 |
 |
Dear Tor, > I have hopefully just one more quick follow-up to this discussion. It's > about the differences between specifying the kind of F-contrast we > discussed earlier with regards to Volterra analysis: > > A B A*A A*B B*B > > > 1 0 1 0 0 to see which voxels respond to A, and > > > 0 1 0 0 1 to see which voxels respond to B, > > ... and the alternative kind of F-test, > > 1 0 0 0 0 > 0 0 1 0 0 To test the effects of A, and a similar set of 2 contrasts > for B. > > The first one should test the average of the betas for the linear and > squared term of condition A, and the second should test whether the linear > and squared terms together model a significant amount of variance > (right?). Would one generally be considered more appropriate in a > Volterra analysis than the other? Yes - the second. I am sorry but I did not spot this earlier. A B A*A A*B B*B 1 0 1 0 0 to see which voxels respond to A, and 0 1 0 0 1 to see which voxels respond to B, is wrong and should (as you point out be) 1 0 0 0 0 to see which voxels respond to A, and 0 0 1 0 0 0 1 0 0 0 to see which voxels respond to B. 0 0 0 0 1 > Also, if I want to test the DIFFERENCE between A and B, I have to specify > a different kind of contrast? The way to test for a difference is to remove the commonality. This would be most simply effected by doing another analysis where you modelled A and B as one trial type and B (i.e. the difference) as another and then use the contrasts above. It is possible to work out the F contrast for your design (but I do not know how to - JB does though!). > If I expect a "saturation" effect in the haemodynamic response, that > doesn't mean the quadratic betas will be negative, does it? How would a > "saturation" effect be reflected in the betas of the linear and quadratic > regressors? Actually negative parameter estimates for the quadratic term do usually imply saturation (if the 1st order coeficient is positive). As you imply though this depends, of course, on the basis functions specified. I hope this helps - Karl