Hi Anderson,
Thank you for the reply and my apologies for posting without much explanation - it appears that the text I included in my original message field did not show up in my actual posting. I'm including that text now to provide background on the experiment and design.
I am conducting a study in which participants from either a control or an experiment group undergo repeated fMRI sessions (pre- and post- intervention). The participants complete a block design task with the subject-level contrast of Reasoning>Matching. I'm running into a problem with the highest-level analysis. We are running a 2x2x2 (group x gender x session) mixed design ANOVA and, at the moment, I am attempting to run the analysis in FEAT with a simplified design matrix involving only 8 participants (2 from each combination of factor levels). I'm running a simple model to start, but it looks like I am running into a rank deficiency problem, as you have pointed out. I am unsure how to modify my design matrix -- any suggestions you might have would be greatly appreciated!
Within the FSL GUI I am receiving the warning:
"at least one EV is (close to) a linear combination of the others. You probably need to alter your design. (Design matrix is rank deficient - ratio of min:max eigenvalues in SVD of matrix is 2.0843e-17) Contrasts involving these combinations will be set to zero"
... and I've done the calculation of det(X*X') in Matlab and I am getting a determinant of zero, so we are rank deficient here. I'm not sure how to modify my design to correct this. This matrix was actually generated using the GLMflex program via the example at http://mrtools.mgh.harvard.edu/index.php?title=GLM_Flex_Fast2#How_to_use_GLM_Flex_Fast2, so I had high hopes that it would work.
The matrix I posted has EVs that correspond to (1) group, (2) session, (3) gender, (4) group x session, (5) gender x class, (6) gender x session, (7) group x gender x session, and (8-14) model the individual participants (which is needed as this is a repeated measures study). I am interested in testing for gender and group differences across time. Specifically, I would like to see if one gender responds to the treatment differently if they are in different groups. This is why we are exploring the three-way interaction.
Do you have a suggestion on how I might modify my design matrix to fix the rank problem?
Thank you very much!
Jessica
|