I agree that there's nothing inherently wrong with the orthogonality matrix as it stands, and that looking simply at bivariate correlations among regressors is not the best measure of efficiency. Nevertheless, I'm under the impression that this is a common enough practice for examining different designs for collinearity issues prior to data collection. It doesn't take much more effort to export the design matrix and run corrcoef, but I wonder if many people, even to this day, aren't under the assumption that the orthogonality matrix reflects the correlation between conditions. I for one thought this was true of SPM2 until someone pointed out that the this was not the case.
It seems to me that a very simple solution to all this is simply to revert back to spm99's behavior and detrend the entire design matrix. After all, the nuissance regressors are mean centered, why not mean center the task ones as well?
That way the orthogonality matrix will be identical to corrcoef and everyone gets to have cake!
|