 |
 |
Hi,
First, you need a column of 1s in your design matrix.
Secondly, a general rule of thumb is that unless you are looking at a contrast to get at the overall mean (the column of 1s), then you do not need to mean center. So, if you're only interested in cognition, you do not need to mean center. If you did mean center it wouldn't hurt anything.
I cannot think of a reason why you would also need to scale by the standard deviation in the case of covariates in a linear regression.
Cheers,
JeanetteOn Sun, Apr 15, 2012 at 3:01 AM, Rosalia Dacosta <[log in to unmask]> wrote:
Dear FSL experts,
I have a little doubt about if I have to demean or not demean my EVs. I have been looking to the link of Jeannet Munford web, but I still have some doubts.My objective is to see if there is any gray matter region that correlates with some cognitive tests. So, I have the performance of my sample in a test, and I regress out genderd, age and years of education. Well, any of you could tell me if I have to demean all the covariates, including the raw puntuations of the test of every subject? The matrix design would be like this, where gender (males = 1; females = -1).
SUBJECTS
COGNITION
GENDER
AGE
YEARS EDUCATION
1
5
1
50
8
2
2
-1
60
9
3
4
-1
55
0
4
2
-1
54
0
5
6
-1
52
3
6
4
1
56
6
7
3
-1
53
6
8
6
1
54
6
9
0
-1
54
8
10
2
-1
55
8
11
3
-1
52
11
12
7
1
53
12
13
4
1
60
3
14
9
1
64
8
15
10
-1
65
8
16
12
1
62
0
17
7
-1
63
10
18
8
1
50
6
19
0
1
52
6
Contrasts:
C1 = 1 0 0 0C2 = -1 0 0 0Just one more question: some people introduce z values, instead of mean-centering...and I have not seen this in the web of Jeannette Munford....Any of you could tell me why introduce Z values instead of "simple" mean-centering??Many thanks.Rosalia.