I am trying to understand the 'regress out' using randomize and need your
help on clarification of my approach. Supposedly the hypothesis is that age
has different effect on FA between groups.
To test the hypothesis, age is in the design matrix and demeaned between groups
EV1 EV2
-12.5144662 0
-2.2267942 0
13.9978628 0
0 -9.5596513
0 -10.0473223
0 -0.2555413
-13.6678902 0
5.0526578 0
-3.9555622 0
-1.1561092 0
-2.1692602 0
2.9238908 0
0 0.0485677
0 9.4184307
0 -7.1322543
0 9.7198007
0 5.0787047
0 5.6814447
13.7156708 0
0 -8.2993773
0 5.3471977
The group is in the confound matrix
1 0
1 0
1 0
0 1
0 1
0 1
1 0
1 0
1 0
1 0
1 0
1 0
0 1
0 1
0 1
0 1
0 1
0 1
1 0
0 1
0 1
, and the contrast matrix is
1 0
-1 0
0 1
0 -1
The results of suprathreshold cluster tests showed age is significantly
inversed with FA in clusters over different region in patient group, but not
in control. Can I draw the conclusion that age has more (or different)
inverse effect in FA in patient than controls? Is following mathematical
expression is what randomize does according to above design matrix?
That is considering
FA = lambda 1 + lamda2*group
To control the possible effect of age effect between groups,
FA = lambda 1 + lamda2*group +beta*age
Then the test if age is significant, that is,
Ho: beta=0 vs H1: beta not equals zero.
What not clear to me are those orthogonal EVs, demeaned age between groups,
which are not usually used in generic GLM. How would this approach
different from fitting linear regression models , (without demeaning) to
each group separately for evaluating the age effect ?
Best,
Ping-Hong Yeh
|