Moments of the mean of Dubins-Freedman random probability distributions (Q1397971)

\textit{L. E. Dubins} and \textit{D. A. Freedman} [in: Proc. 5th Berkeley Symp. Math. Stat. Probab., Univ. Calif. 1965/66, 2, Part 1, 183-214 (1967; Zbl 0201.49502)] gave a method for constructing a random probability measure on \([0,1]\). The main ingredient in this construction is the base measure \(\mu\) on the unit square \([0,1]^2\). In this paper a recursive relation is given (Theorem 2.1) from which the moments of the mean \(M=M_{\mu}\) can be calculated for any given base measure \(\mu\). The cases in which \(\mu\) is the uniform distribution on (a) the vertical bisector \(D_1:=\{(x,y):x=1/2\}\), (b) the main diagonal of the unit square or (c) the horizontal bisector \(D_2:=\{(x,y):y=1/2\}\) are studied in detail. The support of \(M\) is considered; if \(\mu\) is concentrated on \(D_1\), then the support of \(M\) is convex. In general, Example 3.1 shows that the support of \(M\) may have an arbitrarily large number of gaps. Finally, if the base measure \(\mu\) is concentrated on \(D_1\), then the map \(\mu\mapsto M_{\mu}\) reverses the stochastic order \(\leq^{\text{st}}\).