Dear Mike,
Firstly you need to have a 2x2 factorial design and
to have 4 regressors which model all the `cells'.
Say your two factors are A and B and the two levels of
each factor are A1, A2, B1 and B2.
Then your four regressors would correspond to the cells
1. A1-B1
2. A1-B2
3. A2-B1
4. A2-B2
Then to test for the main effect of A you'd use the
contrast, c=[1 1 -1 -1]. For the main effect of
B use c=[1 -1 1 -1]. You'd use an
F-test for this because, in SPM, t-tests are always one-sided.
The F-test on this (single-row) contrast corresponds exactly to
a two-sided t-test.
You'd use SPM t-tests to look for signed main effects.
Now to your question (at last !) - to test for an
interaction use c=[1 -1 -1 1]. Again you'd use an F-test.
Or use an SPM t-test to look for a positive or a negative
interaction.
(Note: instead of having the four regressors listed above you can
set up your design matrix to comprise the (two) main effects, the
interaction and a constant term. Then you can use just [1 0 0 0],
[0 1 0 0], [0 0 1 0] contrasts to test main effects and the
interaction. But then you'd need more complex contrasts to look
at simple effects - eg. B1-B2 at first level of A .....)
Best wishes,
Will.
Michael A. Yassa, B.A. wrote:
> Dear John and the SPM list,
>
> During multiple regression analyses in VBM, I would like to look at the
> contribution of the interaction term of two or more variables. Is there a
> way to do this using contrasts (t or F)? Or is there another procedure?
>
> Mike
>
>
>
--
William D. Penny
Wellcome Department of Imaging Neuroscience
University College London
12 Queen Square
London WC1N 3BG
Tel: 020 7833 7478
FAX: 020 7813 1420
Email: [log in to unmask]
URL: http://www.fil.ion.ucl.ac.uk/~wpenny/
|