scientific article; zbMATH DE number 3984326

From MaRDI portal
Publication:3747501





[For the entire collection see Zbl 0597.00024.] Let \(\pi (\theta)=\{G:\) \(G=(1-\epsilon)F_{\theta}+\epsilon H\), H is an arbitrary cdf\(\}\) be the family of cdf's with \(F_{\theta}(x)=F(x- \theta)\) and F unimodal. The set \(g=\cup_{\theta \in {\mathbb{R}}}\pi (\theta)\) will be referred to as the supermodel. For symmetric F \textit{P. J. Huber} [Robust Statistics. (1981; Zbl 0536.62025)] has shown that under g the asymptotically most bias-robust sequence of invariant estimators of the location parameter \(\theta\) is the sequence of sample medians. In the present paper the authors derive these most bias robust estimators without the assumption of symmetry. The estimators turn out to be of the form \(X_{L(n):n}-C\), \(n\geq 1\), for \(X_{i:n}\) being the i-th order statistic in the sample \(X_ 1,X_ 2,...,X_ n\) and for properly chosen L(n) and C.











This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3747501)