Minimal L-space and Halmos-Savage criterion for majorized experiments (Q808119)

Let \({\mathcal E}=(X,{\mathcal A},{\mathcal P})\) be a majorized statistical experiment and \(L_ m({\mathcal E})\) its minimal L-space [cf. \textit{L. Le Cam}, Asymptotic methods in statistical decision theory (1986; Zbl 0605.62002)]. An equivalent majorizing measure \(\mu\) for \({\mathcal E}\) is said to be pivotal if it possesses the property ``a subfield \({\mathcal B}\) of \({\mathcal A}\) is pairwise sufficient and contains supports if and only if each \(P\in {\mathcal P}\) admits a \({\mathcal B}\)-measurable \(\mu\)-density''. [This concept was introduced to extend the Halmos-Savage criterion, cf. \textit{J. K. Ghosh}, \textit{H. Morimoto} and \textit{S. Yamada}, Ann. Stat. 9, 514-530 (1981; Zbl 0475.62003).] The authors show that if D is a maximal orthogonal system in \(L_ m({\mathcal E})\) then \[ \nu_ D=\sum_{w\in D}w \] is a pivotal measure. Conversely, every pivotal measure is of this type. Moreover, a representation of \(L_ m({\mathcal E})\) in terms of \(\nu_ D\) is given. Finally, it is shown that the \({\mathcal P}\)-completion of the subfield generated by all the densities \(dP/d\nu_ D\), \(P\in {\mathcal P}\), is pairwise smallest sufficient.