Dear Chris and Klaas,
thanks for your help and comments.
Spm_dcm_average actually estimates both the posterior mean and
the posterior covariance and stores it to disk as Chris pointed out. I
estimated both parameters for each group separately using spm_dcm_average. Now
I want to look at specific DCM parameters and compare them across groups; for example,
I want to find out if there is a between group difference of a task specific
modulation of a forward connection from one region to another. Therefore, I
entered these specific group parameters manually in the algorithm which
spm_dcm_review uses for computing the posterior density of contrasts of
connections and it worked quite well.
With spm_dcm_review I usually can only compare contrasts of
connections with one group of subjects which I included in a single spm_dcm_average
procedure (resulting in a single DCM_avg_* file). However, I want to do between
group comparisons and this is/was related to my actual question if I can use
spm_dcm_review also for between group comparisons (using parameters from group
specific DCM_avg_* files).
Best, Sascha.
Von: SPM (Statistical Parametric Mapping)
[mailto:[log in to unmask]] Im Auftrag von Christophe Phillips
Gesendet: Mittwoch, 22. Juli 2009 13:33
An: [log in to unmask]
Betreff: Re: [SPM] DCM parameter comparison between groups
Dear Sasha and Klaas,
Here is a snippet of code from spm_dcm_average:
% Average models using Bayesian fixed effects analysis ->
average Ep,Cp
%--------------------------------------------------------------------------
% averaged posterior covariance
final_iCp = sum(miCp,3);
Cp = inv(final_iCp);
% averaged posterior mean
weighted_Ep = zeros(length(cwsel),1);
for model = 1:num_models,
weighted_Ep = weighted_Ep + miCp(:,:,model)*mEp(:,model);
end
Ep = Cp*weighted_Ep;
% Copy contents of first DCM into the output DCM and insert averaged values
into parameter & covariance vectors
DCM
= DCM_first;
DCM.models = P;
DCM.Ep(cwsel) = Ep;
DCM.Cp(cwsel,cwsel) = Cp;
It does look like both posterior mean and covariance are
estimated.
Sasha,
make sure you're using the latest updates for averaging as there used to be
some problem with the averaging of the matrix D parameters, which only appear
when dealing with non-linear models (effect of one area on the connection
between 2 others).
Best,
Chris
Klaas Enno Stephan a écrit :
Dear Sasha,
If I understand your question correctly, then you would like to apply a linear
contrast to the posterior densities of two groups that resulted from using the
fixed effects Bayesian averaging routine in SPM. In principle, this
should be possible. However, you would have to compute this
manually. As far as I remember, the SPM_DCM_average function only stores
the posterior means in the DCM file it writes to disc, not the posterior
covariances; you would need both, however, to compute the contrast.
However, you should ask yourself whether for the particular question you are
interested in a fixed effects or random effects analysis is more
appropriate. Ultimately, this should determine which analysis you are
using at the second level. There were numerous emails on this topic in
the recent past which you should find easily in the archives.
Best wishes,
Klaas
Von: Sascha Fruehholz <[log in to unmask]>
An: [log in to unmask]
Gesendet: Samstag, den 18. Juli 2009, 18:08:23 Uhr
Betreff: [SPM] DCM parameter comparison between groups
Dear SPMers,
I have two groups of subjects and I applied the same DCM model for
both groups. I got DCM parameters for each group which are in accordance with
our hypothesis. I want to do statistical comparisons for these parameters (i.e.
DCM.B) between both groups. Therefore, I entered all single subject parameters
in a classical statistical inference test (t-test). The resulting statistical
group parameters are not the same as those which I get when I use DCM
averaging (spm_dcm_average) where group parameters are actually computed using
a Bayesian fixed effects analysis. More importantly, the group parameters
which result from the classical inference approach did not fit with our
hypothesis as good as those parameters from the Bayesian fixed effects
analysis.
My question is: instead of using classical statistical inference
tests for between group comparisons, is there a possibility to use for example
the same algorithm which is implemented in spm_dcm_review (which computes the
posterior density for contrasts of connections) for between group
comparisons? This algorithm obviously compares parameters within a DCM model
using the DCM.Ep and DCM.Cp parameters. Can I also take the group specific
DCM.Ep and DCM.Cp parameters and compare them by using the algorithm
implemented in spm_dcm_review?
Any help is highly appreciated.
Best, Sascha.
Checked
by AVG - www.avg.com
Version: 8.5.375 / Virus Database: 270.13.22/2253 - Release Date: 07/21/09
18:02:00