[R] problem with se.contrast()

[Prof Brian Ripley]

On Tue, 22 Feb 2005, Christoph Buser wrote:

> Dear Prof Ripley, Dear Prof Dalgaard
>
> Thank you both for your help. I tried it with helmert contrasts
> and got a result that is consistent with lme. I didn't realize
> that the parameterization of the model has an influence on the
> contrasts that I tried to test.
> It seems that I should read a little bit more about this issue
> for my better understanding.
>
> I have one last point to propose:
>
> You could include (as interim solution) a warning (that there might
> be an in efficiency loss) in se.contrast() if one uses
> non-orthogonal contrasts and a multi-stratum model.

If you meant on the help page, I have already done that.
We are still thinking about how to detect the problem well enough, and 
possibly even to see if we can avoid it.

It's trickier than I remembered, although I think I did probably once 
know.

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595