Hi,

 

I am attempting to run do some FIR modeling and I keep running into a pesky error. The line of interest is 947 in spm_sp.m (at least in my version of SPM5):

 

varargout = {all(all( abs(sf_opp(sX)*c - c) <= tol ))};

 

This line checks the residuals of the regression of the contrast weight vectors using the design matrix. The comments have this to say:

 

“A linear combination of the parameter estimates is a contrast if and only if the weight vector is in the space spanned by the rows of X.

 

The algorithm works by regressing the contrast weight vectors using design matrix X' (X transposed). Any contrast weight vectors will be fitted exactly by this procedure, leaving zero residual. Parameter tol is the tolerance applied when searching for zero residuals.”

 

When running my analysis I see that most of the values are very near zero (or at least under the tolerance), however there will be a few (anywhere from one to a handful) that are very high comparatively. There will be, for example, several values of 0.5. As a consequence the script gives an error indicating that the contrast is invalid and quits. If I comment this line out I can run the analysis with no problem and the results are relatively coherent. Can anyone shed some light on what might be causing this problem?

 

Thanks so much!

 

 

Selwyn-Lloyd McPherson
Research Assistant, Brain Imaging & EEG Lab
NCIRE - The Veterans Health Research Institute