T. Zielinski, R. D. Allendoerfer
Aug 1, 1997
Journal of Chemical Education
Least squares fitting of experimental data to a theoretical model is the corner stone of data analysis in many chemical laboratory programs. Even when data is to be fit to non-linear equations, students are routinely taught to linearize the data, apply linear least squares fitting to extract the chemically significant information, and to use the correlation coefficient of this fit to find the best theoretical model for the data. It is shown herein that when the data contains significant random experimental error, the linearization process biases the data giving potentially erroneous results and that the correlation coefficient is an inappropriate statistical measure of the goodness of fit in this situation. Modern personal computers are capable of non-linear least squares fitting and a procedure using Mathcad® is described to fit kinetic data directly to first and second order models and to use the F-test to choose the appropriate kinetic model. Pedagogy and methods of introducing this material into gener...