Hello AFNI board,
I have ended up working with a, for me, non-trivial design and I need help with the group level analysis. Especially the ANOVA aspect.
The design (not mine) is like this:
We have 20 subjects which are conditioned by giving them, in two sessions, a drug in combination of viewing a certain picture and a placebo when viewing another picture. To balance this the background coupled with the drug is shifted between subjects. Then, sometime after the conditioning, an fMRI experiment is preformed where they are shown both Picture 1 and Picture 2 (one at the time). So we end up with:
10 Subjects who have
Pictue1_Drug and
Picture2_Placebo
And
10 Subjects who have
Picture1_Placebo and
Picture2_Drug
The main interest here is simply the contrast between viewing the Drug and Placebo picture and this can be done with e.g. 3dttest++. But we do need to control for the different motives of Picture 1 and Picture 2. So, we would like to see if there is an interaction between the Drug/PBO factor and the Group factor (i.e. just make sure that there were no overall "liking bias" for e.g. Picture 1).
First we thought about using 3dANOVA3. Since we basically have 3 factors:
Factor 1 = Drugs. Levels = RealDrug and Placebo
Factor 2 = Group. Levels = Group that got drug coupled with Picture1 and Group that got drug coupled with Picuture2.
Factor 3 = Subject. Levels =10 (Not 20, right?)
I guess this can be called a “Mixed effects ANOVA with one between subject factor (Group) and one within subject factor (Drug/PBO)”?
But, since we would like to see the main effect of factor A (Drug vs Placebo) and the interaction (AxB) we cannot use 3dANOVA3 since the option:
–fa
is not available for –type 3. Type three is for A = Fixed, B = Random and C = Random. Which is what we have since Drug vs PBO (within subject, fixed), Group1 vs Group2 (between groups, random) and subject (random).
We want to have a map where we see the main effect of the Drug/Placebo factor and also make sure that there, in these areas, are no interaction between it and the group factor (since that can mean our results are due to a “liking bias” towards certain picture).
I have by a college been suggested to use 3dMVM, and that seems to do what we want. Is this a good option? Then I would need to create the main effect of factor A (Drug –PBO), main effect of factor B (Group 1 –Group2) and then the interaction between factor A and B?
If this is a good option, would the between group variance (bsVars) be Group and within subject variance (wsVars) be Drug/PBo?
Many thanks!
Robin
Edited 10 time(s). Last edit at 04/19/2016 11:22AM by Robin.