Dear List,
Apologies if these are rather simplistic questions.
If I have a group of 5 subjects from the same group and want to assess
correlations between FA and psychology, whilst allowing for confounding
variables is the following valid:
Design matrix:
Gp Ev 1 Ev2 Ev 3
1 x y z
where
Ev1 = demeaned psychology score
Ev2 = demeaned confounder 1
Ev3 = demeaned confouder 2
Contrast:
1 0 0
-1 0 0
My questions are (and I have read conflicting posts in the forum):
1) I have seen posts on this topic (020856) where there is an additional Ev
with 1s the whole way down for the group, and within the contrast this is
set to 0. If I did that (eg Ev 4 = gp and set to 1 all the way down, with
contrast now 1 0 0 0 and -1 0 0 0) would that give the same results as bove,
and if not how are they different ?
2) If I had another psychology variable for which I wanted to view possible
correlations, and for which I wanted to use the same confounders, would I
have to redesign the above matrix with Ev1 containing the new demeaned
psychology variable. I presume if I simply added a 4th Ev4 column and put
the data in there, and set up a contrast 0 0 0 1 and 0 0 0 -1, this would
not work as it would treat the other psychology data (EV1) as a confounder
which is not the case.
Thanks.
Mahinda
|