Print

Print
See below.
 
Best

Helmut
 
Hi Helmut,
 
Thanks for your help! Much appreciated. You actually brought up some very interesting questions that I've been considering in theory based on past research, but haven't pinned down as a model. Just to clear up a few things, please see my comments inline:

Well, there are different options. E.g. set up one contrast [1 -1 ...] on single-subject level to test for "first five" > "second five" and a second contrast [0.5 0.5 ...] corresponding to "average" = (first five + second five)/2.
I'm wondering what it would mean to look at 'average'? Is the hypothesis here that first five = last five? (ie that activation doesnt actually change from early to late trials). 
It tests whether the arithmetic mean of the two differs from zero. It doesn't provide any further information on whether a sig. finding is due to the first five, the second five, or a combination of the two "conditions", which is why we also rely on the other contrast to test for differences between conditions.
 
With two conditions A and B, you can set up two contrasts (A + B)/2 = the arithmetic mean (or just A + B = the sum) and another contrast A - B = the difference, which then allows to test for the relevant effects. Other simple effects (A, -A, B, -B) can be determined based on these two contrasts, e.g. A is the arithmetic mean plus 0.5 * the differential contrast = (A + B)/2 + (A - B)/2. In any case, with regard to your two groups the two contrasts allow to assess main effects group, condition, and the interaction group x condition.
 
It's not uncommon to go with other second-level models though, e.g. often the differential contrast A - B is reported separately for the two groups in addition to / as a replacement for the differential contrast across groups.
 
To test for differences between first and second five trials on second level set up a two sample t-test based on the first con images (first bunch of files selected = group 1, second selection = group 2).
Do you mean group 1 would be my young subjects and group 2 my old, right?
Yes.
 
Within this second level model, [0.5 0.5] corresponds to "first five" > "second five" across groups, [-0.5 -0.5] corresponds to "first five" < "second five" across groups,
Got it!
[1 -1] corresponds to "(group 1 first five - group 1 second five) > (group 2 first five - group 2 second five)", [-1 1] corresponds to "(group 1 first five - group 1 second five) < (group 2 first five - group 2 second five)"
That's actually an interesting hypothesis! Basically we'd be asking whether the difference in activation between early and late stages of learning change as a factor of age.. right? That's really great!
Well, this is central to the experiment, otherwise there's no need to manipulate the two factors... not sure what you were planning ;-)
 
To test for group differences across conditions set up a two-sample t-test based on the second con images, the corresponding contrast vector within the second-level model is [1 -1] for group 1 > group 2 and [-1 1] for group 2 > group 1.
I guess this model would be asking how much difference there is between groups across the whole task period, sort of, yes (whether it really targets the "whole task period" depends on how you define period) and doesn't really account for any changes from early to late. It's based on the con images of the second contrast on single-subject level, which result from a certain combination of the beta estimates of condition A (first five) and condition B (second five) = [0.5 0.5 ...]. Thus I wouldn't use the expression "account for". In fact, would this be too different from just having, on the first level, 1 condition with 10 onsets (ie not dividing into early and late conditions), and forwarding this into a simple two sample t-test. with ie [1 -1] for group 1 > group 2? In general I would expect these to be similar, yes.
 
Many thanks for your help!!!
Joelle