R help - Feb 2004 - repeated measures nonlinear regression

Hi, 
I found this email on the R website.  I am trying to figure out how to
analyse a data set that I believe will need to be run through a procedure
involving repeated measures, regression and mixed models.

The data is of insect populations (dependent variable - either 0/1=binomial,
or as counts=poisson) in sites with different characteristics (multiple
independent variables which are both categorical and continuous).  These
populations were repeatedly sampled (an unequal number of times) at different
times during the year.

The standard stats packages are not able to do what I would like to do.  Any 
suggestions for some alternative methods such as those which are specially 
programmed for R and are suited to the type of data which I am analysing?  I 
understand I will need a flexible repeated measures application to be able to 
accept the unequal sampling.  I also will need a gnlm regression to fit the 
general data structure.  And also, I think, the mixed models capability in 
order to fit in the true nature of the data (see next paragraph).

Lastly, I have a question about independent variables.  For example, I have a
measure for host plant density at a site, which I am using as a
predictive/independent variable to compare with the insect population/response
variable.  However, the values for the host plant data at each site are 
produced by many samples taken in the site to derive a mean. The normal method 
in ecology is to then use this mean as 'the' host plant value for that
site in
your analysis.  I don't think this represents the full qualities of the
data,
and would like to be able to include the fact that this is really a mean +/- 
SD into the final statistical model, since this is what it really is, not just 
a single perfect value.  Any suggestions on how to do this?  I thought maybe a 
generalized linear mixed model (glmm) might allow this type of analysis to be 
done.  Perhaps it could be accomplished with some restructuring of the data? 
>From my reading about statistics I gather it is a general problem in 
regression.


Thank you for your time,
Wayne Hallstrom

Wayne Hallstrom
Department of Biological Sciences
CW 405, Biological Sciences Centre
University of Alberta
Edmonton, Alberta
Canada, T6G 2E9 
Telephone: (780) 492-1180
Fax: (780) 492-9234

Wayne Hallstrom <wayne.hallstrom at ualberta.ca> writes:
> I found this email on the R website.  I am trying to figure out how
> to analyse a data set that I believe will need to be run through a
> procedure involving repeated measures, regression and mixed models.
> 
> The data is of insect populations (dependent variable - either
> 0/1=binomial, or as counts=poisson) in sites with different
> characteristics (multiple independent variables which are both
> categorical and continuous).  These populations were repeatedly
> sampled (an unequal number of times) at different times during the
> year.
> 
> The standard stats packages are not able to do what I would like to
> do.  Any suggestions for some alternative methods such as those
> which are specially programmed for R and are suited to the type of
> data which I am analysing?  I understand I will need a flexible
> repeated measures application to be able to accept the unequal
> sampling.  I also will need a gnlm regression to fit the general
> data structure.  And also, I think, the mixed models capability in
> order to fit in the true nature of the data (see next paragraph).
It sounds like you should be fitting a generalized linear mixed
model.  Look for the functions GLMM in package lme4 or glmmPQL in
package MASS.