I would not substitute zero for a NaN. Beta is a value of a slope. It
tells whether there is a linear relationship. If Beta = 0, then there is
no linear relationship, which is the null hypothesis in a one-sample test.
So substituting a zero for a NaN would prejudice the results in favor of
the null hypothesis. I think it is best to set breakpoints in the spm code
and see where the NaN is coming from when Beta is being computed and why
some voxels and resulting in a NaN and others aren't.
> Dear all,
> in order to correlate a test-score with the beta-values of my subjects, I
> read beta-values from the images, before averaging over my ROI for each
> subject. Betas in my ROI include some NaNs for some subjects. Why is this
> the case (different anatomy even after normalisation?)?
> How should I handle this in averaging? (set them to 0 as I do now or
> somehow take these voxels out of the analysis (how could this be done
> using Matlab-code?)?)
> Thanks. Best wishes,