Deat FSL experts,
I apologize in advance for this simple question on the randomise design matrix and contrast set up. It is my first experience of TBSS and FSL analysis and I just wanted to make sure I've set up the design correctly.
I have two groups of subjects (young and old) and one behavioral covariate (number of errors in navigation task).
I'm interested in finding the wm areas where there is any atrophy (YOUNG>OLD) contrast and which are positively correlated with the error parameter.
So I set up the design matrix with 4 columns to make sure we have separate variance for each group:first two columns (b1-b4) - behavioral parameter, lst two columns - first old group (3rd column) and next the young group (4th column).
b1 0 1 0
b2 0 1 0
0 b3 0 1
0 b4 0 1
etc...
The contrasts I requested were [1 1 1 -1] and [1 1 -1 1] Is that correct? Specifically I expect to find significant positive correlation for the second contrast.
As I understood from the previous posts I do not need to demean my data in that case but only when I use behavioral data for one-group analysis (see below)
Also, do I need to set up the group membership (-e <design.grp> option in that case?) I assumed yes.
A second related question is whether there any positive/negative correlation between beh data and FA across two groups.
in that case I demeaned my behav data and entered all the error scores in one EV. Since it is now a one-sample Ttest, do I just need one column (for behavior EV) in the design matrix and the contrasts of [1] and [-1]?
I greatly value your time and responses.
|