On high breakdown point estimation. (English) Zbl 0933.62015

The linear regression model \(Y_i=\sum_{j=1}^pX_{ij}\beta_j+e_i\), \(i=1,2,...\), is considered. The aim of paper is to clarify the behaviour of high breakdown point estimators for this model. The performance of new algorithms for the least median of squares and the least trimmed squares is discussed. Some numerical examples are considered and it is shown that some caution is inevitable.


62F10 Point estimation
62J05 Linear regression; mixed models
62F35 Robustness and adaptive procedures (parametric inference)