Hi Sanne,
If you have a column of 1's in the design, you don't need to demean
scanner, and any dummy coding (0 and 1, or -1 and 1, or 1 and 2,
whatever) will work. Please, have a look at Jeanette Mumford page on
mean centering for details: http://mumford.fmripower.org/mean_centering
All the best,
Anderson
On 14/01/2014 14:08, S Menning wrote:
> Hi,
>
> I am using randomise to analyse whole brain correlations between DTI measures (e.g. FA) and other measures (e.g. cognitive scores). Since data acquisition was performed on two scanners I want to include scanner site as a dummy variable.
> Regressors of interest should be demeaned in this type of analyses, should a regressor of no interest (scanner site dummy variable)also be demeaned? When comparing two analyses, one with the demeaned dummy variable and one with the 'normal' dummy variable, I found some differences.
> I hope that someone here can tell me which method to use (and why).
>
> Thanks!
>
> Sanne
|