This question is closed. Reopen it to edit or answer.
Making a Fit Procedure Which Satisfies:
1 view (last 30 days)
I wonder if there exists a fit procedure which satisfies (or better.. could be built):
1) Predicts Linear Relation to Data (linearity)
2) Goes through Origin (0,0) [to be clear]
3) Takes Error into account (Poisson, btw.. it would be great if there were an advantage to knowing this a priori as well.. one can dream, no?)
4) Iterative--not necessarily in the purest form of performing the fit itself.. but merely a process whose accuracy tends to increase (such as a series sum might) when 'bad' data are cut (maybe some data are taken when machinery were technically failing)
Does such a procedure exist? Is it already posted to file exchange? Do you know of a reference which might address these issues and know the author or title.. or both?! ;)
Thank you in advance for any advice (Or code) you may provide! If it does I'd be interested in automatically computing the chisq, residuals, uncertainties due to the fit (and data in whichever case it may apply).
P.S. This is more like a wishlist that I may gather elements for over a long period so please don't be shy if you can provide an answer to any portion of this question.. BUT, I am aware that polyfit0(xdata,ydata,1) [no error form assumed] does a fine job of computing the fits and errors assuming data go through the origin, even if their dof is incorrectly calculated (off by 1).
Thanks again for reading!