Hi There,
I was hoping you could help me with some questions about graphing outputs from afni MVM models with the correct standard error bars. For example, I have a model that includes a significant genderxgroup interaction. I would like to create a histogram for each cluster for which this interaction is significant with % signal change on the y-axis and 4 groups on the x-axis (female MDDs, female NCL, male MDDs, male NCL).
My question is: how does one extract the adjusted means and correct standard errors for this comparison? In asking around, it appears that many people extract mean % signal change for significant clusters for each subject and then graph the crude group means using the raw data, and their error bars are standard errors of those means (in manuscripts, it's often unclear exactly what folks have used). Is there a way to do this that actually reflects the adjusted MVM model?
Does a contrast statement within my MVM (e.g. a t-test for MDD vs Control within females) give me the adjusted means and standard errors consistent with the interaction term groupxgender (i.e. would those means/SEs be similar to what comes out of a least squares mean comparison from a GLM in R or SAS and reflect the underlying interaction model)?
That gets me to my last question, which is: is it correct to extract the %signal change for each significant cluster for each subject, bring that into SAS, and then run GLMs (with the same model specification as in the AFNI MVM model) and extract least squared means and SEs from those models for the purposes of graphing? While i see this as an improvement over graphing the crude means, I'm concerned that this might lead to smaller standard errors because the original AFNI models were run at the voxel level, and the SAS model would be at the cluster level.
Thanks so much for any insights or links to tutorials about this.
Best,
Kaja