There can be some value in comparing S values for different models with
the same data (eg ranking at least). A value much less than 1 (eg 1E-3)
indicates that you lost an "esd" column (or scaling factor) somewhere in
your data processing, and you should really try to find it again. Having
S>>2 means the model could somehow be improved, but not necessarily
during your lifetime, and perhaps without changing your conclusions
significantly.
Are you having trouble with a referee? Putting the data and model in a
powder CIF file for submission with the article might be better than
worrying about fit statistics. Having access to the refinement gives a
reviewer an opportunity to help...
Best,
Jon
Lubomir Smrcok wrote:
Hi,
S is, strictly speaking a valid criterion only under special
conditions which are not fulfilled in "Rietveld". My advice is to
forget about the value of S and think of physics, i.e. bond distances,
angles, occupancies and in favorite cases, also of displacement
parameters.
Lubo
On Tue, 3 Jun 2008, Franz Werner wrote:
Dear Rietvelders
I've a basic question on the Rietveld agreement index S (goodness of
fit) . In "The Rietveld Method" (ed. R.A. Young) it says in chapter
1.4 "An S value of 1.3 or less is usually considered to be quite
satisfactory. An S value of 1.7, for example, probably is a warning
that you should look further into the reasons and question the
adequacy of your model." In the literature, however, papers are
published quite frequently with considerably larger S values (I'm
just looking at an Acta Cryst. B paper with S values between 6 and 8.5).
The question is now how strict has the 1.3 limit to be applied, what
is the "largest acceptable" S value? How does one assess S>1.7 and
justify it to referees?
I should add that I'm not a follower of the R-value cult but think
that the most important criterion is chemical and physical plausibility.
Thanks for your advice.
Franz Werner
--
Der GMX SmartSurfer hilft bis zu 70% Ihrer Onlinekosten zu sparen!
Ideal für Modem und ISDN: http://www.gmx.net/de/go/smartsurfer
