Get access

An overview of the variables selection methods for the minimum sum of absolute errors regression

Authors

  • Carmen D. S. André,

    Corresponding author
    1. Instituto de Matemática e Estatística, Universidade de São Paulo, Rua do Matão, 1010, CP-66281, CEP 05315-970, São Paulo, Brazil
    • Instituto de Matemática e Estatística, USP, Rua do Matão, 1010, CP-66281, CEP 05315-970, São Paulo, Brazil
    Search for more papers by this author
  • Subhash C. Narula,

    1. School of Business, Virginia Commonwealth University, Richmond, Virginia 23284-4000, U.S.A.
    Search for more papers by this author
  • Silvia N. Elian,

    1. Instituto de Matemática e Estatística, Universidade de São Paulo, Rua do Matão, 1010, CP-66281, CEP 05315-970, São Paulo, Brazil
    Search for more papers by this author
  • Rodrigo A. Tavares

    1. Banco Itau S.A., Rua Boa Vista, 176-8o, andar-Corpo 2, São Paulo, CEP 01014-919, Brazil
    Search for more papers by this author

Abstract

The minimum sum of absolute errors regression is a robust alternative to the least squares regression whenever the errors follow a distribution for which the sample median is a more efficient estimator of location parameter than the sample mean, the errors follow a long tailed distribution, there are outliers in the values of the response variable in the data or the absolute error loss function is more appropriate than the quadratic loss function. Often an initial model may contain a large number of variables. However, in many situations, it is neither necessary nor important to include all the variables in the model. The methods for variable selection for the minimum sum of absolute errors regression are not as well documented and known as for the least squares regression. Our objective is to present an overview of the procedures to fit models with fewer variables and some criteria for selecting a model. Copyright © 2003 John Wiley & Sons, Ltd.

Get access to the full text of this article

Ancillary