[R] Re: Repeated Measures, groupedData and lme

[Ignacio Colonna]

Emma,
	I am not an expert, but I have been trying to fit similar models.
Adding to Keith's reply to your question, I can suggest what I concluded was
the most reasonable model for my case. Based on Keith's Model1, you might
also want to allow for a correlation among years within each experimental
unit (I am assuming the experiment was conducted at the exact same location
over the 3 years). 
	Say you want to impose an autoregressive, order 1 structure (you can
change this to any other structure you may consider appropriate) 
	To do this at the block*treatment unit (the smallest experimental
unit size in your experiment), you can add to keith's code:

correlation=corAR1(form=~1|block/treatment)

thus the entire code would be
Model1<-lme(mg~treatment + year + treatment:year, random=~1|block,
correlation=corAR1(form=~1|block/treatment),data=magnesium)

This results in a model with a certain covariance among all exp.units within
the same block, plus another covariance between pairs of years within the
same exp.unit, and this covariance decreases as the difference in time
increases.

You can see graphically the structure of this covariance by doing:
rho<-0.3
ar1<-corAR1(value=rho,form=~1|block/treatment)
ar1<-Initialize(ar1,data=yourdata)
m1<-corMatrix(ar1)
plot(m1$"1/name of first treatment"[,1])

Now, I really hope someone more knowledgeable is checking this out there and
will point out whether this is incorrect, as I have used it for my analysis
assuming was the correct approach. 

Ignacio


-----Original Message-----
From: r-help-bounces at stat.math.ethz.ch
[mailto:r-help-bounces at stat.math.ethz.ch] On Behalf Of Keith Wong
Sent: Friday, March 18, 2005 5:41 PM
To: r-help at stat.math.ethz.ch
Subject: [R] Re: Repeated Measures, groupedData and lme

Hello,

I'm an R-newbie, but I've been learning to use lme for repeated measures
experiments as well.

If I understand correctly: 
  Outcome variable: Mg (Kg/ha)
  Subject/grouping variable: block

  Condition/treatment: treatment (19 levels)
  Repeated factor: time (3 levels: 99, 02, 04)


I think if you specify the model formula in the lme call, then the formula
structure specified in the groupedData object is ignored.

One suggestion for the model:

Model1<-lme(mg~treatment + year + treatment:year, random=~1|block,
data=magnesium)

If the question of interest is the treatment:year interaction

Or
Model2 <- lme(mg~treatment, random=~1|block, data=magnesium)


Hope this helps ... and hope the experts chime in at this point to give more
guidance.

Keith


------quoting original post---
Hello

I am trying to fit a REML to some soil mineral data which has been
collected over the time period 1999 - 2004. I want to know if the 19
different treatments imposed, differ in terms of their soil mineral
content. A tree model of the data has shown differences between the
treatments can be attributed to the Magnesium, Potassium and organic
matter content of the soil, with Magnesium being the primary separating
variable.

I am looking at soil mineral data were collected : 99, 02, 04. 

In the experiment, there are 19 different treatments (treatmentcontrol,
treatment6TFYM, treatment 12TFYM etc),  which are replicated in 3
blocks.

For the magnesium soil data, I have created the following groupedData
object: 

magnesium<-groupedData(Mg~year|treatment, inner=~block) 
Where mg=magnesium Kg/ha

As it is a repeated measures I was going to use an lme.  I have looked
at Pinherio and Bates : Mixed-Effects models in S and S-plus and I am
getting slightly confused.  In order to fit the lme, should I specify
the data to use in the model as the grouped structure model?

If so is the following command correct:

Model1<-lme(mg~treatment, random=block|year, data=magnesium)? 

I am slightly worried that it isn't, because in model summary, instead
of listing the 19 different treatments in the fixed effects section, it
writes intercept (as normal), then treatment^1, treatment^2 etc.

However if I don't specify the groupedData object in the model, then in
the fixed effects section, it names the treatments (i.e. intercept,
treatmentcontrol, treatment6TFYM.

Should I be fitting the model using the whole data set rather than the
groupedData object?


Thank you very much for your help


Emma Pilgrim

______________________________________________
R-help at stat.math.ethz.ch mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide!
http://www.R-project.org/posting-guide.html