Dear SPM experts and community members:
I used the following codes to the estimate (beta) and variance for
parameter i in a PET study using design of SPM96: Multi-study, replicated
conditions (3 groups, each group has 11 subjects, each subject
has 2 scans for each of 3 conditions):
%asssume GLM is y=X*beta+e. The location, l, is in matrix space
beta=spm_sample_vol(smp_vol([sprintf('beta_%04d.img',i)]),l(1),l(2),l(3),0);
s2=spm_sample_vol(smp_vol(['ResMS.img']),l(1),l(2),l(3),0);
load SPM
p=diag(xX.Bcov);p=p(i);
variance=s2*p;
Question: When X is not full rank, what is the best interpretaion of beta
and its variances (especially if parameter i is a one not uniquely
estimatable)?
Thanks,
Kewei
|