ERROR ANALYSIS OF LINEARIZATION METHODS IN REGRESSION OF DATA FOR THEVANLAAR AND MARGULES EQUATIONS

Linearization is used extensively in regression and model testing of e xperimental data. In this work, Margules and Van Laar equation paramet ers have been calculated, for one sample data set, using several diffe rent linearization methods. The resultant parameter values were compar ed with values obtained by nonlinear regression. The increase in absol ute error and change of its distribution due to the transformation of the data were also analyzed. It is concluded that linearization may le ad to inaccurate or even completely incorrect parameter values which d o not describe the original data adequately. The commonly used statist ical tests may not detect the inaccuracy of the calculated parameter v alues. The ultimate test is the accurate recovery of the original acti vity coefficient data.