Hi all,
I am using S-Plus 2000, and trying to fit an lme model, where:
weight = body mass of study animal
id = code for individual animal
age = age of individual animal
year = (factor) year in which measurement was taken
logEPG = an index of parasite infestation.
Individual animals where measured repeatedly over different years, and change
therefore in weight, logEPG and obviously age from one year to an other (one
measurement of each variable for each individual in each year.
I use an lme because I would like to take account of the fact that repeated
measurements on the same animals where taken, and therefore I set random = ~ 1
| id. To account also for year effects (due for environmental variability from
one year to the next), I included year in the model as a fixed effect.
My fixed effects part of the model, therefore looks like this: weight ~ age +
age^2 + logEpg + year + year:age + year:logEpg.
This works and gives me plausible results, I am however not sure if I should not
rather consider year nested in id in the random part of the model (eg. random =
~ 1 | id/year.
A complication is that the design, is unbalanced because the data come from
field observations, and therefore not each individual is rappresented in each
year and also the age range for each individual can be different. the factors
id and year therefore overlap and are not really nested one into the other.
I hope I was clear enough in explaining my problem...can anybody help me in
clearing my doubts?
thanks
Achaz von Hardenberg
|