A new proof of the approximate sufficiency of sparse order statistics (Q1082749)

Let \(X_{i:n}\) denote the ith order statistic of a random sample of size n from a differentiable density f with support (0,1). When f is uniform use the notation \(U_{i:n}\) instead of \(X_{i:n}\). Take \(1\leq r_ 1<...<r_ k<n\), \(r_ 0=0\), \(r_{k+1}=n\) and let \(\phi\) : (0,1)\({}^ n\to [0,1]\) be any critical function. Then it is shown that \[ | E\phi (X_{1:n},...,X_{n:n})-E{\tilde \phi}(X_{r_ 1:n},...,X_{r_ k:n})| \leq \] \[ c[\sum^{k+1}_{j=1}(r_ j- r_{j-1}-1)(r_ j-r_{j-1}+1)^ 2/(n+1)^ 2]^{1/2}. \] Here \({\tilde \phi}=\int \phi (x_ 1,...,x_ n)dK_ r\) where \(K_ r\) is the conditional distribution of \(U_{1:n},...,U_{n:n}\) given \(U_{r_ i:n}=x_ i\), \(1\leq i\leq k\), and \(c=\sup | f'(y)| /\inf f^ 2(y)\). This is an improvement over Theorem 2.2 of the author, \textit{M. Falk} and \textit{M. Weller} [Statistical extremes and applications, Proc. NATO Adv. Study Inst., Vimeiro/Port. 1983, NATO ASI Ser., Ser. C 131, 597-610 (1984; Zbl 0583.62039)].