ABSTRACT
The problem of outliers in random data sets is a very interesting, important and common one. Nevertheless there is no formal and generally accepted definition of what is meant by an outlier. Terms like outlier, spurious observation, contaminant, gross error and others are used with different and overlapping meanings. Two main purposes are in the focus when dealing with outliers in statistical data. One concerns the estimation of unknown parameters or other statistical inference which should be performed in an optimal way despite the possible occurrence of some outliers in the data. The other is the detection or identification of outliers in some sample as a main issue. The chapter presents a brief survey on outlier generating models and to show that different procedures can turn out depending on the choice of the outlier generating model. Under the assumption of the outlier generating models, all formulated as supermodels including the working hypothesis H0, one can derive tests for outliers.
