Tests to Evaluate Potential Outliers

Abstract

This last chapter deals with the evaluation of potential outliers. Outliers are different from out-of-specification data points and represent very extreme aberrant values. Outlies can occur with univariate or multivariate samples and may have disastrous effects on the inferential statistic when analyzing small data sets. Outlier tests are very valuable for in these situations, but may not be needed for larger data sets because the other values will soften the effects of potential outlies on the results of any statistical evaluation. The primary tests described in this chapter are Grubbs’ test and Dixon’s Q test for univariate situations and an evaluation of the studentized residuals for linear regression-type situations. Minitab applications for evaluating data points as potential outliers are presented.