Klaas - thanks for your helpful reply.
Already I have another query. A DCM that ran quite happily and gave
sensible results for the first 17 subjects (and the first 2
sessions of this subject) suddenly gave an error I have never seen before....
E-Step(-): 16 F: -2.515579e+03 dp: 1.120090e-02
E-Step(+): 17 F: -2.515579e+03 dp: 9.870216e-04
E-Step(+): 18 F: -2.515579e+03 dp: 6.959794e-05
E-Step(+): 19 F: -2.515579e+03 dp: 4.504073e-06
??? Input must be real.
Error in ==> erf at 21
y = erfcore(x,0);
Error in ==> spm_Ncdf at 86
F(Q) = 0.5 + 0.5*erf((x(Qx)-u(Qu))./sqrt(2*v(Qv)));
Error in ==> spm_dcm_estimate at 72
pp = 1 - spm_Ncdf(T,abs(Ep),diag(Cp));
Error in ==> DCM_batch at 48
spm_dcm_estimate (name)
Error in ==> dcm_overnight_target_PCCvmPFC at 53
DCM_batch(inDCM);
I am presuming that sqrt() there turned something imaginary...any ideas
why that might happen?
many thanks as always -
Chris
Christopher Summerfield
[log in to unmask]
On Thu, 7 Dec 2006, Klaas Enno Stephan wrote:
> Dear Chris and Julie,
>
> You should adjust your data for anything that is a known confound,
> e.g. realignment parameters. If you deliberately zoom in on a
> particular set of experimental manipulations and disregard other
> manipulations, then the latter could be regarded as confounds and
> removed through adjusting your data during VOI extraction. I would
> be a little careful with this though since it questions the rationale
> of the experimental design that you chose in the first place. There
> may be very good reasons to do this but it always depends on the
> specific application. I'd say there is no general rule here (except
> that all sessions should be treated identically in this regard).
>
> BTW, congratulations on your recent DCM paper in Science, Chris!
>
> All the best
> Klaas
>
>
>
> At 08:21 06/12/2006, you wrote:
> >Hi Klaas and other experts,
> >when defining a VOI, SPM asks whether we want to "adjust data for...
> >effects of interest" and several mails on this list indicate that it
> >is important to
> >do this when defining a VOI for DCM.
> >e.g.
> >http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind05&L=SPM&P=R580878&I=-3
> >
> >Our question is: in the situation
> >where you have several regressors in your design matrix, but not all of
> >them are included in your DCM model, should you correct for effects of
> >interest across ALL regressors, or only those which you want to include?
> >
> >Similarly, given that the VOI is defined on a session-by-session basis,
> >should you correct for effects of interest only for that single session,
> >or for all the sessions in your design matrix?
> >
> >And what are the various implications and assumptions of making these
> >different types of correction?
> >with thanks
> >Chris Summerfield and Julie Grezes
> >
> >
> >
> >Christopher Summerfield
> >[log in to unmask]
>
|