Dear Steve and list,
I recently encountered a similar misspecification when I tried to use
the "PET: Multi-Group: Conditions & Covariates" model. See also
http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind0501&L=spm&P=R53992&I=-1
For more complex designs under the PET module answering the two
questions for the non-sphericity correction sometimes leads to
non-sensical variance components. So, when using the PET module, make
sure that you always check the variance components in SPM.xVi.Vi before
estimation.
In your special case, I think you can get around this erroneous matching
of subjects from different groups in the variance components when you
use the Two-Sample T-Test in the fMRI module (under Basic Models).
There you have select ALL images at once and then specify a group vector
indicating which subject belongs to which group (e.g. [ones(1,15)
2*ones(1,10)]). For the non-sphericity options you should choose
"replications are over obs(ervations)" and NO correlated repeated
measures. Then you will obtain the correct variance components that
Alexa described previously.
If (in very special cases) you need other variance components that those
specified by SPM, you can add them to SPM.xVi.Vi prior to estimation.
Cheers,
Jan
On 2005-02-17 (Thu) at 11:29:19 -0000, Alexa Morcom <[log in to unmask]> wrote:
> Steve/ list -
>
> What exactly seems to have gone wrong with the nonsphericity specification?
>
> As I understand it, subject/ replication *is* a factor, it's just the random
> factor, and thus the one that covariance components for the group/s are
> estimated over.
>
> Therefore I believe you should have xVi.Vi (later SPM.xVi.Vi) = {[25x25
> double] [25x25 double]}, where: xVi.Vi{1} is [eye(15) zeros(15,10);
> zeros(10,25)] representing the covariance component for variance of the
> first group, and xVi.Vi(2) is [zeros(15,25); zeros(10,15) eye(10)]),
> representing the equivalent for the second group.
>
> But I can't see where subject 1 of group 1 is identified with subject 1 of
> group 2
>
> Any help?
>
> Alexa
>
>
>
> | -----Original Message-----
> | From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]]On
> | Behalf Of Stephen J. Fromm
> | Sent: 16 February 2005 14:24
> | To: [log in to unmask]
> | Subject: [SPM] Non-sphericity, PET "Compare-populations" model, and
> | factor indices
> |
> |
> | I'm trying to use the PET "Compare-populations: 1 scan/subject (two sample
> | t-test)" model.
> |
> | The confusing part is when I get to the non-sphericity questions.
> |
> | It's also not clear to me that spm_spm_ui('Files&Indices',...) is
> | correctly computing the factor indices.
> |
> | I have two groups, of size 15 and 10. This produces a 25x4 factor
> | matrix, "I". The last two columns are all ones. The first column,
> | representing "subject" according to D.sF, is of the form
> | [1:15, 1:10]'
> | The second column, representing "group," is
> | [ones(15,1); 2*ones(10,1)]
> | Subsequent code in spm_spm_ui seems to interpret the first column
> | literally as factor levels for factor "subject." But that doesn't seem
> | right. First, I don't see why "subject" is a factor with distinct levels
> | in a two-sample t-test (in the context of subjects drawn from two
> | populations, with inferences to be drawn at the population level).
> | Second, the code seems to identify subject 1 of group 1 with subject 1 of
> | group 2, and on up to subject 10 of group 1 and 2, when in fact these are
> | distinct subjects. This then leads to what appears to be incorrect
> | specification in the non-sphericity module. (Since the subjects are
> | distinct, there are no repeated measures, but of course the two groups can
> | have different variances.)
> |
--
Jan Gläscher Neuroimage Nord
+49-40-42803-7890 (office) Dept. of Neurology, Bldg S10
+49-40-42803-9955 (fax) University Hospital Hamburg-Eppendorf
[log in to unmask] Martinistr. 52
20246 Hamburg
Germany
http://www.uke.uni-hamburg.de/zentren/neuro/neurologie/mitarbeiter/glaescher_jan.html
--------------------------------------------------------------------------------------
GnuPG/PGP key id: FEC4B55C
fingerprint: 5A36 1EF6 8472 117E 805A F240 3146 A410 FEC4 B55C
--------------------------------------------------------------------------------------
|