Is There a Better Curve Fit for NDIR Calibration Data than 4th Order Polynomials?

This paper presents a technique for examining “Goodness of Fit” of polynomial least square curves using “errorless” data. (The errors in “real world” data tend to mask a polynomial curve's lack of fit). A method of assessing the “quality” of real data before attempting to fit a curve to the data is also presented. If data “quality” proves satisfactory, a cubic spline curve can be generated which provides a much better fit to the data than can ever be attained using polynomial curves.