It depends on what your model/question is. Presumably what you're interested in is:
(1) What is the main effect of training (training minus baseline)?
(2) Is there an age effect? I.e., does the difference between training and baseline vary with age?
So, abstractly, (for a single voxel) you'd plot age on a horizontal axis, and the difference of training and baseline on a vertical axis. The effect of age would be the slope.
I'm not sure how to do this "conveniently" (*), except in the case where you can make a single con image representing
training minus baseline
If you can't do that, you can use ImCalc to subtract the baseline con image from the training con image.
In either case, you would then have one image per subject. At the group level, you'd do a simple correlation or regression. There'd be two columns in the design matrix: one is subject age, the other is a constant.
To see the effect of age on the change from baseline and training, you'd have a [1 0] or [0 1] contrast, the 1 being for the column for age.
To see the average change from baseline to training is a little more complicated. I think the correct expression is (assuming the order of columns in the design matrix is age, then constant (all one's))
[avg_age 1]
where avg_age is the average age of your sample. Or, you could do a paired t-test without age as a previous commenter stated.
---------------------
(*) For the case of a covariate with different values between pre and post scans (unlike your age), there's a way to do it and avoid having to use ImCalc, described in this post:
https://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=SPM;bd57195c.00
But I'm not sure is there's a similar trick when the covariate is the same for both scans.
Best
|