[R] repeated values, nlme, correlation structures

[Patrick Giraudoux]

Dear listers,

As an exercise, I am trying to fit a logistic model with nlme.  Blue tit 
pulli (youngs) were weighted occasionnally (for field reasons) along 
time in 17 nestboxes. Individuals where not idenfied but their age was 
known. This means that for a given age several measurements were done 
but individuals could not be identified from a time to the other. This 
makes repeated values for a given age group in each nestbox. The aim is 
to get an acceptable growth curve (weight against age).

As far as repeated values cannot be handled with standard coStruct 
classes of nlme, I have done a first fit with nlme using the mean of 
each group. Comparing several models, the best fit is:

modm0c<-nlme(pds~Asym/(1+exp((xmid-age)/scal)),
    fixed=list(Asym~1,xmid~1,scal~1),
    random=Asym+xmid~1|nichoir,data=croispulm,
    start=list(fixed=c(10,5,2.2)),
    method="ML",
    corr=corCAR1()
    )

with pds = weight, age = mean age of each age group, nichoir = nestbox 
(a factor of 17 levels)

Based on the empirical autocorrelation function of the normalised 
residuals drawn from this model one can acceptaly assume that  the 
normalized residuals behave like uncorrelated noise.

Though this could be quite satisfying at first sight,  I am quite 
frustrated with starting from the mean weight of each age group, thus 
not being capable to manage and incorporate the variability around the 
mean weight of each age group in the model. My bible is Pinheiro & Bates 
(2000), but I did not find an example to board this problem.

Is there an affordable way (=not that much complicated for a biologist 
familiar to some general statistics) to handle this in nlme?

Any hint?

Patrick