Hello Mark, 

Thank you for your help.  

I understand, I will mean center the data as one group and then split the covariate for each group, and have four columns (column 1 group 1, column 2 group 2 , column 3 mean centered continuous variable for group 1, column 4 mean centered continuous variable for group 2).  I would not include contrasts 1 and 2, I will run them separately without the -D option.

Thanks again, 

Bedda

On Sat, Mar 23, 2013 at 3:35 AM, Mark Jenkinson <[log in to unmask]> wrote:

Hi,

This looks generally right, although you should not use -D in randomise here (as you are explicitly modelling the mean in your design matrix so there is no reason to demean the data - and if you do you cannot have contrasts 1 and 2).  The other thing is that you should not demean each subgroup separately for your covariate but should take all of the original covariate values as a single set of numbers, calculate the mean of this set, subtract this mean from all the numbers, and then split it into different groups and put them into the EVs.  Apart from that everything looks fine.

All the best,
        Mark

On 20 Mar 2013, at 00:17, Bedda Rosario <[log in to unmask]> wrote:

> Hello,
>
> I would like to run a TBSS analysis for FA data.  I have a continuous variable and two groups.  My interest is on the interaction, that is, continuous x group.  My design matrix has for EVs: EV1 - 1's for group 1, EV2 - 1's for group2, EV3 - demeaned covariate for group1 and EV4 - demeaned covariate for group 2.  EV1 and EV2 are not ordered as 0's and 1's or 1's and 0's because the files (as listed as ls) are in different order defined by another group variable.  I would appreciate if someone can take a look at my design files.and let me know if the design contrasts are correct (Interest: contrasts 3 and 4).  I will run the following command
>
> randomise -i all_FA_skeletonised -o tbss mean_FA_skeleton_mask -d design.mat -t design.con -n 500 -D -T
>
> Thank you,
> Bedda
>
>

> <design.mat><design.con>