Description: COMP is an interactive nonlinear least squares routine written in BASIC language and used to obtain estimates of parameters in nonlinear functions and to approximate their associated statistical errors. The program uses the linearizing (or Taylor Series) expansion of partial derivatives outlined in Draper and Smith, (1966, pp. 267 to 270). Therefore, partial derivatives must be supplied (as well as the function) by the user for any new models not currently contained in the programs. When a linearizing method is used to estimate parameters in a nonlinear model, all the usual procedures of linear regression theory can be applied. However, the results so obtained are only valid insofar as the linearized form approximates the true model. All of the statistics computed by COMP should be viewed with this restriction in mind. The output of COMP includes the variance-covariance matrix, t-tests for parameters, Von Neumann's ratio, observed, predicted and residual values, the error mean square, and an optional procedure to evaluate heteroscedasticity.
Date: November 1, 1977
Creator: Thomas, J. M.; Cochran, M. I.; Watson, C. R. & Eberhardt, L. L.
Partner: UNT Libraries Government Documents Department