Dear Sunali,
I fully agree with Klaas that you'd better estimate all your DCMs with
the same SPM version.
I think that the DCM.ID field (data identifier) has been introduced only
in SPM8 so you can use this information to figure out whether a
particular DCM.mat was estimated with SPM5 or SPM8.
Re-estimating everything with the latest SPM version would however still
be the best option.
Best regards,
Guillaume.
Klaas Enno Stephan wrote:
> Dear Sunali,
>
> It is not a good idea to apply BMS to models that have been estimated
> with different versions of SPM. There have been a number of
> improvements of the inversion scheme over the past few years, and these
> changes may lead to numerical differences in the free-energy
> approximation to the log model evidence. I would suggest that you
> re-run all models using the same SPM version (preferably the latest one).
>
> All the best,
> Klaas
>
>
> ------------------------------------------------------------------------
> *Von:* Sunali Wadehra <[log in to unmask]>
> *An:* [log in to unmask]
> *Gesendet:* Montag, den 1. Februar 2010, 3:57:48 Uhr
> *Betreff:* [SPM] Bayesian Model Selection for DCM in SPM8
>
> Hi everyone,
>
> I am having some problems with Bayesian model selection for DCM in
> SPM8. I have been using the random and fixed effects analysis, but the
> results have been unusual. Each model selected for the random effects
> analysis appears to have a very low log evidence in the fixed effects
> results, so much so that the model is not even reflected in the
> graphical representation. A possible problem is that some models may
> have been estimated in SPM5 and some in SPM8, though all models have
> been specified in SPM8. Could this affect the BMS result? Is there a
> way to check to see which models were estimated in SPM8 versus 5? I
> have tested the results out using different groups of models to see if
> the same pattern arises, particularly by grouping those that were
> definitely estimated in one version of SPM, and because it does, it
> seems as if there may be an additional technical issue. Does anyone
> have any suggestions on how I could approach this problem?
>
> Thanks for your help.
> Sunali
>
>
> __________________________________________________
> Do You Yahoo!?
> Sie sind Spam leid? Yahoo! Mail verfügt über einen herausragenden Schutz
> gegen Massenmails.
> http://mail.yahoo.com
--
Guillaume Flandin, PhD
Wellcome Trust Centre for Neuroimaging
University College London
12 Queen Square
London WC1N 3BG
|