>But there is one question which we didn`t ask you yesterday: accepting
>the given-letters of yesterday we got a resting condition (A) and want
>to compare this to an activation (B) under the reconsideration of a
>covariate (performance), in this case we do not want to compare to a
>Choosing the contrast we are asked to give three numbers, considering as
>you told us the first two contrasts would be -1 1 , but if the object of
>interest is the influence of the covariate, which number is suppost to
>be the third?
>We are looking forward to another satisfying solution.
>thanks timm strotmann
(..and after I sought clarification...)
>Only during task B we could assess a subject`s performance. Now, we
>would like to
>use the scores as covariates. If I just compare condition B to the
>covariates I am
>still asked to give a contrast, and I do not know which kontrast to give
>1 or -1.
>i hope it makes our problem understandable
This sounds to be a not entirely straightforward paradigm: a hybrid of
block design and parametric design. There is an immediate problem with
having two different rows in your design matrix, one of which models
condition B as an 'epoch', the other of which models a series of scores
which only occur during condition B, with a score of zero given to the
model for every scan in condition A. Clearly these two rows would be quite
similar, in that both have a row of zeroes during condition A and a row of
positive numbers in condition B. To some extent, therefore, the two rows
will be confounded.
Is it possible to adjust your scores so that the mean score is zero? In
this case the parametric row will have a series of scores during condition
B which will sum to zero, and a series of zeroes during condition A, and it
will therefore be completely orthogonal to either of the 'epoch' rows.
If you have done this, and you have three covariates of interest in your
design matrix: 1. positive baseline shift during condition A (the rest
condition); 2. positive baseline shift during condition B (the task
condition) and 3. parametrically varying row, scores summing to zero, then
the questions which you could ask are as follows:
Contrast -1 1 0
This identifies voxels in which there is MORE activation during condition B
(task) than condition A (rest).
Contrast 1 -1 0
This identifies voxels which are LESS active during condition B (task) than
condition A (rest).
Contrast 0 0 1
This identifies voxels in which the signal has a significant positive
correlation with the subject's performance (regardless of whether there is
an overall change in baseline in condition B).
Contrast 0 0 -1
This identifies voxels in which the signal has a significant negative
correlation with the subject's performance.
Thus when you said that you have '... a resting condition A and want to
compare this to an activation (B) under the reconsideration of a covariate
(performance) ...' I think that part of the reason why I didn't understand
exactly what question you are asking of your data is that your question
breaks down into two distinct questions (with a positive and negative
contrast for each).
Personally I would consult one of the spm experts (I am a relative novice)
before embarking on such an analysis, as it is relatively complex. I will
copy this back to the spm discussion list, and perhaps you will receive a
more expert opinion,
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.