 |
 |
Dear Drew,
Re point 1. To deal with the multiple sessions per subject you can either (a) concatenate all sessions
into a single session and fit a single DCM for each subject or (b) fit a DCM to each session and then use spm_dcm_average to combine them into a single model. Perhaps others can comment on which is currently easiest in the software (Hanneke ?)
Re point 1, second part. The multiple subjects in each group are used in the paired t-test eg. 10 subjects
at day 1 and at day 3 gives you a paired t-test with 10 connection values in one group (day 1) and 10 connection values in another (day 3).
Re point 2. The point of having a modulatory input is to allow the intrinsic connections to be different
for one condition versus another. If your condition is a between session effect (ie day 1 versus day 3) then you should get the same results either (i) concatenating day1 and day 3 data into single time series and having a modulatory input that tells DCM which
are from day 3, (ii) just fitting separate DCMs to days 1 and 3 (with intrinsic connections only).
Best,
Will.
From:
Sevel,Landrew S [mailto:[log in to unmask]]
Sent: 14 January 2013 20:13
To: Penny, William;
[log in to unmask]
Subject: Re: Repeated-measures, group DCM
Professor Penny and SPMers,
I see how this would be a more manageable approach. In doing so, would there need to be any special treatment
in the averaging procedure to deal with multiple sessions per subject and multiple subjects in each group?
Additionally, if I were to compare models with a modulatory input for day 3 and it was found to have significantly
more evidence than a model without, would this invalidate comparison between purely intrinsic models on days 1 and 3?
Thanks for your suggestions,
Drew
From:
<Penny>, William <[log in to unmask]>
Date: Monday, January 14, 2013 7:16 AM
To: Landrew Sevel <[log in to unmask]>, "[log in to unmask]" <[log in to unmask]>
Subject: RE: Repeated-measures, group DCM
Dear Drew,
It may be easiest to forget about modulatory input and fit separate models on day 1 and 3 where both
have input and intrinsic connectivity only.
Your inference at the group level is then just paired t-tests on individual connections of interest.
If there is some uncertainty about the model you could integrate this out by doing within-subject Bayesian
model averaging and then running the paired t-tests on the averaged parameters.
Best,
Will.
From:
SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]]
On Behalf Of Sevel,Landrew S
Sent: 11 January 2013 19:01
To:
[log in to unmask]
Subject: [SPM] Repeated-measures, group DCM
DCM experts and SPMers,
Are there any extant resources describing repeated-measures DCM comparison for group-level
DCMs?
Specifically:
We are conducting a study with visits on days 1 and 3 (with three sessions on each day).
This is a baseline vs. stimulus on design. On day two, we will be introducing a manipulation. We are interested in seeing how this manipulation alters network connectivity at the group level.
However, we are having difficulty conceptualizing how to compare both time points.
We hypothesize that the day 2 manipulation will function as a B input on day 3. As I understand,
this would prevent direct parametric comparison between models on day 1 (no B input) and 3 (B input), regardless of structural similarities between the models.
Either way, it seems necessary to perform model selection on day 3 rather than assuming
the best-fitting model on day 1 holds on day 3.
Does anyone have experience with similar questions?
Best,
Drew