied to the large healthy majority.
The prior probability of the condition, not the test's accuracy alone, determines what a positive result actually means.
Why it matters
Base-rate neglect produces predictable, systematic errors in high-stakes domains. Physicians in classic studies overestimated the posterior probability of cancer following a positive screening test by roughly tenfold, a distortion that translates directly into unnecessary interventions, patient anxiety, and misallocation of diagnostic resources. 2 In legal contexts, jurors who encounter compelling evidence routinely underweight the prior improbability of guilt, inflating conviction confidence beyond what the evidence warrants.
For decisions that depend on probabilistic reasoning, the lesson is structural. Reframing problems in natural frequency terms consistently improves accuracy without formal statistical training, because the representation makes prior probabilities explicit rather than hidden. 3 Treating current beliefs as provisional estimates, revisable in proportion to each piece of incoming evidence, is the practical discipline Bayes' theorem prescribes. Whether the brain computes the rule explicitly or approximates it through other mechanisms remains an open question; the normative standard the theorem provides is not. 4
What is Bayes' theorem in simple terms?+
Bayes' theorem is a rule for updating a probability estimate when new information arrives. You begin with a prior probability based on background rates, observe new evidence, then calculate how much more likely that evidence would be if the hypothesis were true. The result is a revised, posterior probability.
Why do people struggle with Bayesian reasoning?+
The main obstacle is base-rate neglect. Human reasoners anchor on vivid, immediate evidence and systematically underweight background frequency information. When a positive medical test is salient and concrete, the low prevalence of the underlying condition is psychologically overridden, producing posterior estimates far higher than the mathematics supports.
How do natural frequencies make probability problems easier?+
Natural frequencies restate conditional probabilities as counts within a defined population, making the Bayesian structure visible. Saying '8 out of 1,000 people with the condition test positive' is mathematically equivalent to the percentage form but reduces the cognitive work required to identify the numerator and denominator of the posterior calculation.
Is the brain actually Bayesian?+
Behaviour in many perceptual and decision tasks approximates Bayesian predictions, but whether the brain explicitly computes Bayes' rule at an algorithmic level is actively debated. Recent frameworks distinguish behaviourally Bayesian outputs from explicit Bayesian computation, suggesting the correspondence may reflect other mechanisms rather than direct implementation of the theorem.
