As that is not purely nested, lme cannot (easily) be applied.
On Thu, 20 Jan 2005, Paul Quataert wrote:
From: "Prof Brian Ripley" <ripley@stats.ox.ac.uk>
You need to specify the nesting, probably by Plant/Wplot/Splot.
Thank you for your answer. However Wplot and Splot are on an equal footing
and not nested within each other. I think I have a similar design (but
replicated) as mentioned in "Design and Analysis of Experiments" of Douglas
C. Montgomery (5th Edition). There on page 583 it is written:
"The strip-split-plot design has had extensive application in the
agricultural sciences, but if finds occasional use in industrial
experimentation. In the simplest case we have two factors A and B. Factor A
is applied to whole plots just as in the standard split-plot design. Then
factor B is applied to strips (which are really just another set of whole
plots) that are orthogonal tot the original whole plots used for factor A.
Figure 13-9 illustrates a situation in which both factors A and B have three
levels. Note that the levels of factor A are confounded with the whole
plots, and the levels of factor B are confounded witht the strips (which can
be thought of as a second set of whole plots).
B1 B2 B3
A1
A2
A3
Rephrasing my original question in this terminology the model with aov is:
Afit <- aov ( Gas ~ Treatment * FactorA * FactorB
+ Error (Block / (FactorA + FactorB) ), data = Wset)
Block because strip-split design is replicated and Treatment is an
additional factor assigned randomly to the blocks.
This works fine! Apparently Error() allows to put FactorA and FactorB on an
equal footing. However this is not possible any more with lme.
Specifically, following statement does not work:
Afit <- lme ( Gas ~ Treatment * FactorA * FactorB, data = Wset,
random = ~ 1 | Block / (FactorA + FactorB) )
In vain I tried to find another way, but any attempts results in nesting
FactorA and FactorB.
Thanks for any help,
Paul Quataert
Institute for Forestry and Game Management
Belgium
----- Original Message -----
From: "Prof Brian Ripley" <ripley@stats.ox.ac.uk>
To: "quataepa" <paul.quataert@lin.vlaanderen.be>
Cc: "Snews" <s-news@lists.biostat.wustl.edu>
Sent: Wednesday, January 19, 2005 2:12 PM
Subject: Re: [S] Strip-split-plot analysis with lme
On Wed, 19 Jan 2005, quataepa wrote:
Dear S-Plusers,
How to perform a strip-split-plot analysis in S+ (version 6.2) with lme?
Specifically I want to perform following analysis
Afit <- aov ( Gas ~ Treatment * Wplot * Splot +
Error( Plant / (Wplot + Splot) ) , data=Wset)
where Wplot and Splot are the whole plot and strip plot variables.
With aov no problem, however I do no succeed to translate this to lme.
I tried:
Afit <- lme ( Gas ~ Treatment * Wplot * Splot,
random = ~1 | Plant / (Wplot + Splot) , data=Wset)
but this is not allowed by the syntaxis.
You need to specify the nesting, probably by Plant/Wplot/Splot.
Brian D. Ripley, ripley@stats.ox.ac.uk
--
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 272866 (PA)
Oxford OX1 3TG, UK Fax: +44 1865 272595
|