International Conference of the ERCIM Working Group on Computational and Methodological Statistics (ERCIM) edition:7 location:Pisa (Italy) date:6-8 December 2014
Regression methods that are robust against outlying data points are now widely used in applied data analysis. Such estimators simultaneously identify observations with a large residual and downweight them for estimating the regression parameters. Even if only one component of an observation causes the large residual, the whole observation is downweighted, which results in a loss of information. We propose the shooting S-estimator as a first step towards regression in situations where a large number of observations suffer from contamination in a small number of (not necessarily the same) variables. This new estimator combines the ideas of the coordinate descent algorithm (also known as shooting algorithm) and simple S-regression, making it robust against elementwise contamination.