Dear Anderson,
I highly appreciate your quick answer. Kindly I have the following followup questions:
1. Why there's no need to demean age and sex in this case? Demeaning the variables keep confusing me as when to demean and when to not demean? I highly appreciate any input on this.
2. In GLM we have a column named groups (the first column in the model bellow). In GLM (gui) when we save a design ( e.g. design) this will output ( design.con, design.com and design.grp). "Design.grp" can be fed into randomise using the flag "grp".
Kindly, I would like to learn when I must add numbers to the groups column in the model. I mean (1) for group one, (2) for group two and so on .... I guess that when we add in GLM /groups column number (1) to all the groups, then we are dealing with one sample group mean (OSGM). is this correct? So when I must add a number for each group (i.e. 1,2,3 ,.....).
Best,
Jon
Hi Jon,
Yes, this is all fine.
There's no need to demean age and sex in this case, but doing so won't hurt either.
All the best,
Anderson
On 14 August 2016 at 03:11, Jon Anderson <[log in to unmask]> wrote:
Dear FSL experts,
I want to build design matrix and contrast for my TBSS data so I can use it in the command "randomise". The data is FA maps ( for 50 subjects) and the design model is for three groups and two covariates ( one continuous-AGE and the second one categorical-gender).
Regarding design.mat ( the example bellow is just for 9 subjects and 3 groups for illustration purposes)
Group EV1 EV2 EV3 EV4 EV5
input1 1 1 0 0
input2 1 1 0 0
input3 1 1 0 0
input4 1 0 1 0
input5 1 0 1 0
input6 1 0 1 0
input7 1 0 0 1
input8 1 0 0 1
input9 1 0 0 1
.
.
EV1, EV2 EV3 are the groups 1 , 2, 3
EV4 demeaned age
EV5 demeaned gender ( male =1, female=2)
The contrasts are
EV1>EV2 1 -1 0 0 0
EV2>EV3 0 1 -1 0 0
EV1>EV3 1 0 -1 0 0
positive EV4 effect 0 0 0 1 0
negative EV4 effect 0 0 0 -1 0
positive EV5 effect 0 0 0 01
negative EV5 effect 0 0 0 0 -1
Kindly is this model correct?
I highly appreciate your help!
Jon
