Bias-robust estimates of regression based on projections (Q688398)

The authors introduce a new class of regression estimators which are based on (univariate) regressions of the response variable with respect to all one-dimensional projections of the carriers. They are called projection estimators \((P\)-estimators). \(P\)-estimators are regression, affine and scale equivariant and are shown to have a ratio of consistency of \(\sqrt n\). These estimators are highly bias-robust and possess both bounded influence and high breakdown point. Numerical computations of the asymptotic bias and Monte Carlo estimations of the mean squared error under contamination for finite sample sizes show that \(P\)-estimators compare very favourably with other robust estimators.