Re: What's left for a 4.2 release?

[Bastian M. <bma...@we...>]

Lucas Hart wrote:
=2E..
> I don't recall having seen any mention of incorporating patch=20
>  1445064  Gnuplot fitting improvements  2006-03-07
> submitted by Thomas Mattison after a number of exchanges on the
> mail list between he and HBB.
>=20
> The discussions re that patch brought out the point that currently=20
> the documentation and the fit output label are consistent, but those=20
> are inconsistent with the values output by the fit routine, i.e.,=20
> the documentation and fit output explicitly refer to parameter=20
> errors as "asymptotic standard error", i.e., as calculated from the=20
> variance-covariance matrix, with the intent that notation might cause
> the less experienced fitter to consult the documentation and become=20
> aware of the limitations in using those values to determine confidence =

> levels.
>=20
> However, fit.c has the comment
>  * HBB (br...@ph...) : fit didn't calculate the error=
s
>  * in the 'physically correct' (:-) way, if a third data column
>  * containing the errors (or 'uncertainties') of the input data was giv=
en.
>=20
> and the code
> 	/* scale parameter errors based on chisq */
> 		chisq =3D sqrt(chisq / (num_data - num_params));
> 			for (i =3D 0; i < num_params; i++)
> 				    dpar[i] *=3D chisq;
>=20
> Somehow, that discrepency was overlooked in earlier releases.
>=20
> I would present the asymptotic standard error and leave it to the user
> to determine any relationship between parameter errors as determined by=
=20
> the curvature of the chisquare hypersurface in the region of the minimu=
m=20
> and confidence limits for those parameters rather than apply an
> arbitrary scaling to the asymptotic standard errors.
>=20
Seconded. This scaling has been a major source of irritation to
to students in our lab courses.
But the scaling is not arbitrary. Effectively the errors of the
datapoints are scaled such that  reduced chi^2 =3D=3D 1.
That's very sensible if no data errors are supplied.

Btw. other widely used data analysis packages (including ROOT and
Origin) do not scale errors by default.

> That may be a point of contention - IIRC, the patch has a control
> variable which allows the user to select unscaled errors, as documented=
,
> or scaled errors, if such are desired, or for consistency with earlier
> gnuplot versions.

Personally, I would prefer new options to `set fit` instead of numerous n=
ew
FIT_xxx control variables. This would be much more consistent with the re=
st
of gnuplot.

--=20
Bastian M=E4rkisch
Physikalisches Institut, Universit=E4t Heidelberg