deriv example for nlmer

Dear Helen,

Won't it be easier to stick with Sslogis to run the model and then
convert scal to K for your report?

HTH,

Thierry

------------------------------------------------------------------------
----
ir. Thierry Onkelinx
Instituut voor natuur- en bosonderzoek
team Biometrie & Kwaliteitszorg
Gaverstraat 4
9500 Geraardsbergen
Belgium

Research Institute for Nature and Forest
team Biometrics & Quality Assurance
Gaverstraat 4
9500 Geraardsbergen
Belgium

tel. + 32 54/436 185
Thierry.Onkelinx at inbo.be
www.inbo.be

To call in the statistician after the experiment is done may be no more
than asking him to perform a post-mortem examination: he may be able to
say what the experiment died of.
~ Sir Ronald Aylmer Fisher

The plural of anecdote is not data.
~ Roger Brinner

The combination of some data and an aching desire for an answer does not
ensure that a reasonable answer can be extracted from a given body of
data.
~ John Tukey
  

-----Oorspronkelijk bericht-----
Van: r-sig-mixed-models-bounces at r-project.org 
[mailto:r-sig-mixed-models-bounces at r-project.org] Namens Helen Sofaer
Verzonden: dinsdag 19 oktober 2010 4:07
Aan: r-sig-mixed-models at r-project.org
Onderwerp: [R-sig-ME] deriv example for nlmer


Hello modelers,
I am working on a growth rate analysis in lme4, where my goal 
is to test for differences in growth rates between two 
populations of birds while incorporating random effects for 
nests (to deal with the lack of independence between 
siblings) and for each nestling (to deal with repeated 
measures of individuals). 

I'd like to build a model that is parameterized slightly 
differently than the SSlogis self-starting model, where K = 
1/scal (this is simply due to convention in the bird world). 
I can tweak the SSlogis code (given in the help file for 
selfStart) and the model runs, but cannot test for 
differences between populations. Since determining reasonable 
starting values is not a problem, and I'd like to be able to 
easily adjust the formula to incorporate additional 
covariates, I would like to avoid using a self-starting function.

It is my understanding that nlmer requires a function, but 
not necessarily a self-starting one. However, I can't find an 
example of any code that does this. 

I can build a very simple function that allows for 
differences between populations (referred to as site, which 
is a 0 1 dummy variable) in both the inflection point (xmid) 
and the growth rate (K). I'm trying to start with the 
simplest reasonable model, so I didn't allow for differences 
in the asymptote. This function works in nls:

logisKsite = function(Age, site, Asym, xmid, K, Kdiff, middiff){
	Asym/(1 + exp((xmid+middiff*site - Age)*(K+Kdiff*site)))
	}

startsite = c(Asym = 9, xmid = 3, K = .5, Kdiff=0, middiff=0) 
nls_logisKsite = nls(weight ~ logisKsite(Age, site, Asym, 
xmid, K, Kdiff, middiff), growth, start = startsite)

The nlmer model I am trying to run is (for simplicity, this 
model includes only a random effect of nest, only on the K parameter):
nlmer_logisKsite = nlmer(weight ~ logisKsite(Age, site, Asym, 
xmid, K, Kdiff, middiff) ~ (K | Nest_ID), growth, start = 
startsite, verbose = TRUE)

Running this model in nlmer produces the error: gradient 
attribute of evaluated model must be a numeric matrix Dr. 
Bates' posted lectures note that the model must provide 
derivatives, via the deriv function. I haven't been 
successful getting this to work, even in the model without 
site effects.

If there is an example of how to run nlmer without a 
self-start function, could someone please point me to it? 
Advise on how to implement the deriv function would also be 
very helpful. 

Thank you very much,

Helen

Helen Sofaer, PhD Candidate in Ecology, Colorado State University