Print

Print
Hi Jie Yu,

Please, see below:

On 3 December 2014 at 15:58, Jie Yu <[log in to unmask]> wrote:
Dear Anderson

Sorry for the bothering again.

I've decided to use my original design.mat, design.con, and your design2.grp.

Using randomise -C 2.3, I then used fslmeants to get the contrast estimates and draw graphs…


The previous design is fine, but it will make it more difficult to study the directions of the slopes for each group, something you'll want to look at if the interaction is significant. In any case, it's up to you to decide...

 

****************************************************************************

But then there is a problem with the age again…

We are doing a adolescent's study. Presumably, age matters a lot.

Can I create a new EV with the "time 1 age of all participants" and make it 0 in these contrasts?


The age is already modelled in your design with the subject-specific EVs. There is no need to model it separately. Likewise, there's no need to model, for instance, the height of the subjects, their weight, IQ, SES, nutritional status... all these are already taken care of by the EVs that are specific for each subject, and which are already in your model (both the first and the second).

So, although it may seem counter-intuitive, given that you actually need age in the model, it turns out that you don't need to model it explictly. It's already there. And if you try to include age, you'll receive an error message about the design being rank-deficient.

All the best,

Anderson