Dear SPMers, Dear Karl,

Many thanks for your prompt, kind and precise response (as always !).
However, since then I am scratching my head without finding a solution.
In fact I suspect that I misunderstood some basics about PPI 
(psychophysiological interaction).
Thanks to your file (spm_reagions), a small patch of an activated region 
under one contrast has been extracted.
11 subjects, scanned twice for a memory task : retrieval vs Lecture (R-L). 
A slightly different task has been asked between session n?1 and 2 (target 
vs context T-C).

First performing R-L
Extracting a small region using spm_regions
Multiplying it with the T-C regressor (centered)
Putting all (Y, Y*T-C, T-c) as covariate of interest.

I thought that Y and T-C should explain all the variance of the little 
region that was taken (there contrast have been set to 0).
In fact, this region persist to be activated (in fact it is the only part 
of it that still activated) !
How should I interpret that ?
The activity of this region is potentiated by is own activity under the 
factor T-C ?
Could it be explained by the fact that the 1rst eigenvector was taken ?
(By the way, I heard about some concern on the number of observation NO - 
number of scan - for a given number of variable NV - nbr of voxel. Is it 
true that NV should still greater than NO (I haven't found it in my book 
yet) ? Since my region only contain 26 vx and there is about 550 scans, 
could the use of the simple mean be more reliable ?).

Is it possible to mask the results with a write filtered statistic ?

A (last) question about the way to specify a regressor :
I intend to specify sessions too when performing this PPI. However I must 
than give 22 regressors *3 (3 for each session). Is there a simplest way to 
do that ?

Again I am loss for a word too say how thankful I am (and I may not be the 
only one) for the very high quality support that is provided on this list.
Thank you all, thank you Karl,
Very best regards



Jack


________________________________________________________________
|	Jack Foucher				Universite Louis Pasteur	|
|	Institut de Physique Biologique		UPRES-A 7004 du CNRS	|
|	4 rue Kirschleger			Tel: 33 (0)3 88 77 89 90	|
|	67085 STRASBOURG			Fax: 33 (0)3 88 37 14 97	|
|	France									|
|	Faster E-mail:	[log in to unmask]						|
|	Other		[log in to unmask]				|
|_______________________________________________________________	|



-----Message d'origine-----
De:	Karl Friston [SMTP:[log in to unmask]]
Date:	lundi 11 octobre 1999 19:19
A:	[log in to unmask]
Cc:	[log in to unmask]
Objet:	Re: Matrice visualization and covariates of no interest

Dear Jack,

> Is it possible to visualize the matrix of covariate of no interest ? and
> how to do it using SPM99 UI ? Or should it be done using image(xX.iG)?
>
> Is there any possibility to put a self designed covariate in the above
> cited matrix by UI or should it be done manually in xX.iG (can spm_append 
> be use)?

It is best to enter regressors using the GUI becasue there are other
flags and names that are created automatically.  To see the confounds
design matrix use

imagesc(xX.X(:,xX.iG))

> The regressor should be as explicative within the H or G matrix ? 
However,
> by setting its contrast to 0 when performing the T-statistics is its
> explicative power taken into account in the residual error calculation ?

Yes it is.  The contrast simply specifies the particular linear
combination or compound of parameter estimates you want to make an
inference about.  The estimates themselves are based on the full
model.

> By the way, what is block effect refer to ? sessions & subject ?

Both.  They are both examples of blocks.  In fMRI the basic block is a
session that may or may not correspond to a subject.

> What remains in C if conditions are modeled in H ? user specified
> regessors, voltera's ?

The only distinction between H and C is that H contains dummy or
indicator variables that code for condition-specific effects.  C
contains covariates.  In fMRI only C is used.


> Is there any reason to separate H&C and B&G ?
>
> % xX.iH         - vector of H partition (condition effects)      indices,
> % xX.iC         - vector of C partition (covariates of interest) indices
> % xX.iB         - vector of B partition (block effects)          indices
> % xX.iG         - vector of G partition (nuisance variables)     indices

Not mathematically no.

> At last are low frequency removed through regressor as in SPM97 or using
> classical filter (seems to be the case if spm_filter is used) ?

They are now removed using spm_filter.  However the mathematical
procedure used is identical.

With very best wishes - Karl





%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%