Dear experts,
In the last few weeks we repeatedly ran into trouble using ANOVA within subjects, that we cannot understand. We'd appreciate any ideas that might help to solve this or explain what is happening here.
The problem: After running an ANOVA within subjects (i.e. estimation finished without errors), we are unable to generate contrasts.
Looking into the SPM.mat will yield unusual values in SPM.xX :
trRV = 0
trRVRV = 0
erdf = nan
Bcov = zeros(41)
Apparently something went very wrong during model estimation. However, there were no error messages or warnings.
We are a bit lost on interpreting this. Is this happening due to a problem within the data, is this a bug in SPM, or is this a relevant statistical result in itself?
A few more specific details :
We run a 2nd Level ANOVA over a total of 19 contrasts and 22 subjects. We have used this model to explore the data quite a bit and without problems. After some discussions we decided to test a few minor variations of the evaluation, that led to this problem, namely:
- Using a small eplicit mask in the 2nd level.
- Changing some a durations in the first level, recalculating the 1st level contrasts and re-running 2nd level.
The differences in the data between the working settings and the non-working settings seem pretty minor. Other 2nd level evaluations, e.g. simple 2nd level t-tests, run fine with all frist level contrasts and explicit masks.
Also: In the past few years we have used a number of similar models for exploring our experiments. We never ran into this problem before.
We are currently using SPM12(7219) and MATLAB 2017a on MacOS 10.12.6.
Thank you for reading and best regards,
Thomas
|