Print

---

Wang,

I can answer:

 

The reason you cannon generate xVi.V from xVi.h and xVi.Vi as you have shown above is because spm does not do a final update of xVi.V after the xVi.h hyperparameters converge.  In otherwords, the calculation of the xVi.V matrix that is actually saved to the SPM.mat file used the xVi.h hyperparameters from the iteration immediately prior to conversion.  It turns out this is not much of a problem, however, as the ReML algorithm converges when the hyperparameters change by less than a very small percentage.  That is, your final results will not be affected but as you have found you will not be able to generate the exact xVi.V from xVi.h and xVi.Vi.  To verify this effect is negligilbe just calculate a difference matrix using your equation above and the xVi.V in SPM.mat and you'll see very small differences throughout.

 

Best,

Brian

 

On 6/17/07, Volkmar Glauche <[log in to unmask]> wrote:

Dear Wang Liang,

the part of the SPM structure describing the estimated error covariance is
similar for all SPM designs, so it doesn't matter whether one looks at
first or second level statistics. The number of covariance components
estimated is just the number of elements in xVi.Vi or xVi.h.
I don't know what your second question is about... Of course, you will
only find xVi.h and xVi.V after you have estimated your model.

Volkmar

On Mon, 18 Jun 2007, Wang Liang wrote:

> Dear SPMers,
>  I have two questions for your help.
> 1).How to determine the number of variance components SPM provided when
> proceeding a first level statistic
> 2) Based on the xVi.h, xVi.Vi{i} and xVi.V in the SPM.mat file, I cann't
> acquire the equation xVi.V = xVi.h(1)*xVi.Vi{1}+ …… which is hint in
> spm_spm.m
>
> with best regards!
>
> wangliang
>

--
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                  http://fbi.uniklinik-freiburg.de/