Dear Becky,

I did not fully understand how you extracted your VOIs. Did you define
an F-contrast ("Effects of Interest") in order to adjust the data? And
if so, have you done this for both - the previous and the actual study?

The reason I ask is that you will need to adjust the data with respect
to their "individual" fit to the general linear model. If you look in
the code of spm_regions.m, which is called when extracting the data, you
will find some details in the section entitled "Computation".

If you choose to adjust your data, this will be done by applying the
parameter estimates of the specified F-contrast. These estimates of
course depend on your data and should not be used when changing the
time-series. So if you change the data, you should first fit the data to
the GLM and specify your "Effects of interest" with an F-contrast. After
that you can adjust your data using the according parameter estimates
and apply temporal filtering as well.

I am not sure if this is the problem, but it might be the case...

Kind regards,
Thilo

On Wed, 2012-05-09 at 19:40 +0100, Becky van den Honert wrote:
> Dear all,
> 
> I have a question regarding the field  DCM.X0. I would like to know how this field is created and what it does.  My understanding is that it is created during VOI extraction and is used to adjust the extracted timeseries to take into account confounds in the data (e.g. drift). In my study I am running DCM on timeseries from peak voxels that have already been identified in a previous study.  As such, I am replacing the timeseries extracted from VOIs identified in SPM with my timeseries of interest. The data used to extract VOIs in SPM are the same that were used to create my timeseries of interest, but different voxels may have been selected.
> 
> What should I do about X0 in this case?  I have tried using the DCM.X0 that is created by SPM. But perhaps this is problematic, because this DCM.X0 was created based on a timeseries that I am not using in my analysis.  I saw in spm_dcm_estimate.m that X0 could be replaced with ones instead. I have tried running my DCM analysis both ways (using X0 created by SPM and replacing it with 1s).  If I use X0 created by SPM, the 'actual' timeseries used to fit the DCMs look very strange; there is a huge spike at the beginning and end of my run (e.g. X0default.tiff).  If I replace X0 with 1’s, the spikes go away (e.g. X0ones.tiff).  Using X0 created by SPM versus replacing X0 also leads to strikingly different results in Bayesian Model Selection.
> 
> Please let me know if there is any additional information I can provide. Thanks for your help.
> 
> Becky