We recommend demeaning your confounding variables in both cases. Although it is also possible to get some analyses to work without demeaning it is usually easier and safer to demean your confounds. However, it is also important that you model the mean for randomise appropriately. For example, either (i) have separate group EVs - with 0's and 1's, or (ii) a group difference EV of -1's and +1's together with a mean EV of all +1's. The first case is often more natural and easier to setup.

On 27 Aug 2017, at 03:34, huyang <[log in to unmask]> wrote:

Dear FSL experts,

(1) In fMRI preprocessing, I want to regress out confounding variables like motion parameters and WM/CSF signals before functional connectivity analysis. I find that the command " fsl_regfilt -i input -o output -d design.mat" can do the work for me. My question is whether the variables (one variable one column) in the design matrix should be demeaned?

(2) To test two group difference by using randomise, the command is like "randomise -i input -o output -d design.mat -t design.con -n 5000 -T" and I also include (continual and categorical) covariates (such as age and gender) in the design matrix. My question is whether the covariates need to be demeaned without the use of randomise's "-D" option?

Many thanks!

Yang Hu