Re: Least-Squares Fitting with asymmetric errors

["Stewart V. Wright" <svwright+lists at amtp dot liv dot ac dot uk>]

G'day Adam,

Firstly let me thank you for your comments regarding my question.

* Adam Kleczkowski <ak133@cam.ac.uk> [040506 10:54]:
> This is getting very off-topic (not related directly to GSL), so probably
> should be moved to private e-mail rather than being discussed in the GSL
> group.

I would offer a alternative opinion.  This question relates to fitting
of data using GSL.  If the functionality is not available then a
discussion of what is needed will help someone (possibly even me) to
write code to implement it.  My problem is a particular case of a more
general fitting issue.

For example a lot of experimental observables have a hard bound on one
side of a measurement (excluded due to kinematic constraints) and yet
on the other a 1 \sigma error bar which is quite large.  Admittedly
these results come from a boot strap analysis of a number of
measurements and that would be the ideal way to approach this problem.
However one doesn't always have access to _all_ the details of a
previous measurement...


Saying that, I don't want to start an argument over "what is relevant
for GSL", but I did want to point out that the real world is somewhat
more interesting (and limited) than the idealised cases that have been
implemented.   :-)


Cheers,

S.

Attachment: signature.asc
Description: Digital signature