 |
 |
Robin, If I've understood you correctly then you are after paired effects at the second level (done separately for each subject). You'll then group average across these subject paired effects at the 3rd level. One immediate problem that comes to mind is that this requires you to have different COPE outputs from a first level session GLM as inputs into the second level GLM (e.g. nov3 and repNov3 would both come from the same first level session analysis and would be different inputs into the same second level GLM). Strictly speaking this is a problem, because in FLAME we do not pass up the covariance structure between the COPEs from the first level (it is not straightforward to do so). However, I can not see any way round this with the experimental data that you have, so I would be inclined to take the head in the sand approach to this and carry on regardless. So the extra bit you are after I think is that you want to look for a trend (linear) in the paired differences at the second level?? If so then I think you can add an extra regressor alongside the one that normally models the pairwise effect mean, to model the linear trend. This should be the pairwise effect mean regressor modulated by a linear trend and orthoganalised wrt the pairwise effect mean regressor. Mean pairwise effects and linear trends in the pairwise effects can then be passed up to the third level for group averaging. have fun! Cheers, Mark. On Wed, 2 Nov 2005, Robin Goldman wrote: > thoughts anyone? > thanks. > R > > On Oct 31, 2005, at 6:10 PM, Robin Goldman wrote: > > > Hi there, > > > > I'm not sure how to set up a comparison in a 3 level design. Subjects > > underwent ten 5 minute > > scans, and in each, they were presented novel stimuli. These novel > > stimuli were repeated 2 scans > > later, so the novel items (nov) in scan 1 were presented again > > (repeatNov) in scan 3 (along with > > more nov items), the ones in scan 2 were repeated in scan 4, etc. > > > > So, for 15 subjects, I have 10 scans each, and want to do a contrast > > looking at repeatNov-Nov, so > > repNov3 vs nov1, repNov4 vs nov2, repNov5 vs nov3, ... > > > > I assume I set this up in some way as a paired t-test in a second > > level for each subject, then > > combine these in a 3rd level.... > > > > But things get trickier... I want to look not just at mean changes, > > but also at linear decreases > > across the scans and exponential decreases across the scans for this > > contrast. I know how to do > > this if the repeats were in the same scan and I could set up the > > contrast in the first level, but... > > > > um... > > > > Thanks! > > Robin > > > -- Mark Woolrich. Oxford University Centre for Functional MRI of the Brain (FMRIB), John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK. Tel: (+44)1865-222782 Homepage: http://www.fmrib.ox.ac.uk/~woolrich