 |
 |
Typo that could lead to confusion: "those4" should just read "those" Jim Porter TRiCAM Lab Coordinator Elliott Hall N437 612.624.3892 www.psych.umn.edu/research/tricam James N. Porter wrote: > Howdy FSLers- > > I have a question about proper design matrix procedures. > > First, say I've performed lower-level FEATs and obtained 8 copes (A-H) > for my 8 task conditions for each subjects' repeated scans. Next, I > run a higher-level analysis that obtains the subjects' means for the 8 > copes across their scans. Then I run a higher-level analysis to obtain > contrasts between these 8 copes. My inputs are cope images, of which > each subject has 8. Here is my EV matrix for this set up (for > simplicity, as if I only had 2 subjects): > > A B C D E F G H > cope1.feat/stats/cope1.nii 1 0 0 0 0 0 0 0 > cope1.feat/stats/cope2.nii 1 0 0 0 0 0 0 0 > cope2.feat/stats/cope1.nii 0 1 0 0 0 0 0 0 > cope2.feat/stats/cope2.nii 0 1 0 0 0 0 0 0 > cope3.feat/stats/cope1.nii 0 0 1 0 0 0 0 0 > cope3.feat/stats/cope2.nii 0 0 1 0 0 0 0 0 > cope4.feat/stats/cope1.nii 0 0 0 1 0 0 0 0 > cope4.feat/stats/cope2.nii 0 0 0 1 0 0 0 0 > cope5.feat/stats/cope1.nii 0 0 0 0 1 0 0 0 > cope5.feat/stats/cope2.nii 0 0 0 0 1 0 0 0 > cope6.feat/stats/cope1.nii 0 0 0 0 0 1 0 0 > cope6.feat/stats/cope2.nii 0 0 0 0 0 1 0 0 > cope7.feat/stats/cope1.nii 0 0 0 0 0 0 1 0 > cope7.feat/stats/cope2.nii 0 0 0 0 0 0 1 0 > cope8.feat/stats/cope1.nii 0 0 0 0 0 0 0 1 > cope8.feat/stats/cope2.nii 0 0 0 0 0 0 0 1 > > Now say I wanted to know the results of the contrasts > 1-- ((A-B)-(C-D)) - ((E-F)-(G-H)) > 2-- ((A-C)-(B-D)) - ((E-G)-(F-H)) > > Would it be mathematically/statistically correct to simply perform a > distribution of the signs/operators and end up with contrasts coded as: > A B C D E F G H > 1-- 1 -1 -1 1 -1 1 1 -1 > 2-- 1 1 -1 1 -1 1 1 -1 > > Under this method, I can imagine two conceptually distinct contrasts > that multiply out to have the same ±1 values, which I also imagine > would be a bad thing. If so, would my analysis path be required to > follow the old PEMDAS rules and find the values of each parenthetical > contrast in the prescribed order? In other words, is it best to > > 1) define the /elements/ of a higher-order contrast in the lower-level > analyses, obtain subject means on the elements in a higher-level > analysis, then contrast those results in one shot in a group means > higher-level analysis, or > > 2) set up all desired higher-order contrasts from the get-go at the > single-subject level, then simply obtain subject and group means for > those4 copes (performing no extra contrasting) with the higher-level > analyses? > > Thanks for all your help, > > -- > Jim Porter > TRiCAM Lab Coordinator > Elliott Hall N437 > 612.624.3892 > www.psych.umn.edu/research/tricam