I may have read the initial email too quickly. Here are some points to
clarify the choices that I would make:
(1) If the conditions are the same, then you could use the second contrast:
a single row F-test replicated across sessions.
(2) All of my studies include multiple conditions and thus all use a
multiple row F-contrast for removing the non-neuronal signal from the BOLD
measurements - hence the assumption of the multiple row F-contrast in your
case. For each condition, and each derivative term (if in the model) and
parametric/time modulator (if in the model), you want a separate row. Thus,
most cases involve an F-contrast that is multiple rows. Each of these rows
will contain the contrast vector for a single condition.
The replicate feature only works if the column order is the same in all
sessions.
The difference between your two contrasts is a slight difference in how the
null space is computed.
Here is the formula:
null-space (e.g. Y0) = sX.X*(eye(spm_sp('size',sX,2)) -
spm_sp('xpx-',sX)*Q*b
sX=SPM.xX.xKXs;
b are the beta weights
Fc is the contrast structure.
If Fc.X0 is a structure:
Q=(spm_sp('ox',spm_sp('set',Fc.X1o.ukX1o))' *
spm_sp('cukx',sX))'*(spm_sp('ox',spm_sp('set',Fc.X1o.ukX1o))' *
spm_sp('cukx',sX))
If Fc.X0 is not a structure:
Q=Fc.c * pinv(Fc.X1o' * Fc.X1o) * Fc.c'
Best Regards,
Donald McLaren, PhD
On Mon, Jul 27, 2015 at 5:20 AM, Antonio Díaz Parra <[log in to unmask]>
wrote:
> Hi Donald, many thanks for your reply.
>
> Could you give me any detail else? I assume that the second contrast just
> removes the noise that is common for the three sessions whereas the first
> one individually removes the noise of each session . Nevertheless, when I
> use the contrast manager module (with weights matrix equals "1" and
> replicate over sessions equals "Replicate&Scale" ), the program
> automatically sets the F-contrast as a vector rather than as a matrix. That
> is the reason why I am a little confused.
>
> Kindest regards,
>
> Antonio
>
>
