Dear Rosalia,
If you have an unbalanced set of subjects (e.g. 12 males and 7 females) then you must demean or otherwise this regressor will have a non-zero mean and any non-zero mean within the dataset will drive this difference. An alternative, as I said in my previous email, is to include a regressor that is all ones (a simple mean) which would have zero associated with it in the contrasts. This is an equivalent way of dealing with the problem. But the situation you describe is no different from any other, and you must always either (a) demean both data and regressors, or (b) include a single mean regressor (all ones) and have zero associated with this in the contrasts.
All the best,
Mark
On 28 Aug 2012, at 03:07, Rosalia Dacosta <[log in to unmask]> wrote:
> Dear Mark,
>
> For example, if your regressors are gender (male 1, female -1)...then...I think there is no sense in demeaning them, is not there? All other regressors that refers to scoring in some test or scale should be demeaned...this is what I mean. For my design matrix, of course my regressors were all demeaned, but in the case you have "yes/not" variables which refer to a condition (gender, smoker, diabetic....where is the sense of demeaning?, is not the same?
>
> Kind regards,
>
> Rosalia.
>
>
|