ABSTRACT
Whereas Chapter 1 was concerned with mild outliers, this chapter will focus on gross outliers. Their major characteristic is that they do not appear to be sampled from a population. Otherwise, they could be modeled by an additional mixture component. A modification to the previous estimators will now be introduced that turns out to be effective not only against noise but also against gross contamination: trimming. Trimming is probably the oldest method of robustification, having already been known to Newcomb [387]. It provides an omnibus outlier protection. The data identifies its own outliers. Trimming was introduced to cluster analysis by Cuesta-Albertos et al. [106].
