Dear Dr Strotmann,
>I have got same questions working on a study, which has a design as
>follws: two groups (I and II) with two conditions (A and B) each, I
>would like to campare. Condition A is the same in both groups and
>condition B differs.
>I choose the contrast -1 1 -1 1, because condition A is a
>rest-condition. But what shows the result? Is this also the difference
>between group I and II?
Since what you have called 'condition B' differs between groups I and II,
it would perhaps be clearer to give it a different letter in each case, ie.
condition B for group I and condition C for group II (otherwise it is
tempting to think that you are describing a 2x2 factorial design!).
Your study therefore consists of two essentially different experiments, one
of which has been used to assess the effect of condition B vsa rest and the
other of which looks at the effect of condition C vs rest. As it happens
they share the same baseline rest condition. Let's assume that your
conditions are ordered group I cond A, group I cond B, group II cond A,
group II cond C.
The contrast -1 1 -1 1 is a sort of sub-optimal 'cognitive conjunction',
looking for voxels which are activated both by B and by C. However, this
would not be a good way of calculating a 'conjunction', since a voxel
within which there is a large difference between B and A may show up in the
spm even if this voxel shows no difference at all between C and A.
The best way to perform this analysis within SPM is to obtain the two
separate contrasts -1 1 0 0 and 0 0 -1 1 and then at the results stage
request the conjunction between them. This approach will eliminate from
the spm any voxel within which there is a significant interaction (at a
statistical level which you specify), ie. any voxel within which the
comparisons B vs A and C vs A yield very different results. However, it is
worth being aware that the ideal way to calculate a 'conjunction' is still
under review.
Obviously conditions B and C differ not only with regard to the task or
stimulus, but also with regard to the group of subjects. If you are using
fMRI, then you really need to use a random effects analysis,
Best wishes,
Richard Perry.
from: Dr Richard Perry BM BCh MA PhD MRCP(UK),
Clinical Research Fellow, Wellcome Department of Cognitive Neurology,
Darwin Building, University College London, Gower Street, London WC1E 6BT.
Tel: 0171 504 2187; e mail: [log in to unmask]
Pager: 04325 253 566.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|