Robustness weight by weighted median distance (Q1775967)

The author deals with the problem of estimating a regression function by nonparametric (weighted) regression methods from a set of data contaminated by errors with heavy-tailed distributions. He proposes a new method for defining robustness weights based on the weighted median distance (WMD) of the response variables and compares its performance with the commonly used locally weighted regression (lower) due to \textit{W. S. Cleveland} [J. Am. Stat. Assoc. 74, 829--836 (1979; Zbl 0423.62029)]. The WMD method occurs to be rather simple in computation, it does not require iterations and can be easily generalized to higher dimensions. Its efficiency was compared with lower by a simulation experiment which demonstrated the advantages of the new method, particularly for heavy-tailed cases.