Error Analysis of Linearization Methods in Regression of Data for the Van Laar and Margules Equations

Linearization is used extensively in regression and model testing of experimental data. In this work, Margules and Van Laar equation parameters have been calculated, for one sample data set, using several different linearization methods. The resultant parameter values were compared with values obtained by nonlinear regression. The increase in absolute error and change of its distribution due to the transformation of the data were also analyzed. It is concluded that linearization may lead to inaccurate or even completely incorrect parameter values which do not describe the original data adequately. The commonly used statistical tests may not detect the inaccuracy of the calculated parameter values. The ultimate test is the accurate recovery of the original activity coefficient data.