Hi Dr. Cox,
Thank you very much for your elaborate reply. I have been trying to orthogonalize the behavioral response regressors from our experiment to the SCM in hopes of getting from these regressors unique beta coefficients which represent unique variance as compared to the mean or IRF. I have been getting IRFs that change significantly as compared to those IRFs without the added regressors especially when the TS had great variability (unsmoothed). I felt it was the precision because of this but when I look at how orthogonal the vectors are from the SCM, I find they are no better than 10-6 when doing the dot product. Again I suppose they round to float, although MATLAB I think uses double precision also in its internal calculations. I was confused that maybe something was done in single precision.
thanks,
Linda Heidinger