Hi
The output is a regression, not a correlation (that is, it's also sensitive to magnitude). Think of it as running a GLM against the stage 1 output matrix. The _Z files are then pseudo-Zs, i.e. effect size estimates divided by residuals.
If your 3rd stage randomisation step does not include the behavioural measure in the design then yes, in principle you can still play the 'null hypothesis' game and correlate the within cluster stats against behaviour.
hth
Christian
On 28 Jul 2013, at 21:52, "Carlos P. Preto" <[log in to unmask]> wrote:
> Hi FSL folks,
>
> I have a question about the dual regression stage2 output files. I understand that the output maps indicate the strength of correlation between a given voxel and the timeseries obtained at stage1 from the input component map, but what exactly do the voxel values in the dr_stage2_ic0000x and in the dr_stage2_ic000x_Z files represent?
>
> I'd also like to know if it would be methodologically sound to perform group comparisons using the dr_stage2_subject00000 file and subsequently use the significant clusters to obtain the correlation values from the Z-stat maps (rather than from the non-Z maps as those are not normally distributed) to perform correlations with behavioural measures outside FSL.
>
> Many thanks in advance,
>
> Carlos
|