A method of reducing the effect of systematic and/or random noise during determination of dependent variable values, (eg. pseudo "n" and "k" and/or ellipsometric PSI and DELTA or mathematical equivalent vs. wavelength or angle of incidence), involving selecting a mathematical function and an independent variable subset range combination so that a square error best fit with total summed square error over the independent variable subset range being minimized, zero or within an acceptable range near zero, are achieved.