Dear all,
I still have the problem I described before. I run various version of my design matrix but FEAT always stops at the post-stats stage with the following error message:
/opt/fsl_502/bin/contrast_mgr stats design.con
An exception has been thrown
Logic error:- detected by Newmat: index error: requested indices = 2, 1
MatrixType = Rect # Rows = 1; # Cols = 56521
Trace: ContrastMgr::GetCorrection; ContrastMgr::ComputeNeff; ContrastMgr::run.
In order to rule out any errors with my design matrix, I ran various parts separately. My design matrix includes 87 regressors, each modeling a specific phase (anticipation) of one trial of my experiment, 9 regressors, modelling a different phase (outcome) of the trials, each regressor containing all onsets of this phase of one condition (9 condition in total), and one confound regressor, including e.g. missing trials. When I run the matrix only with the first 87 regressors it works perfectly fine; if I run the 9 condition regressors with or without the confound regressor, it works as well. If I only run a couple of the 87 trials regressors, e.g. three of them, together with the condition regressors and the confound regressor, it works as well. But taking all 87 regressors plus taking one condition regressors, doesn't work as well as all 87 trial regressors and all other regressors together.
I tried these analyses with FLS 5.0.2. and with the newest patch 5.0.4.; I also tried it on different machines, one is a huge linux server (40GB working memory + up to 120GB swap; working memory is not fully used during the analyses), the other a Mac with Lion. No success.
I'm getting a bit desperate because I have no idea what could be the problem or how to continue. The only option I see at the moment, is to do separate FEATs for each single trial I want to model (i.e. 1 single trial regressor of the anticipation phase, combining the remaining 86 trials in one regressor and include the condition and confound regressors, then permutating the single trials to be modeled). However, this would mean that I have to do 87 FEATs for each of 24 subjects, which I think is a quiet inconvenient and time consuming solution.
Any comments/suggestion are more then welcome!
Cheers,
Susanne
|