Bayesian inference updates belief as each new clue arrives
Bayesian inference uses Bayes' theorem to turn a prior guess about a hypothesis into a posterior once evidence lands. Science, medicine, law and sports all lean on this loop: multiply prior belief by how well the data fit, then renormalize.
Statistical inference in the Bayesian style computes the probability of a hypothesis after seeing data, then revises again as more observations arrive. A prior distribution encodes belief before the new evidence; a likelihood from a statistical model measures how compatible that evidence is with each hypothesis; Bayes' theorem yields the posterior. The pattern matters especially when data arrive in sequence. Philosophically it ties to subjective or "Bayesian" probability in decision theory, and it appears across engineering, psychology, law and clinical work.
In symbols, P(H|E) = P(E|H)P(H)/P(E). H is the hypothesis under test among competitors; P(H) is the prior; E is evidence unused when the prior was set; P(H|E) is what we want—the posterior. P(E|H) is the likelihood: as a function of H with E fixed it scores compatibility. P(E), the marginal likelihood or model evidence, does not depend on which hypothesis is labelled and therefore cancels when comparing relative posteriors—so long as it is not zero. Intuitively the posterior rises with both the hypothesis's prior plausibility and its fresh likelihood.
When H and not-H form an exclusive pair, the denominator expands into two weighted likelihood terms, and an algebraically rearranged form highlights a single composite factor. If that factor is near 1, the posterior sits near even odds; if it shrinks toward 0, the hypothesis looks nearly certain given the evidence; if it balloons, the hypothesis looks improbable. An unlikely prior already pushes that factor large. The multiplication rule P(E ∩ H) = P(E|H)P(H) = P(H|E)P(E) is a compact memory aid.
Bayesian updating is convenient yet not the only coherent learning rule. Ian Hacking noted that classic Dutch-book arguments pin probability axioms without forcing the dynamic Bayesian step. Richard C. Jeffrey's probability kinematics updates when the evidence itself is uncertain. Still, when exclusive, exhaustive models share a prior that sums to one, observing i.i.d. events lets the whole belief distribution shift together under Bayes' rule.
Source: Bayesian inference
