A new multiple outliers identification method in linear regression (Q2175218)

The authors consider the problem of multiple outliers identification in the normal linear regression model \[ Y_{i}=\beta_{0}+\beta_{1}x_{i1}+\cdots+\beta_{m}x_{m1}+\epsilon_{i}, \quad i=1,\ldots,n, \] where \(\epsilon_{i} \sim N\left(0,\sigma^{2}\right) \) are i.i.d. random variables. The authors' abstract: ``A new method for multiple outliers identification in linear regression models is developed. It is relatively simple and easy to use. The method is based on a result giving asymptotic properties of extreme studentized residuals. This result is proved under rather general conditions on estimation procedure and covariate distribution. An extensive simulation study shows that the proposed method has superior performance as compared to various existing methods in terms of masking and swamping values. Advantage of the method is particularly visible in case of large datasets and (or) large numbers of outliers. The analysis of several well-known real data examples confirms that in most cases the new method identifies outliers better than other commonly used methods.''