Hi Helmut,
>2) what does
>[-1 1;
>0 0 -1 1]
>
>versus
>
> [-1 1 -1 1] test for?
Unfortunately I am not very familiar with multidimensional contrasts and
actually I do not know how to interpret an F-contrast like
[-1 1 0 0;
0 0 -1 1]
In case you weight only one column of your design matrix with 1 (or -1) at a
time (i.e. all other entries in the same row are zero), you will get an
answer to the question:
What do these columns in your design matrix contribute to explaining variance
in your data which cannot be accounted for by the other regressors?
In this case the columns with the ones are the reduced model (e.g. the effects
of interest span all columns of interest in your design), whereas the full
model includes all other regressors as well (e.g. constant terms).
But back to your question: If I were forced to interpret the F-contrasts
1) [-1 1 0 0;
0 0 -1 1] and
2) [-1 1 -1 1]
where column 1+3 belong to condition A and column 2+4 to condition B
my guess (!) would be:
1) Show any differences (positive and negative) between condition A and B
(both 1st session) AND any differences between A and B (2nd session).
2) Show any differences (again, pos and neg) between the "mean activation" in
condition A (irrespective of the session) and the "mean activation" in
condition B (irrespective of the session)
(I think mathematically "mean activation" is actually the sum of the parameter
estimates of conditions A and the same for condition B, respectively. And the
whole contrast is then the difference of these sums)
Cheers,
Thilo
On Wednesday 21 September 2005 17:41, Helmut Laufs wrote:
> Dear Thilo (anybody else welcome),
>
> Thank you for your reply. To clarify:
>
> 1) what does the contrast I suggested then exactly test for? (You said
> "compare a reduced model to the full one")
> 2) what does
> [-1 1;
> -1 1]
>
> versus
>
> [-1 1 -1 1] test for?
>
> Thank you again!
>
> Helmut
>
> > Dear Helmut,
> > I guess your contrasts compare a reduced model to the full one, which is
> > probably not what you want. To find out differences between conditions
> > irrespective of their signs I think it's appropriate to specify the
> > contrast-
> > "matrix" for the F-test as if it were a t-contrast, i.e:
> > [-1 1] or
> > [-1 1 -1 1]
> > but still doing an F-test.
> >
> > Please correct me, if I'm wrong....
> > Thilo
> >
> > On Tuesday 20 September 2005 19:47, Helmut Laufs wrote:
> >> Dear SPMers,
> >>
> >> to test whether there is any difference (i.e. irrespective of the
> >> direction
> >> of signal change) between two conditions, is it valid to do an F
> >> contrast like this:
> >>
> >> [-1 0;
> >> 0 1]
> >>
> >> ?
> >>
> >> ..and extending this to two sessions (repeated conditions) in one design
> >> (let's assume, there are only 2 conditions/colums per session):
> >>
> >> [-1 0;
> >> 0 1;
> >> 0 0 -1;
> >> 0 0 0 1]
> >>
> >> ?
> >>
> >> It looked straight forward to me in the first place, but now I am
> >> confused.
> >>
> >> Any help appreciated.
> >>
> >> Thank you.
> >>
> >> Helmut
> >
> > --
> > Thilo Kellermann
> > RWTH Aachen University
> > Pauwelstr. 30
> > 52074 Aachen
> > Tel.: +49 (0)241 / 8089977
> > Fax.: +49 (0)241 / 8082401
> > E-Mail: [log in to unmask]
--
Thilo Kellermann
RWTH Aachen University
Pauwelstr. 30
52074 Aachen
Tel.: +49 (0)241 / 8089977
Fax.: +49 (0)241 / 8082401
E-Mail: [log in to unmask]
|