Abstract
Well known least squares procedure may have problems in several situations. In this paper we are interested ecpecially in two such features of the data multicollinearity and contamination presence.
This contribution recalls negative consequences of these two situations for the least squares method and also tries to explain shortly the reasons of the problems. Advanced methods which are being used for dealing with multicollinearity as well as robust methods used for outlier detection are also recalled as well as its main properties.
These methods work well on the type of data for which they were invented. However, when we have both problems present in the data, i.e. for exaple the situation where majority of the data suffers from multicollinearity and the rest is contamination, it shows up that presented methods are not longer suitable.
In the end of the paper we give also motivation why it is important to find some method which would be dealing with multicollinearity and outliers at the same time.