Hi Leighton

Typically, a repeated measures or within-subject design (e.g. comparing differences in connectivity across two sessions for each subject) will already be intrinsically controlling for main subject effects (as both sessions for each subject can be expected to be affected similarly by those subject-effects, e.g. age). When entering age as a control factor in those analyses what you are controlling for is not for potential differences in connectivity with age (i.e. main age effects; does connectivity vary with a subject's age), but rather for potential differences in connectivity-change between the two sessions with age (i.e. age-by-session interactions; do changes in connectivity between the two sessions vary with a subject's age?). While that is a bit less common control to perform, it is also perfectly fine. In your case, what I think may perhaps be causing a big difference in the results when you control for age is that you are controllling for age at the level age=0, which may not be appropriate if your age covariate is not centered (e.g. if age is encoded in years you are asking the question of what would the expected change in connectivity be if extrapolated linearly to a 0-years-old subject, which may not be terribly meaningful if your sample is far from that). If that is the case, then I would suggest to either: a) center the 'age' covariate (subtract its mean value) before entering it in your second-level analysis; or b) change your between-subjects contrast to [1 ###] entering in ### the average age in your sample instead of 0 (both (a) and (b) should produce exactly the same results).

Hope this helps

Alfonso

Originally posted by Leighton Barnden:

Hi CONN folk,

Some advice please on a nuisance covariate (age) in a paired T-test.

For a cohort of subjects we acquired two task fMRI runs, each run acquired on the same day (different subjects on different days). In CONN we have two sessions, one each for run1 and run2. In RESULTS (2nd level) the default 'Conditions' contrast of [1 -1] gives us paired t-rest results. We have used both seed-based and group ICA designs.

I'm concerned about discounting the effect of age. Clearly, for each subject the age associated with each run is the same. Should I account for different ages across the cohort?

I have loaded Age via SETUP: Covariates (2nd-level) and it appears in RESULTS Subject Effects beneath 'AllSubjects'.

We have selected 'AllSubjects' and ‘Age’ and set the contrast to [1 0], (and left Conditions contrast [1 -1]).

Is this correct? It makes quite a difference to not discounting for age!

Leighton