Kris,
please reply to the list, since these questions may occur to several
people.
The disappearance of the saved job and the SPM.mat seems strange to me...
But for the covariance part: You can review the covariance components
also before estimation. On the page stating "Covariance not available",
there is a pagination button as on other SPM multi-page outputs. This
shows the specified covariance components. For 28 subjects, in my
experience there should be no more than 3-4 covariance components.
Btw - I don't really understand your design - do you have 28 different
regressors (i.e. columns in your design matrix)? This would not work
anyway, because there are no degrees of freedom left for the regression
analysis.
On Thu, 26 Jan 2006, Kris Knutson wrote:
> I can only review the design before estimating it, as when I estimate it, the
> SPM.mat file and the spm job I saved both disappear.
>
> When I do Review --> Design --> Explore covariates, that looks fine. It shows
> the covariates, one per fdg PET scan (person). I am only using one covariate
> (a score on a psychological test-- integers between 0 and 12) per subject.
> I've run it with 28 subjects and 28 covariate scores, and for debugging
> purposes with 4 subjects and 4 covariates scores, and get the same error each
> time.
>
> When I do Review --> Design --> Explore covariance structure, it says,
> "Covariance not available".
>
> I also varied the options for global normal, normalization, global
> calculation, and internal masking but always got the same error.
>
> Any help would be appreciated.
>
> Kris
>
> On Jan 26, 2006, at 10:19 AM, Volkmar Glauche wrote:
>
> > This message usually occurs, if you have specified unestimable covariance
> > components. You can review the specified components with "Review
> > design|Explore Covariance structure". Most probably, there are too
> > many
> > components. Try to restrict your assumptions about unequal
> > variances/dependent errors to the absolutely necessary factors.
> >
> >
> > On Thu, 26 Jan 2006, Kris Knutson wrote:
> >
> > > During a multiple regression of PET data, estimation step, I
> > repeatedly get the same error:
> > >
> > > Temporal non-sphericity (over voxels)... REML estimation.
> > > Warning: Matrix is singular to working precision.
> > > In spm_remL at 87
> > > spm_spm at 860
> > > spm_spm at 299
> > > ??? Error using ==> svd
> > > Nan or Inf prevents convergence
> > > Error in ==> pinv @ 29
> > >
> > > The error occurs even though the same PET scans work fine in a one
> > sample t-test.
> > >
> > > Any help pointing to the source of the error would be greatly
> > appreciated.
> > >
> > > Also, for the multiple regression (PET) option, why are they two
> > different places to enough
> > > covariates? What is the difference?
> > >
> > > Thanks.
> > >
> > >
> >
> > --
> > Volkmar Glauche
> > -
> > Department of Neurology [log in to unmask]
> > Universitaetsklinikum Freiburg Phone 49(0)761-270-5331
> > Breisacher Str. 64 Fax 49(0)761-270-5416
> > 79106 Freiburg
> >
> >
> Kris Knutson
> Psychologist
> CNS/NINDS/NIH
> 10 Center Dr.
> Bethesda, MD 20892-1440
> 301-402-6920
>
>
--
Volkmar Glauche
-
Department of Neurology [log in to unmask]
Universitaetsklinikum Freiburg Phone 49(0)761-270-5331
Breisacher Str. 64 Fax 49(0)761-270-5416
79106 Freiburg
|