An outlier may be detected by:
(1) a graphical plot of the data
(2) calculating the number of standard deviations it is from the mean
(3) an established statistical test such as the Chauvenet criterion
The outlier may represent:
(1) an error in measurement
(2) an error in data entry or transcription
(3) an error in the units used to express the result
(4) an error in calculation
(5) an issue irrelevant to the current study
(6) an unexpected value that can provide a new and valuable insight
A single outlier may affect the results of univariate and bivariate data analysis. Multiple outliers can distort an analysis, especially if the study size is small.
Responses to an outlier may include:
(1) repeating the analysis
(2) double-checking the data and its units
(3) increasing the size of the data set
(4) selecting a method of analysis that is not significantly affected by an outlier
(5) repeating the analysis with and without the outlier and comparing the results
(6) excluding the outlier (throwing it out)
Simly throwing the data out is the simplest to do but the least justified and the most subject to bias. The criteria for valid data should be established before starting a study.
