If you want to do it that way (assuming group1=1, group2=2, etc), then the contrast would be

-0.5 0.0 0.5 0 0 0 0 0 0

This looks a little odd because it ignores group2, but that is the way the math works out. I would probably use one group and use 1, 2, 3 as a third covariate

doug

On 11/28/2012 12:53 PM, Mahinda Yogarajah wrote:

Hi,

I think (hope) i am finally getting to grips with how to analyse things but wanted to check one last thing.

Say I have 1 factor with 3 levels (3 groups of subjects) and 2 covariates (age and icv).

My design matrix would include:

Regressor 1 - ones for subject in Gp 1, 0 otherwise, codes intercept/mean for gp 1 Regressor 2 - ones for subject in Gp 2, 0 otherwise, codes intercept/mean for gp 2 Regressor 3 - ones for subject in Gp 3, 0 otherwise, codes intercept/mean for gp 3 Regressor 4 - age for subjects in Gp 1, 0 otherwise, codes age slope for gp 1 Regressor 5- age for subjects in Gp 2, 0 otherwise, codes age slope for gp 2 Regressor 6 - age for subjects in Gp 3, 0 otherwise, codes age slope for gp 3 Regressor 7 - icv for subjects in Gp 1, 0 otherwise, codes icv slope for gp 1 Regressor 8 - icv for subjects in Gp 2, 0 otherwise, codes icv slope for gp 2 Regressor 9 - icv for subjects in Gp 3, 0 otherwise, codes icv slope for gp 3

In terms of contrast would the following be correct:

1.5 0.5 -1.5 0 0 0 0 0 0 - to look for linear change in dependent variable across groups after correcting for age and icv

0 0 0 1.5 0.5 -1.5 0 0 0 - to look for linear change in age slope against dependent variable across groups while correcting for icv ie whether age-dependent measure correlations get stronger/weaker across gps

Thanks

Mahinda