[Ifeffit] Consultation on Reduced Chi2

[Matt Newville]

Jesus,



On Mon, May 23, 2016 at 7:08 AM, Jesús Eduardo Vega Castillo <
jevecas at gmail.com> wrote:

> Dear list,
>
> I am back with a new consultation on EXAFS fitting within Artemis.
>
> I have made a fit and obtained a reduced Chi2 value of 391 and R-factor of
> 0.014 using 9 variables. Then I have added two more paths increasing the
> number of variables up to 12 and then I obtained a reduced Chi2 of 3039 and
> R-factor of 0.007.
>
> I am a little surprised by the huge increase of reduced Chi2 while
> R-factor decreases down to half.
>
> What could be the cause of this large increase?
> Does it mean that it is not worth to add these new paths?
>
>
Reduced chi-square is scaled by dividing by the number of free parameters,
(Nidp - Nvarys).  Increasing the Nvarys from 9 to 12 without also
increasing Nidp (by adding data k or R range), could definitely make
reduced chi-square increase even if the R-factor and chi-square decrease.

One of the main uses of reduced chi-square is to determine if additional
variables are improving the fit well enough to justify their inclusion in
the model.

--Matt
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://millenia.cars.aps.anl.gov/pipermail/ifeffit/attachments/20160523/6d53d4ab/attachment.html>