 |
 |
Hi Anna, Please, see below: On 16 November 2015 at 18:00, Anita van Loenhoud <[log in to unmask] <javascript:_e(%7B%7D,'cvml',[log in to unmask]);>> wrote: > Dear all, > > > > I want to do a simple voxelwise correlation (both negative and positive) > with education (i.e. only one predictor) in a group of 511 subjects. > Education here is actually an ordinal variable, in which 1=lowest education > and 7=highest education. My design.mat looks like this: > > /NumWaves 1 > > /NumPoints 511 > > /Matrix > > 7 > > 6 > > 6 > > 4 > > 4 > > 4 > > 7 > > [etc] > > > > My design.con looks like this: > > /NumWaves 1 > > /NumPoints 2 > > /Matrix > > -1 > > 1 > > > > My randomise command looks like this: > > randomise -i 4D.nii -o edu_correlation.randomise -d design.mat -t > design.con -n 500 -D -T > > > > I have a few questions: > > 1. Are my design.mat/con correct? Should I add PPheights=7 in design.mat > and PPheights=1 in design.con? > Fine as is, no need for PPHeights. > 2. Should I demean my data or not? I had a hard time finding that out in > the manual/mailing list. > Yes, need -D since no intercept is modelled, and the interest is on the continuous variable. > 3. Should I have added a GM explicit mask using –m (since I am only > interested in gray matter voxels)? In principle, the images in the 4D.nii > already only display GM voxels, so maybe a mask is not necessary? > If the input is already masked or already contains just what is of interest, no need for a mask. > 4. I read that when you do a ‘one sample t-test’ you should use -1; does > that apply to my data since I have one instead of two groups (I thought it > doesn’t)? > A permutation test works fine for this design, even with just 1 group, with the continuous variable (so, no need for the -1). All the best, Anderson > > > Thanks a lot, > > > > Anna > -- Sent from mobile. Please forgive the conciseness.