If you have missing covariate values for shape analysis (eg for correlating shape change in different groups) can you model as you would for VBM, with an extra EV per missing value, or would you recommend that the input image created by first utils only contains data from the subjects who have a value for that covariate?
I have two behavioural values I’m interested in looking at between 3 groups. I have a complete data set for the first behavioural value, so the input into randomise contains all subjects' bvecs in that model. For the second value, not all participants have a value, so I need to model them out, or not include them, but I’m not sure which to do.
Can you advise which is more appropriate and why please?