A New Method for Examining Rounding Error in Least-Squares Regression Computer Programs

Abstract

A new, easily applied method for examining any least-squares computer program for internal consistency is given. Only one data set is necessary, and by making appropriate transformations on these data results are given to provide an internal check on the accuracy (or consistency) of the given program. Accuracy becomes an increasing problem, of course, when the original data set is ill-conditioned. The theorems and proofs necessary to the method are presented as well as a sample data set for illustrative purposes.