Goodness of fit and residuals
Tue, 04/20/2021 - 08:40 am
Just a general question.....When we attempt to fit a spectrum with various Gaussian bands, we can get an idea about the goodness of our fitting through observation of the residuals given on the top in Igor. The general idea is that the residuals should be more or less randomly scattered around 0 and i think that they should show a constant variance throughout the whole fitting range. Assuming that i would like to compare two different fitting efforts, would it make sense to claim that if fitting 1 develops non constant variance (funnel shape) after certain point or if it keeps constant variance but with a higher magnitude than fitting 2, then fitting 1 is definitely worse than 2? Is there anything else that should be also considered? There is the option of comparing the corresponding chi square values but i think that this is not necessarily informative if i am interested in determining the fitting quality ONLY FOR A PART of the whole fitting range. I hope my question was not very confusing.