A problem oriented language for linear statistical analysis of full rank and non-full rank models



Journal Title

Journal ISSN

Volume Title



Multiple linear regression is a commonly used procedure for testing the fit of experimental data to an assumed model. There are many regression programs available that process data for the general linear model of full rank, however, it is unusual for these programs to also handle the non-full rank model. Often the rules for input specification to these programs is quite rigid making problem definition difficult. The University of Houston Analysis of Variance (UHAV) program is designed to overcome these disadvantages. The theory presented in this study is implemented in UHAV so that it will handle the non-full rank model as well as the full rank model. Linear combinations of the regression coefficients, such as the residuals and the regression sum of squares, are computed for the non-full rank model so that analysis can proceed. Included in UHAV is a problem oriented language for simplified problem specification for either model. The desired model can be selected automatically or specified by a UHAV statement. UHAV is written in Fortran IV so that it is relatively "machine independent".