Hi Mauricio,
For the covariates that you are not interested in testing the
interaction, they don't need to be split according to the groups. Just
one column for each.
All the best,
Anderson
On 15/01/2014 11:30, Mauricio Delgado wrote:
> .......................
> Dear FSL group,
>
> We are aiming to use Randomise for voxelwise inferences on TBSS data; however, we are struggling to set up the GLM design required for our hypotheses.
>
> We have two groups (patients and controls) and 3 covariates. We would like to assess mean FA differences between groups adjusted for Age, IQ, and Sex, while also examining a Group x Age interaction effect. Based on the examples on the GLM page of the FSL website, I suppose that for a test of interaction one should split the covariate of interest. So that would mean we should split Age in two EV’s (one for patients one for controls). However, I am wondering how the other two covariates should be defined in the model, should they also be split or is one EV per covariate sufficient?
>
> When setting up the contrasts, would the following be correct?
>
> Group Controls Patients Age_C Age_P IQ_C IQ_P Sex_C Sex_P
>
> 1 1 0 1.8 0 -14.98 0 0.3 0
> 1 1 0 0.8 0 -1.98 0 0.3 0
> 1 1 0 -2.2 0 -5.98 0 0.3 0
> 1 1 0 -2.2 0 3.02 0 0.3 0
> 1 1 0 -2.2 0 -6.98 0 0.3 0
> 1 1 0 -2.2 0 -1.98 0 0.3 0
> 1 1 0 -1.2 0 1.02 0 0.3 0
> 1 1 0 0.8 0 -1.98 0 0.3 0
> 1 1 0 0.8 0 8.02 0 0.3 0
> 2 0 1 0 -2.2 0 -1.98 0 -1.7
> 2 0 1 0 -2.2 0 -2.98 0 0.3
> 2 0 1 0 -1.2 0 17.02 0 0.3
> 2 0 1 0 0.8 0 -0.98 0 0.3
> 2 0 1 0 0.8 0 9.02 0 0.3
> 2 0 1 0 1.8 0 9.02 0 0.3
> 2 0 1 0 0.8 0 14.02 0 0.3
> 2 0 1 0 -2.2 0 -3.98 0 -1.7
> 2 0 1 0 -2.2 0 12.02 0 -1.7
>
>
> Contrast_1 Con>Pat 1 -1 0 0 0 0 0 0
> Contrast_2 Con<Pat -1 1 0 0 0 0 0 0
> Contrast_2 Group X Age 0 0 1 -1 0 0 0 0
>
> Thanks for your time and effort!
>
> Best,
> Mauricio
|