[R-sig-eco] Good writting of an unballanced lme model ?

[Claire]

Hello R-helper,

 

Can you help me with specifying a right model for my analysis? I have 
difficulty understand the nuances of the numerous possibilities from the 
on-line helps and I couldn’t get the Pinheiro and Bates (2000).

I took different measures on birds in 4 sites with 2 different forest 
treatments (2 sites per treatments). Thus I have a nested design where site is 
nested in treatment.

Additionally I have an unbalanced design (different number of birds capture pet 
sites and per treatments), and I have heteroscedasticity in my responses 
variable between my treatments. Moreover my sample size is not huge: 35 
individual in total.

 

I suppose that my model should be something like:

mod<-lme(measure1~treatment, random=~1|site, data=x, weights = 
varIdent(form=~1|treatment), method=”REML”)

Anova (mod, type=”III”)

 

I use the nlme package in R version 2.13.1. My mains questions are:

1) I have to put “sites” in random and take in count that it is nested in 
treatments. So should I write it: random = ~1|site   or   random = 
~1|treatment/site ?

2) The use of the function weights=varIdent allows me to specify that I have a 
case of heteroscedasticity, but I don’t get if I have to write it: 
varIdent(form = ~1|treatment)   or   varIdent(form = ~1|site)   or   
varIdent(form = ~site|treatment)   ?

3) I didn’t get which circumstances correspond to the best method function. 
Most of people said that method = “REML” is the best because unbiased but 
in all applied examples that I found used method = “ML”. Thus which one 
should I take?

 

Just in case this is an example of my data:

treatment           site         measure1          
  age

T              M20       13792    old

T              L5            14192    old

C             PERK      15424    old

C             G5          28112    old

C             G5          36592    young

T              L5            42192    young

T              M20       65856    young

C             G5          66064    old

T              L5            66288    young

C             PERK      73664    old

 

Thank a lot for your help.


        [[alternative HTML version deleted]]


_______________________________________________
R-sig-ecology mailing list
R-sig-ecology@r-project.org
https://stat.ethz.ch/mailman/listinfo/r-sig-ecology