Robust Regression: A Diagnostic Tool

This report discusses the usefulness of robust regression when distorted observations have undue influence on the normal least squares regression fit. Instead of throwing out these distorted observations, Iteratively Reweighted Least Squares IRLS techniques are used to reduce the influence these outliners have on the fitted line. These robust techniques were utilized through a Robust Statistical Estimation Computer Package ROSEPACK that was developed at Massachusetts Institute of Technology. A detailed description of how to run this statistical package of the BRL CYBER system is also presented.