EVALUATION OF LINEAR LEAST SQUARES COMPUTER PROGRAMS

Abstract

Two linear least squares test problems, both fifth degree polynomials, have been run on more than 20 different computer programs in order to assess their numerical accuracy. Among the programs tested were representatives from various statistical packages as well as some from the SHARE library. It was found that those programs using orthogonal Householder transformations or Gram- Schmidt orthonormalization were much more accurate than those using elimination algorithms. Programs using orthogonal polynomials proved to be superior to those using elimination algorithms.