Christian Büchel wrote:

> Dear Russ,
> 1.5 out of 3 answers to your questions.
> 
> 
>>1. Upon reviewing the results following estimation, one is asked for a
>>threshold specified in Hz.  It's not clear exactly what is being
>>thresholded here, and why it is specified in terms of frequency?  Is
>>there a principled way to determine this threshold?
>>
> 
> The coupling is specifiued in terms of activity/s (Hz). In Bayesian inference,
> you specify a threshold of your effect (Hz) and you get a probability that the
> effect is equal or bigger than that. This is the same, for Bayesian inference in
> standard SPM2, where you can now ask questions like, "what is the probability
> that the signal change in any region is greater than 2%".
>

> 
>>2.  Right now it appears that DCM in SPM2b can only deal with a single
>>session at one time.  I would like to include data from multiple
>>sessions in order to increase my power; is it reasonable to
>>concatenate multiple sessions together and include a nuisance
>>regressor to model for session effects? I know the high-pass
>>filtering will not be as optimal, but I think that would be a fair
>>tradeoff.  However, I worry about the non-stationarity of the noise.
>>Alternatively, will the final SPM2 version of DCM allow multiple
>>sessions?
>>
> 
> Concatenation of time-series is certainly an option. I was wondering whether it
> would be possible to include further constraints in the identification of the
> system in that one constrains all effects (elements of matrices A, B, C)  to be
> the same for all sessions. This should be fairly straightforward to implement at
> the optimization stage.
> 


You could do a Bayesian fixed effects 
analysis over sessions by combining the 
posterior distributions over a given 
connection strength from multiple 
sessions. The resulting 'session 
posterior' then has a mean and variance 
that reflects the typical session 
response and (within-session) variability.

Alternatively you could treat the 
session as a random effect by simply 
doing a t-test on the estimated 
connection strengths (ie. treat the 
means like con* images).


> 
>>3.  I'm having a hard time getting my head around the construction of
>>contrasts in the DCM context - can one of you provide a brief
>>description of how this works?
>>



It does'nt ! There is no facility to use 
contrasts in DCM.

All variables that you wish to have 
either a direct or modulatory effect on 
the neurodynamics must appear explicitly 
as a separate column in the design 
matrix of the analysis.

Best wishes,

Will.


> 
> Me too!
> 
> -Christian
> 
> 
> 
> Dr. Christian Büchel
> Neurologische Universitätsklinik, Haus B
> Universitäts-Krankenhaus Eppendorf
> Martinistr. 52
> D-20246 Hamburg
> Germany
> Tel.: +49-40-42803-4726  Fax.: +49-40-42803-9955
> [log in to unmask]
> www.uke.uni-hamburg.de/kliniken/neurologie/pages/mitarbeiter/buechel_c.htm
> 
> 
> 


-- 
William D. Penny
Wellcome Department of Imaging Neuroscience
University College London
12 Queen Square
London WC1N 3BG

Tel: 020 7833 7478
FAX: 020 7813 1420
Email: [log in to unmask]
URL: http://www.fil.ion.ucl.ac.uk/~wpenny/