A characterization theorem for externally Bayesian groups (Q794356): Difference between revisions

This paper addresses itself to the problem of combining subjective probability density functions \(f_ 1,...,f_ n\) (with respect to some measure \(\mu\) on \(\Theta)\) of n individuals for a parameter \(\theta \in \Theta\). The author requires that the pooling procedure T commutes with the process of updating the \(f_ i's\) in the sense that \[ T(\ell f_ 1/\int \ell f_ 1d\mu,...,\ell f_ n/\int \ell f_ nd\mu)=\ell T(f_ 1,...,f_ n)/\int \ell T(f_ 1,...,f_ n)d\mu \quad [\mu] \] where \(\ell:\Theta \to [0,\infty)\) denotes the group's common likelihood function. Such a T is called externally Bayesian. It is shown that, under a simple regularity condition, \[ T(f_ 1,...,f_ n)=\prod^{n}_{i=1}f_ i^{w_ i}/\int \prod^{n}_{i=1}f_ i^{w_ i}d\mu \quad [\mu] \] where \(\sum^{n}_{1}w_ i=1\) and each \(w_ i\geq 0\), is the only externally Bayesian pooling operator such that \(T(f_ 1,...,f_ n)\propto G(f_ 1(\theta),...,f_ n(\theta))\) [\(\mu]\), where \(G:[0,\infty)^ n\to [0,\infty)\) is an arbitrary Lebesgue measurable function.