Accession Number : ADA117022

Title :   Robust Regression Procedures for Predictor Variable Outliers.

Descriptive Note : Technical rept.,

Corporate Author : SOUTHERN METHODIST UNIV DALLAS TEX DEPT OF STATISTICS

Personal Author(s) : Dorsett,Dovalee ; Gunst,Richard F

PDF Url : ADA117022

Report Date : Mar 1982

Pagination or Media Count : 47

Abstract : Least squares estimators of regression coefficients can be overly sensitive to violations of certain error assumptions; e.g., outliers in the response variable. One solution to the presence of outliers in a data base is to apply univariate robust estimation procedures to the residuals of estimated models. Equally problematic as outliers among the response variable are outliers or aberrant values for the predictor variables. Extreme values on individual predictor variables or an unusual combination of predictor variable values for a few observational units can distort least squares estimators even if the error assumptions are valid. This article discusses robust regression procedures, with special emphasis on techniques which are resistant to extreme predictor variable values. (Author)

Descriptors :   *Regression analysis, *Variables, Mathematical prediction, Coefficients, Estimates, Residuals, Errors, Resistance, Least squares method, Observation, Value, Abnormalities

Subject Categories : Statistics and Probability

Distribution Statement : APPROVED FOR PUBLIC RELEASE