Hi,

If the interaction isn't significant you can look at the group differences in that region.  If it is significant it doesn't make much sense since a significant interaction mean the group difference changes as a function of your continuous variable.

So you can either carefully interpret your interaction model, or if the interaction isn't significant use a single column and change the group column to all 1's.  The importance of that depends on how difference the between subject variance is for your two groups.

Cheers,
Jeanette

On Fri, Jan 27, 2012 at 3:57 AM, Torsten Ruest <[log in to unmask]> wrote:
Hi there,

I had setup 2 groups in the groups column, and when I added the single column covariate to the interaction model I'll have the problem of not separable EVs. Maybe an important piece of information is that the different acquisition parameters are only in the Pat group (indexed with 1, concerns 3 out of 20 in the Pat group, and out of 50 in total). So I am not sure whether to just use 1 group in the group column, or keep the 2 groups but split the nuisance variable into 2 columns. My 1st scenario would look like this:

EV1: Con
EV2: Pat
EV3: Acquisition
EV4: Age_Con
EV5: Age_Pat

C1: 0   0       0       1       -1
C2: 0   0       0       -1      1
C3: 0   0       0       1       0
C4: 0   0       0       0       1

and following Jeanette's advice, I'd be only interested in the interactions (C1 C2) and the individual group slopes (C3 and C4). Note that I am intending to run randomise and feat OLS.

Thanks for any suggestion!

Cheers,

Torsten