Hi Sabine,
I don't think that specifying an identity V matrix will really help, if
none of your voxels survive the initial F test and masking procedure. You
should rather go and check whether your data and design is set up properly
(and perhaps lower any threshold masking options and defaults).
Yes, non-sphericity is estimated after the first pass of spm_spm. Then, a
second pass is performed using weighted design matrix & data. If you
choose non-sphericity options similar to SPM2, behaviour should be
similar.
There is a problem in non-sphericity specification that leads to a crash
in spm_reml.m, but your error message does not have its origin there.
There will be a bug fix for this with the next set of updates.
Volkmar
On Tue, 22 Aug 2006, Sabine Klein wrote:
> Hi all,
>
> it seems that a lot of people often have problems during the estimation of
> 2nd level statistics with SPM5. Sometimes estimation is just not possible:
>
> A warning popup appears:
> "Please check your data
> There are no significant voxels
> The globals are plotted for diagnosis"
>
> and then a crash:
> Error running job:
> Error using ==> plot
> Vectors must be the same lengths.
>
> There are a lot of unanswered emails regarding this problem in this SPM
> mailbase.
>
> I tried to find a workaround:
>
> After setting up the model for the 2nd level statistic in SPM5 the
> corresponding SPM.mat does not contain the non-sphericity matrix SPM.xVi.V,
> but it contains an array of non-sphericity components:
> SPM.xVi.Vi = {speye(nScan,nScan)}
>
> Setting up the same model in SPM2 yields to a structure with the
> non-sphericity matrix SPM.xVi.V, but without SPM.xVi.Vi:
> SPM.xVi.V = speye(nScan,nScan)
>
> Now I modified the SPM.xVi for SPM5 in the same way.
> This can be done e.g. by inserting
> if ~isfield(xVi,'V'),
> Vi = struct( 'form','i.i.d.','V',speye(nScan,nScan));
> end;
> after line 407 in spm_spm.m.
>
> This results in omitting the temporal non-sphericity estimation (behaviour
> like in SPM2). Can anyone tell me if this workaround is feasible? Or is the
> temporal non-sphericity estimation necessary? Maybe somebody has other ideas
> for a workaround...
>
> I have another question. Sometimes 2nd level statistic estimation with SPM5
> is possible. But I cannot say in which configuration. Sometimes a simple
> t-test works, sometimes not. When estimation is possible, the estimation
> procedure ist done twice with temporal non-sphericity estimation in between
> (the routine spm_spm.m is called twice!). Does anybody know why this is done
> twice and whether this is intended/necessary???
>
> Regards
> Sabine
>
>
--
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
|