I'm not sure I understand the question.
So I will answer a different but possibly related one !
If regressors A and B are correlated then you can test for what they
explain together using a [1 0; 0 1] F-contrast and what each explains
*over and above what is explained by the other* using a t-contrast [1 0]
and [0 1].
The issue of correlated variables is perhaps best dealt with by thinking
about statistical tests rather than parameter estimates.
cyril pernet wrote:
> Dear Will,
> Sorry to bother you directly but I didn't get any answers from a
> previous post and indeed you may be one of the few people how can help
> me ...
> I was wondering how the GLM is performed when fitting dummy and
> continuous variables - say by hand without convolution, something easy.
> The regression coefficients of continuous variables are unchanged
> whether I fit a whole model that include dummy (1-1) variables and
> continuous variables or only those continuous variables - by contrast
> the coef of the dummy variables are different between the full model and
> a model with only dummy variables - is there any particular way to do
> the fit?
> thank you for your help
William D. Penny
Wellcome Trust Centre for Neuroimaging
University College London
12 Queen Square
London WC1N 3BG
Tel: 020 7833 7475
FAX: 020 7813 1420
Email: [log in to unmask]