<研究論文(原著論文)>グッドによるベイズ主義的な確証度

Abstract

Bayesian measure c(H, E | K) of confirmation is the degree to which known evidence E supports a given hypothesis H relative to background knowledge K. This measure should be positive when E confirms H, negative when E disconfirms H and zero when E is irrelevant to H. There are many Bayesian measures of confirmation which satisfy this condition. log Pr(E | H & K)/Pr(E | ~ H & K), which was proposed by Good, is one of them. In the present paper, it is shown that only Good's measure is valid in the case where the measure c(H, E1 & E2 | K) of confirmation is the sum of c(H, E1 | K) and c(H, E2 | K) when, in Reichenbach's terminology, H screens off E1 from E2.