Standard Article

Trimming and Winsorization

Statistical and Numerical Computing

  1. Paul S. Horn1,
  2. Karen Kafadar2

Published Online: 15 SEP 2006

DOI: 10.1002/9780470057339.vat027

Encyclopedia of Environmetrics

Encyclopedia of Environmetrics

How to Cite

Horn, P. S. and Kafadar, K. 2006. Trimming and Winsorization. Encyclopedia of Environmetrics. 6.

Author Information

  1. 1

    University of Cincinnati Ohio, OH, USA

  2. 2

    Indiana University, Bloomington, IN, USA

Publication History

  1. Published Online: 15 SEP 2006


Frequently, a sample may contain observations that differ substantially from the other data values. There are various ways to deal with the possibility of such values so that the resulting analysis will not be adversely affected if they are present. All of these methods assign weights to observations according to some well-conceived algorithm. In this article we concentrate on two forms of assigning weights to observations. The first is trimming, which corresponds to assigning zero weight to a prespecified fraction of the extreme observations and equal weights to the rest. The second is Winsorizing, which also assign zero weights to the extreme observations, as in trimming, but assigns extra weight to the most extreme observations in the retained sample. In this article we will define these concepts for a single sample of observations, and briefly discuss the applications to more complex problems like regression.