We have discussed the handling of outliers here with the conclusion that removal of outliers requires some cautiousness and considerations of proper testing. By all means, it is not appropriate to remove data from a group simply because they seems to someone to be unreasonably extreme. However, if with proper testing and scientific reasoning the data appear to be incorrect, they should be eliminated from the group. In case of uncertainty in the existence of outliers, we can apply statistical procedure less sensitive to outliers. In the extreme case, the existence of dubious data should encourage us to repeat an experiment.
MaxStat offers the Grubbs‘ test for outliers applicable only to normally distributed data. This implies that one has to check whether the data show a normal distribution before applying the Grubbs test, and MaxStat is doing that automatically.
MaxStat offers the Grubbs‘ test for outliers checking normally distributed data for outliers. This implies that one has to check whether the data show a normal distribution before applying the Grubbs test. The test always determine the value which shows the largest absolute deviation from the mean. The procedure of the test is quite straightforward: we search the maximum of the absolute differences between the values and the mean. The result is divided by the standard deviation of the data group. If the resulting test statistic g is greater than a critical value, the corresponding value can be regarded to be an outlier. The critical value depends on the data size n and significance level. If an outlier has been identified and removed, the test must not be repeated even so some other, and expensive, statistical software packages practice this iterative approach.