Dear Cyril,
Thanks for your answer!
The correlation values I reported come from the SPM design orthogonality review, so I believe they correspond to the formula you mentioned. Just to make sure, I ran the correlations again using that formula and I got the same values SPM reports in the design review.
After defining the three contrasts of interests, I calculated the efficiency using the formula you pointed to
1/(c'*pinv((tmp.xX.X)'*tmp.xX.X)*c)
The values I get for (C vs D) within A and (C vs D) within B are very similar. The actual numbers hover around 0.22, but as far as i understand the numbers per se do not tell me very much, besides allowing me to compare different versions of the design matrix. Is this correct? When I do so by playing around with the SOAs, I've been noticing that the efficiency values for the contrasts I am interested in actually increase slightly as the correlation between individual regressors in the design matrix increases. Does that make sense?
|