No label defined (Q4936007)

The aim of this paper is to clarify under which additional assumptions the Cox's famous theorem (1946) holds. Cox considered a function Bel that associates a real number with each pair \((U,V)\) of subsets of a domain \(W\) such that \(U\neq\emptyset\). \(\text{Bel}(U,V)\), also written as \(\text{Bel}(U|V)\), is thought of as the belief, credibility, or likelihood of \(V\), given \(U\). The Cox's theorem claims that if Bel satisfies two reasonable restrictions A1 and A2, then Bel must be isomorphic to a probability measure. The two restrictions are as follows: there exist functions \(S:\mathbb{R}\to \mathbb{R}\) and \(F: \mathbb{R}^2\to \mathbb{R}\) such that (with \(\overline V\) denoting the complement of \(V\)): \[ \text{Bel}(\overline V|U)= S(\text{Bel}(V|U))\quad\text{if }U\neq\emptyset,\text{ for all }U,V\subseteq W,\tag{A1} \] \[ \text{Bel}(V\cap V'|U)= F(\text{Bel}(V'|V\cap U),\text{Bel}(V|U))\quad\text{if }V\cap U\neq\emptyset,\text{ for all }U,V,V'\subseteq W.\tag{A2} \] It is well known that A1 and A2 by themselves do not suffice to prove Cox's theorem. For instance, D. Dubois and H. Prade (1990) gave an example of a function Bel, defined on a finite domain, that satisfies A1 and A2, with \(F(x,y)= \min(x,y)\) and \(S(x)= 1-x\), but is not isomorphic to a probability measure. Thus, in order to prove Cox's theorem, to A1 and A2 there are needed some additional assumptions. This is the main purpose and contents of the present analysis. Various sets of assumptions under which the Cox's theorem can be proved are discussed, examined and provided, and the general conclusion of the author is that all of them are ``rather strong and, arguably, not natural''.