Hello There,
I'm having some trouble with a second-level F-test that I'm running using
FLAME1 mixed-effects analysis across a group of subjects. Each subject's
bold signal is represented by 5 basis functions. The 2nd-level EVs were set
up to estimate the mean value for each basis function coefficient across the
group, the contrast matrix is an identity matrix, and the 2nd level F-test
includes each of the EV's. I get a "matrix is singular" error (log excerpt
is below).
When I run the exact same design, but omit the F-test, everything works
fine. The same behavior occurs with OLS and FLAME1+2 analysis, but FE
analysis seems to work fine.
One non-standard thing with this data is that the registrations were done
ahead of time, so all the copes/varcopes were placed in a single "stats"
directory, with the reg/example_func2standard.mat set to identity matrix.
I'm attaching the design.con, design.mat, design.grp, and design.fts files
for reference in case there's a mistake there.
Thanks very much for any help you can provide, we appreciate it!
--Patrick
>>>>> BEGIN ERROR LOG HERE
Progress Report / Log
Started at Thu Sep 11 15:59:40 EDT 2008
Higher-level stats
cat ../design.lcon | awk '{ print }' > design.lcon
/usr/pubsw/packages/fsl/current/bin/fslsplit mask tmpmask -z
/usr/pubsw/packages/fsl/current/bin/fslsplit filtered_func_data tmpcope -z
/usr/pubsw/packages/fsl/current/bin/fslsplit var_filtered_func_data
tmpvarcope -z
/usr/pubsw/packages/fsl/current/bin/flame --cope=tmpcope0000
--vc=tmpvarcope0000 --mask=tmpmask0000 --ld=stats0000 --dm=design.mat
--cs=design.grp --tc=design.con --fc=design.fts --ols --nj=10000 --bi=500
--se=1 --fm --zlt=100000 --zut=100000
An exception has been thrown
Runtime error:- detected by Newmat: matrix is singular
MatrixType = Crout # Rows = 5; # Cols = 5
Trace: Crout(lubksb); GeneralSolvI; InvertedMatrix::Evaluate;
Gsmanager::f_ols_contrast; Gsmanager::ols_contrasts; Gsmanager::ols;
Gsmanager::run.
|