Hi,
group subjects
EV1 EV2 EV3 EV4
1 1
1 0
0 1
1
2
1 0
0 0
1
3 1
0
0 0
2
4 0
1
0 1
2
5 0
1
0 0
2
6 0
1
0 0
3
7 0
0 1
1
3 8
0 0
1 0
3
9 0
0 1 0
my t-test and F-test contrasts are like
this:
C1 A-B 1
-1 0 0
C2 A-C 1
0 -1 0
C3 B-C
0 -1
1 0
OK, good.
C4
A-Sex 1
0 0 -1
C5
B-Sex 0
1 0 -1
C6
C-Sex 0 0
1 -1
These are junk... they don't make any sense. The gender effect
is in the model, and hence all inferences are controlled for gender (i.e. any
additive gender effects are set aside).
C7 A
1 0 0 0
F1
C8 B
0 1 0 0
F2
C9 C
0 0 1 0
F3
C10 Sex
0 0 0 1 F4
These are fine.
So, my questions are:
1) if the design matrix and contrasts are correctly
set up
See above.
2) Which results reflect the differences among the
three groups after having gender effects
controlled?
All inferences are controlled for gender.
3) Thomas mentioned that I can think about centering
the gender variable with group if I have lots of data. I do have around 80
subjects and I would really appreciate if you would give me more
specific information about this.
If
you would like to control for a group-specific gender effect, you need to fit
a group-by-gender interaction. This entails splitting the gender EV into
three groups *and* ensuring that the gender EV *within* each group are
centered.
For example, using the gender EV you gave above, this would
be
2/3
0 0
-1/3
0 0
-1/3
0 0
0
2/3 0
0
-1/3 0
0
-1/3 0
0 0 2/3
0 0 -1/3
0 0
-1/3
Hope this helps.
-Tom
____________________________________________
Thomas
Nichols, PhD
Director, Modelling & Genetics
GlaxoSmithKline Clinical
Imaging Centre
Senior Research Fellow
Oxford University FMRIB Centre