Dear Giorgia
> we have a design with two sessions and we would like to test between- groups differences on the intrinsic and context-dependent connectivity of our selected model
> the question is: can we run two separate dcm for session 1 and 2 and consider session as a factor when we compare the values later on?
> or rather better to concatenate them?
> if the latter, can we extract one VOI from the concatenated sessions or better to extract two VOIs, one from each session, and later concatenate also the VOIs?
>
> sorry for the trivial problem, but we are having so sparse opinions about that and not being really able to track down the best procedure
This isn't a trivial problem :-) I will assume this is task-related DCM, rather than resting state. As you observe, you have two options.
The first option is to concatenate the data and include a regressor for session. This means that you could have parameters in your DCM coding for the effect of session on each connection. To concatenate sessions and add the session regressor(s) you would follow the instructions here - (https://en.wikibooks.org/wiki/SPM/Concatenation). In principle, this is ideal. However, off the top of my head, I think there would be some practical considerations. I don't think the DCM model specification GUI will allow you to select the session regressors as inputs, so you'd need to script this. Furthermore, I think the nuisance regressor matrix in each of your VOI files (xY.X0) might include the session regressors, meaning that DCM will regress out your session effects by default.
The second option is to specify separate DCMs for each session, and then model the commonalities and differences over sessions using a hierarchical linear regression model, implemented in the PEB framework. You could fit one PEB model to session 1, another to session 2, then a PEB-of-PEBs to compare sessions. See instructions at https://en.wikibooks.org/wiki/SPM/Parametric_Empirical_Bayes_(PEB) .
I think option 1 is more elegant and will be more statistically efficient. However, to get round those practical issues I mentioned, you might need some help with a script. I won't be able to get onto that for another week or so - let me know if that would be helpful.
Best
Peter
|