Outlier

An outlier is an observation in a set of data that is inconsistent with the majority of the data. An [Page 980]observation (i.e., score) is typically labeled an outlier if it is substantially higher or lower than most of the observations. Because, among other things, the presence of one or more outliers can dramatically alter the values of both the mean and variance of a distribution, it behooves a researcher to determine, if possible, what accounts for their presence. An outlier can represent a valid score of a subject who just happens to represent an extreme case of the variable under study, it can result from failure on the part of a subject to cooperate or follow instructions, or it can be due to recording or methodological error on the part of the experimenter. If the researcher has reason to believe that an outlier is due to subject or experimenter error, he or she is justified in removing the observation from the data. On the other hand, if there is no indication of subject or experimental error, the researcher must decide which of the following is the most appropriate course of action to take:

• a.
  Delete the observation from the data. One disadvantage associated with removing one or more outliers from a set of data is reduction of sample size, which in turn reduces the power of any statistical test conducted although in the case of commonly employed inferential parametric tests (e.g., t test and analysis of variance), the latter may be counteracted by a decrease in error variability.
• b.
  Retain the observation, and by doing so risk obtaining values for one or more sample statistics that are not accurate representations of the population parameters they are employed to estimate.
• c.
  Retain one or more observations that are identified as outliers and evaluate the data through the use of one of a number of statistical procedures that have been developed to deal with outliers, some of which are discussed in the last part of this entry.