		More Information
	Back to Aabel 3 Overview Back to Aabel Statistics

Outlier Analysis

Outlier detection is import for two reasons:

Outliers can provide valuable information on data quality or highlight atypical patterns.
Outliers tend to pull the mean vector towards themselves and inflate the variance in their direction, and can hence influence the result of tests that are based on comparing means or include variances in computational procedures.

Mahalanobis Distance and Jackknifed Mahalanobis Distance

Mahalanobis distance is a widely used method of multivariate outlier detection.
Jackknifing is a process where the multivariate distance for each observation/object is calculated using means, variances, and covariances that did not consider (i.e., not influenced by) the given observation.

Outlier Analysis of Proportion or Percentage Data

Outlier analysis of proportion or percentage data, i.e., data where the row sums of the selected variables have a constant sum, requires log center transformation of the data.

Aabel provides a number of pre-processing data transformation methods accompanying the outlier analysis, which include log centering and other methods that are appropriate for outlier analysis using Mahalanobis distance and Jackknifed Mahalanobis distance.