Re: possible bug in errors in fitting routine

[Hans-Bernhard B. <HBB...@t-...>]

On 08.12.2012 21:00, Arun Persaud wrote:

> according to http://arxiv.org/abs/1210.3781 page 42 (between formulas D7
> and D8) gnuplot outputs the wrong errors when errors in the raw data are
> used during the fitting process.

It's a good deal less simple than the author of that paper makes it out 
to be.  The error reported by "fit" (as of the current development 
version: in its default state) only differs from the result desired by 
Mr. Young in the case where the data errors are clearly _not_ adequate, 
yielding a reduced chisq way off target.  Reporting the errors he wants 
would do the user an even bigger disservice in those cases than what 
gnuplot does now.

As a matter of fact, the Particle Data Book itself, possibly one of the 
biggest fitting jobs regularly performed in all of physics, uses the 
exact same procedue as gnuplot in those cases where the fits are too bad 
to be taken at face value.