Dear Ralf,
> I have a question regarding the use of basis functions. I have modelled
> 2 conditions with 2 basis functions respectively (Cosine set, epoch
> length 2, convolved with default hrf). Additionally I have modelled a
> time x condition interaction. In the design matrix we get 8 columns
> condA(1)[0] condA(1)[1] condA(2)[0] condA(2)[1] condA(1)[0] condA(1)[1]
> condA(2)[0] condA(2)[1].
>
> The first BF seems to me cos 0 (Box-car ?) and the second BF cos 1. Is
> it statistically valid to set up a contrast beween condition A and B
> with both BFs (1 0 1 0 -1 0 -1 0) or can I only compare the BFs
> separately (1 0 0 0 -1 0 0 0 or 0 0 1 0 0 0 -1 0) ? The same question
> arises with the time x condition interaction.
> I would be glad if anyone can give me some comments.
Normally one would restrict T-contrasts to test for effects expressed
in terms of the same basis functions. This is because the coefficient
of one basis function may not have any interpretable meaning in
relation to the coefficient of another. Therefore [1 0 0 0 -1 0 0 0]
and [0 0 1 0 0 0 -1 0] is the correct pair of contrasts or use an
F-contrast
[1 0 0 0 -1 0 0 0;
0 0 1 0 0 0 -1 0]
[1 0 1 0 -1 0 -1 0] places constraints (i.e. both coeficients should be
positive or negative] on the relative contribution of the two basis
functions that may have no sensible motivation.
I hope this helps - Karl
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|