Analysis of statistical outliers with application to whole effluent toxicity testing

In this analysis, low-value outliers were detected in five data sets obtained from laboratory records. The effect of removing the outliers by three methods of data rejection (asymmetrical and symmetrical trimming and Winsorization) revealed that all three methods slightly increased the mean and reduced the variance of the data sets. These opposing effects on the results of a hypothesis test of means were examined in the context of passing or failing a regulatory requirement of no significant difference between an effluent concentration and a control. Analysis by statistical resampling of one data set showed that while all data rejection methods reduced the level of Type II error in a hypothesis test of no difference in reproduction between a test concentration and a regulatory limit, asymmetrical trimming was the best in this regard.