Using lme() for split plot

That's a good example with a couple levels of nesting (similar to the
examples in the other book), but they still only have one factor,
'Variety', nested in each block. ?Am I missing something? ?Should I
make up a psuedofactor with four levels to code my two two-level
factors?


On Thu, May 7, 2009 at 5:46 PM, Rub?n Roa-Ureta <rroa at udec.cl> wrote:

Joshua Stults wrote:

Hi,

I'm trying to figure out how to use lme() for analyzing a split-plot
experiment. ?I've been looking at the examples from the 'R Book',
those are nested but with only one factor at the whole-plot level, my
test is 2^2 at the whole-plot level, with a single many level factor
at the sub-plot level. ?My question is about properly specifying the
random effects part of the model,

lme( y ~ block + a*b*poly(c, n), random=~ ? )

Where 'a' and 'b' are my two level whole-plot factors and 'c' is the
many level sub-plot factor. ?I'm not sure what to use to get the right
error terms. ?Do I use two error terms:

random = ~ 1 | block/a + 1 | block/b

or one:

random = ~ 1 | block/a*b

or something else entirely? ?I haven't been able to find any relevant
examples on Google. Thanks for any suggestions/pointers.

Have you checked Pinheiro and Bates 2004 Mixed-effects models in S and
S-PLUS? They have a split-plot example starting on p. 45.
Rub?n

--
Joshua Stults
Website: http://j-stults.blogspot.com

Joshua Stults wrote:

Hi,

I'm trying to figure out how to use lme() for analyzing a split-plot
experiment. ?I've been looking at the examples from the 'R Book',
those are nested but with only one factor at the whole-plot level, my
test is 2^2 at the whole-plot level, with a single many level factor
at the sub-plot level. ?My question is about properly specifying the
random effects part of the model,

lme( y ~ block + a*b*poly(c, n), random=~ ? )

Where 'a' and 'b' are my two level whole-plot factors and 'c' is the
many level sub-plot factor. ?I'm not sure what to use to get the right
error terms. ?Do I use two error terms:

random = ~ 1 | block/a + 1 | block/b

or one:

random = ~ 1 | block/a*b

or something else entirely? ?I haven't been able to find any relevant
examples on Google. Thanks for any suggestions/pointers.

Have you checked Pinheiro and Bates 2004 Mixed-effects models in S and
S-PLUS? They have a split-plot example starting on p. 45.
Rub?n

Crossed Random effects are difficult using lme (wasn't designed for it). Try
lmer in the lme4 package if you need this.


Bert Gunter
Genentech Nonclinical Biostatistics


-----Original Message-----
From: r-help-bounces at r-project.org [mailto:r-help-bounces at r-project.org] On
Behalf Of Joshua Stults
Sent: Thursday, May 07, 2009 3:02 PM
To: Rub?n Roa-Ureta; r-help at r-project.org
Subject: Re: [R] Using lme() for split plot

That's a good example with a couple levels of nesting (similar to the
examples in the other book), but they still only have one factor,
'Variety', nested in each block. ?Am I missing something? ?Should I
make up a psuedofactor with four levels to code my two two-level
factors?


On Thu, May 7, 2009 at 5:46 PM, Rub?n Roa-Ureta <rroa at udec.cl> wrote:

Joshua Stults wrote:

Hi,

I'm trying to figure out how to use lme() for analyzing a split-plot
experiment. ?I've been looking at the examples from the 'R Book',
those are nested but with only one factor at the whole-plot level, my
test is 2^2 at the whole-plot level, with a single many level factor
at the sub-plot level. ?My question is about properly specifying the
random effects part of the model,

lme( y ~ block + a*b*poly(c, n), random=~ ? )

Where 'a' and 'b' are my two level whole-plot factors and 'c' is the
many level sub-plot factor. ?I'm not sure what to use to get the right
error terms. ?Do I use two error terms:

random = ~ 1 | block/a + 1 | block/b

or one:

random = ~ 1 | block/a*b

or something else entirely? ?I haven't been able to find any relevant
examples on Google. Thanks for any suggestions/pointers.

Have you checked Pinheiro and Bates 2004 Mixed-effects models in S and
S-PLUS? They have a split-plot example starting on p. 45.
Rub?n

--
Joshua Stults
Website: http://j-stults.blogspot.com