Multiple Regression Technique for Detecting Outliers

摘要

The least median of squares regression technique to lessen the impact of outlying observations is presented and alternatives are given. The output of a FORTRAN implementation of the technique, called PROGRES is illustrated. The results can be interpreted by graphical representation of the standardized residuals. It is shown how PROGRES can be used as a diagnostic tool in regression analysis. Simulations which compare robust and nonrobust regression estimators in different design situations are discussed.