Dear Hanneke,
(I hope that Maarten doesn't mind that I am writing...)
what you describe is exactly what Maarten did: He specified an "effects of
interest" F-contrast spanning all relevant columns of the design matrix
(EXcluding e.g. the intercept and realignment parameters).
Then Maarten extracted the time course of a region by clicking the
Eigenvariate-button and adjusting the data with respect to the above
mentioned contrast.
We were quite surprised that the mean of the time course did NOT vary close to
zero, because we expected the intercept (as well as variance related to
movement) to be subtracted from the raw data (i.e. the removal of the null
space of contrast
y=y-spm_FcUtil('Y0',SPM.xCon(xY.Ic),SPM.xX.xKXs,beta);
So the question is, if we should worry about the fact that the extracted time
course has a non-zero intercept, though we 1) chose the option to remove the
variance accounted for by the model-intercept and 2) applied the default
high-pass filter (as implemented in spm_regions).
Best regards,
Thilo
On Tuesday 24 January 2012 22:10, Hanneke den Ouden wrote:
> Dear Ashley,
>
> For the purposes of the inputs to DCM, you can set up the design matrix
> whichever way you like - you do not estimate this model, so parameter
> (in)estimability is not an issue. What happens is simply that SPM cycles
> through the regressors in the design matrix and asks you for each of them
> which one you want to use as modulatory or direct inputs.
>
> Hanneke
>
> On 6 January 2012 12:50, Ashley Safford <[log in to unmask]> wrote:
> > Dear SPMers,
> > I have a couple of questions regarding setting up a design matrix to use
> > in a DCM analysis. I have a 2x2 factorial design and for the DCM
> > analysis I would like to include the two main effects, the interaction,
> > all conditions and each level of the first factor separately.
> >
> > Main Effect A Main Effect B Int ALL A1
> > A2
> > FA-L1,FB-L1 1 1 1 1
> > 1 0
> > FA-L1,FB-L2 1 -1 -1 1
> > 1 0
> > FA-L2,FB-L1 -1 1 -1 1
> > 0 1
> > FA-L2,FB-L2 -1 -1 1 1
> > 0 1
> >
> > To set up these contrasts in the first-level fMRI model specification I
> > created a single condition that included all trials then created a
> > Parametric Modulation for each effect with an array of 1s, 0s and -1s for
> > values. My questions are:
> > 1. Is this the correct way to go about achieving the desired design
> > matrix? 2. This specification shows under parameter estimability that
> > some of the betas are not uniquely specified. Is this an issue for DCM?
> > And if so, how might I set up the model differently to avoid this but
> > still have all the desired effects to use as modulatory effects in DCM?
> >
> > Any help or suggestions would be much appreciated.
> > Sincerely,
> > Ashley Safford
> >
> >
> >
> > Ashley Safford
> > Neuroscience PhD. Candidate
> > George Mason University
> > [log in to unmask]
> > http://archlab.gmu.edu/people/ahamlin2/
--
Thilo Kellermann
RWTH Aachen University
Department of Psychiatry, Psychotherapy and Psychosomatics
JARA Translational Brain Medicine
Pauwelsstr. 30
52074 Aachen
Germany
Tel.: +49 (0)241 / 8089977
Fax.: +49 (0)241 / 8082401
E-Mail: [log in to unmask]
|