scientific article; zbMATH DE number 19007

From MaRDI portal
Publication:3976432





[For the entire collection see Zbl 0731.00018.] A new class of regression estimates is introduced which are highly bias- robust. These estimates, called projection estimates (\(P\)-estimates), are based on univariate regressions of the response variable on all one- dimensional projections of the carriers. A corresponding class of estimates of a multivariate scatter matrix, which are also based on projections and which are needed for the the treatment of one of the projection regression estimates, is introduced, too. The authors give a lower bound for the maximum asymptotic bias of any equivariant and Fisher-consistent regression estimate and review recent results about regression estimates which minimize the maximum asymptotic bias under contamination within the classes of \(M\)-estimates with general scale and \(GM\)-estimates. Finally, the maximum asymptotic biases of the new estimates and the minimax \(M\)- and \(GM\)-estimates are compared numerically. The bias of one of the \(P\)-estimates is close to the lower bound.











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 Q3976432)