This is actually nothing to do with lme, and is affected by the
settings of options(contrasts=). lm fits work in the same way.
The default treatment contrasts are Helmert contrasts: see ?contr.helmert.
This is covered in Chapter 2 of the `Guide to Statistics' on-line manual,
at least in S+6.1
On Mon, 18 Nov 2002 Andre.Mery@aventis.com wrote:
>
> Dear all specialist users of S+,
>
>
> I had to use the following code :
>
> > W.lme <- lme(Value ~ Treatment*ordered(TimeNumber), random = ~1 | Rat,
> data = W)
>
> to analyse the so-called "Value", with "TimeNumber" refering to a repeated
> measurement design (2 different days of evaluation), and "Treatment" being a
> factor variable with 3 levels, say X (coded 1), Y (coded 2) and Z (coded 3).
> "Rat" is the subject (different set of subjects in the three treatment
> groups).
>
> The output gave me the following results :
>
> Value Std.Error DF t-value p-value
>
> Treatment1 -3.59722 1.128908 15 -3.18646 0.0061
> Treatment2 -0.78241 0.651775 15 -1.20042 0.2486
>
> So I was left with the interpretation of what "Treatment1" and "Treatment2"
> means ?
>
> I investigated the different outputs produced by "lme" in such a case, by
> changing the coding of the treatment levels. For example, I set X = 2, Y =
> 3, Z = 1, and so on for all the 6 possibilites. From what I obtained, I came
> to the interpretation that :
>
> "Treament1" refers to the comparison of the level coded 2 with the level
> coded 1
> "Treatment2" refers to the comparison of the level coded 3 with the other
> levels 1 and 2 as a whole.
>
>
> Can anybody tell me if this is the right interpretation ?
Part of it.
--
Brian D. Ripley, ripley@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 272860 (secr)
Oxford OX1 3TG, UK Fax: +44 1865 272595
|