I am attempting to analyze an fMRI experiment in which a cognitive always occurs at the offset of a stimulus that is invariant in its duration. Ideally, I would like to deconvolve the BOLD signal in response to the stimulus and then the cognitive task, but I am concerned about collinearity.
When I evaluate the experiemental design using the -nodata option in 3dDeconvolve the program generates the inverse matrix, but the error is sizable and outputs the following message:
++ Matrix inverse average error = 0.0234375 ** WARNING!!! **
So, what is an acceptable average error (<.05 ?). If I were to proceed with this format of deconvolution what trouble might I expect to run into?
Vinny