Hi Steve/List,
Thanks again for the advice. I apologise - I have a few more questions as I
feel my way through this, although I think(/hope!) I am finally getting the
idea.
With respect to your answer regarding the design matrix/contrast design for
analysis of surgery and changes in FA/MD correlations with psychology before
and after surgery, you suggested treating it a bit like a 2 tail test. So
would the following matrix and contrast be appropriate:
Ev1=pre gp Ev2=postgp Ev3=pre score Ev4=post score
Subj1-pre 1 0 a
0
Subj2-pre 1 0 b
0
Subj1-post 0 1 0
c
Subj2-post 0 1 0
d
Contrast:
0 0 1 -1 = where is correlation in presubjects greater than correlation in
postsubjects
0 0 -1 1 = opposite to above
0 0 1 0 = where are +'ve correlations present in pre subjects
0 0 -1 0 = where are -'ve correlations present in presubjects
0 0 0 1 = where are +'ve correlations present in post subjects
0 0 0 -1 = where are -'ve correlations present in post subjects
And would it be correct here to assume that there is no need to use -D as
the group means are modelled in Ev1 and Ev2. Of course Ev3 and Ev4 would
need to contain demeaned scores.
Would these need to be demeaned before (rather than after) padding with 0s ?
Finally if there was confounder such as IQ could one simply add this as
another event (Ev5) and contrast it as 0 in order to remove its confounding
effects in any of the above contrasts ?
Thanks in advance for your advice (+patience)
Mahinda
