For others who might encounter the same problem, Rik was right -- I had duplicate files, and the "EIG" error disappeared/estimation was able to complete once I removed the dups. I still continue to get a warning that "Matrix is close to singular or badly scaled", but here is his response:
"> Is this something I should worry about?
I don't think so. It happens sometimes with my data too. Note that my code assumes (as does SPM in general) a single error covariance component that is constant across ("activated") voxels (see Friston et al, 2002) - estimated by ReML after pooling over such voxels (which is producing the warning). If you don't want to make this assumption, and/or want to partition the error into multiple components as in other stats packages, you could try Donald's Flex ANOVA.
R"
Thank you!!
|