[R-sig-ME] Split-plot Design

[John Maindonald]

I do not think it quite true that the aov model that has an
Error() term is a fixed effects model.  The use of the word
"stratum" implies that a mixed effects model is lurking
somewhere.  The F-tests surely assume such a model.

Some little time ago, Doug Bates invested me, along with
Peter Dalgaard, a member of the degrees of freedom police.
Problem is, I am unsure of the responsibilities, but maybe
they include commenting on a case such as this.

lme() makes a stab at an appropriate choice of degrees
of freedom, but does not always get it right, to the extent
that there is a right answer.  [lmer() has for the time being
given up on giving degrees of freedom and p-values for
fixed effects estimates.]  This part of the output from lme()
should, accordingly, be used with discretion.  In case of
doubt, check against a likelihood ratio test.  In a simple
enough experimental design, users who understand how
to calculate degrees of freedom will reason them out for
themselves.
John Maindonald.

John Maindonald             email: john.maindonald at anu.edu.au
phone : +61 2 (6125)3473    fax  : +61 2(6125)5549
Centre for Mathematics & Its Applications, Room 1194,
John Dedman Mathematical Sciences Building (Building 27)
Australian National University, Canberra ACT 0200.


On 21 Mar 2008, at 9:23 AM, Kevin Wright wrote:

> Your question is not very clear, but if you are trying to match the
> results in Kuehl, you need a fixed-effects model:
>
> dat <- read.table("expl14-1.txt", header=TRUE)
> dat$block <- factor(dat$block)
> dat$nitro <- factor(dat$nitro)
> dat$thatch <- factor(dat$thatch)
>
> colnames(dat) <- c("block","nitro","thatch","chlor")
> m1 <- aov(chlor~nitro*thatch+Error(block/nitro), data=dat)
> summary(m1)
>
> Mixed-effects models and degrees of freedom have been discussed many
> times on this list....search the archives.
>
> K Wright
>
>
> On Thu, Mar 20, 2008 at 12:39 PM,  <marcioestat at pop.com.br> wrote:
>>
>>
>>
>> Hi listers,
>>
>> I've been studying anova and at the book of Kuehl at the chapter
>> about split-plot there is a experiment with the results... I am  
>> trying to
>> understand the experiments and make the code in order to obtain the
>> results... But there is something that I didn't understand yet...
>> I have a split-plot design (2 blocks) with two facteurs, one
>> facteur has 4 treatments and the other facteur is a measure
>> taken in three years...
>> I organize my data set as:
>>
>> Nitro Bloc Year Measure
>> a
>> x
>> 1         3.8
>> a
>>  x
>> 2         3.9
>> a         x         3         2.0
>> a         y         1         3.7
>> a         y         2
>> 2.4
>> a         y         3
>> 1.2
>> b         x
>>  1         4.0
>> b         x
>> 2         2.5
>> and so on...
>>
>>
>> So, I am trying this code, because I want to test each factor and the
>> interaction...
>> lme=lme(measure ~ bloc + nitro + bloc*nitro, random= ~ 1|year,
>> data=lme)
>> summary(lme)
>> The results that I am obtaining are not correct, because
>> I calculated the degrees of fredom and they are not
>> correct... According to this design I will get two errors one for the
>> whole plot and other for the subplot....
>>
>> Well, as I told you, I am still learning... Any suggestions...
>>
>> Thanks in advance,
>>
>> Ribeiro
>>
>>
>>        [[alternative HTML version deleted]]
>>
>>
>> _______________________________________________
>> R-sig-mixed-models at r-project.org mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>>
>>
>
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models