 |
 |
Hi Tim, Please see below: On 10 February 2016 at 05:24, Timothy W. <[log in to unmask]> wrote: > Dear FSL users, > > I'm hoping to receive some guidance in setting up a design matrix for the > following experimental design: > > Mixed ANOVA with 2 within-subjects factors with 2 levels (time: pre and > post, condition: experimental and control) and 1 between-subjects factor > with 2 levels (age: young and old) > There are 2 age groups, and each subject received 4 scans (Exp_pre, > Exp_post, Control_pre, Control_post). > Before going forward, it's necessary to decide whether to consider either 'subject' or one of the two within-subject factors (e.g., condition) as a random effect. This changes the design quite a lot. But see more below. > > What's the best way to set up a design matrix for detecting voxels > exhibiting a significant Age x Condition x Time interaction? > > > What I have tried: > > I understand that a simpler route would be to compute difference images > (via fslmaths) and then run a 2-way mixed ANOVA on the differences > (between-subjects: age group, within-subjects: condition). I've done this > (following previous threads such as: > https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=FSL;e543cdaa.1510); > however, I want to ensure that I am accounting for baseline differences > (within subjects for the 2 conditions). I've tried entering each subject's > pre-scans for both conditions as voxelwise nuisance covariates, which > didn't seem to change the results of the 2-way MANOVA. So, now I'm > wondering whether a more complex model is necessary--where the pre and post > scans for both conditions are entered, rather than just the difference > images. > From this it sounds you want each condition to be considered as a random effect. In that case, the design at the link below. Note this design doesn't give much flexibility to test other things, and the two contrasts test the three-way interaction of age x condition x time (the two age groups are represented by EVs 1 and 2. https://dl.dropboxusercontent.com/u/2785709/outbox/mailinglist/design_tim.ods All the best, Anderson > > FWIW, the scans are resting-state scans, and I'd be submitting > z-transformed correlation maps as inputs to the higher-level analysis > (carried out by flameo in the command line). > > > I'd appreciate any input! > > Tim >