See the following web page, which gives helpful examples on demeaning,
and answers your Q3 (A: demean across all subjects)
http://mumford.fmripower.org/mean_centering/
On Thu, 2011-05-19 at 21:31 +0100, R. Duke wrote:
> Hello Experts,
>
> I have tried my best to confirm my design with the postings, but I feel like I need to confirm with someone on the listserv. Here is my design.
>
> 2 groups (young and old)
> 2 nuisance covariates (cov1, cov2)
> 1 covariate of interest (cov_interest)
>
> I want to test whether there is a groupXcov_interest interaction (i.e., differential slopes between young and old in cov_interest).
>
> Young Old Cov_interest_young Cov_interest_old Cov1 Cov2
> 1 0 20 0 100 500
> 1 0 44 0 200 560
> 0 1 0 33 160 450
> 0 1 0 45 222 720
>
>
> Contrast1 = 0 0 1 -1 0 0
> Contrast2 = 0 0 -1 1 0 0
>
> Questions:
> 1. Does this design/constrast(s) appear correct?
> 2. Should I demean any of my covariates or should I just enter RAW values into each of my four columns?
> 3. If I should demean any of my covariates, should I demean within or between my two groups? This one seems to be highly debated.
>
>
> Cheers!
|