Hi Mirko
> I have situation very similar to that yo described in messages on
> February 4 and 5, 2007; at least I think so. I have two groups (A and B)
> and in each group I did measurements before and after (A1 A2 B1 B2). Now
> when I would like to use in a t-test contrast 1 1 -1 1 or -1 -1 1 1 the
> program is telling me that there is "Invalid contrast". I do not have
> any problems in using -1 1 -1 1 or 1 -1 -1 1. What am I doing wrong?
My mistake, sorry. To recap, as Sterling pointed out, the interaction
you are interested in is the contrast [1 -1 -1 1] which could report B
activating more or deactivating less.
Now, to distinguish the cases, I suggested masking by B>A with [-1 -1
1 1]. As you found, such a contrast is invalid. (Thinking about this
more, it makes sense because the subject effect columns of the design
matrix don't cancel out in such a case, so the contrast can't lie in
the row-space of the design matrix.
Again, as Sterling also said, you can test "the main effect of time
(where activation increases over time) regardless of group" with a
contrast of [-1 1 -1 1], which, as you found, is valid.
So it is this main time effect of overall activation that you can mask
your interaction (inclusively) by in order to distinguish B activating
more (i.e. [1 -1 -1 1] & [-1 1 -1 1]) from B deactivating less (i.e.
[1 -1 -1 1] & [1 -1 1 -1]).
I hope I've got things right this time, sorry for the confusion,
Ged.
|