An Investigation of Robustness to Non#Normality of Criteria for Testing Outlying Observations

摘要

Frequently in statistical analysis, an observational value is encountered that appears to deviate too far from the mean to be considered a part of the population under study. The problem of trying to determine what is “too far” has been the aim of many statistical writers since about 1850. These maverick values might appear at either or both ends of an array or there might be several such values in any sample. These questionable observations have been referred to in the literature as outliers, extreme values, gross errors, contamination and blunders.