Dear AFNI experts,

I'm doing multivariate pattern analysis and computing both within- and between-subject correlation-differences for group analysis. Individual subject maps have a mean of zero and unknown variance under the null hypothesis. Correlations were fisher-transformed before computing differences to make them more 'normal'.

Looking at the between- and within- maps separately (computed using 3dttest++ with -setA), the distribution of t-statistics seem quite similar. However the effect sizes are stronger in the within- than between- maps, in other words the between-maps show lower correlation differences (they are computed using more data) yet also show less variability across participants. A direct comparison with 3dttest++ with -setA, setB and -paired gives a lot of blobs for the within>between just because it has higher mean values.

My question is how to compare the within- and between-subject maps, taking into account differences in variability. I have tried to normalize the variance separately and then run a standard paired ttest as follows:

3dTstat -sos -prefix within_sos.nii.gz within.nii.gz
3dTstat -sos -prefix between_sos.nii.gz between_sos.nii.gz

3dcalc -prefix within_nrm.nii.gz -a within.nii.gz -b within_sos.nii.gz -expr 'a/sqrt(b)'
3dcalc -prefix between_nrm.nii.gz -a between.nii.gz -b between_sos.nii.gz -expr 'a/sqrt(b)'

3dttest++ -overwrite -prefix withinVSbetween.nii.gz -setA within_nrm.nii.gz -setB between_nrm.nii.gz -paired

and these maps look kinda reasonable... but I wonder if this approach if kosher, and if not, whether there are alternatives.

Thanks for you consideration,
best,
Nick