R help - May 2009 - Hierarchical glm with binomial family

Dear members of the R help list,

I want to do a hierarchical glm with binomial family but am unsure
about how to write the syntax which involves nesting.

I want to test whether the risk of being attacked by Herbivores for
Meadowsweet plants is significantly dependent on the Distance to
heterospecific source plants.

Dependent variable = Herbivory (yes/no)
Explanatory continuous variable = Distance to heterospecific source plant

Distance should be nested within Subpopulation which in turn should be
nested within Population.
The number of replicates per subpopulation varies between 8 and 36.
The number of subpopulations per population varies between 4 and 9.

I haven't figured out how to do nesting, but guessing that nesting is
denoted with brackets I guess the syntax should look something like
this (below). Could you please help me to correct this syntax so that
it becomes useful in R?

model<-glm(Herbivory~Distance(Subpopulation(Population)), family=binomial)

Thanks a lot! I deeply appreciate your non-profit engagement!

Johan

Johan Stenberg-2 wrote:
> 
> Dear members of the R help list,
> 
> I want to do a hierarchical glm with binomial family but am unsure
> about how to write the syntax which involves nesting.
> 
> I want to test whether the risk of being attacked by Herbivores for
> Meadowsweet plants is significantly dependent on the Distance to
> heterospecific source plants.
> 
> Dependent variable = Herbivory (yes/no)
> Explanatory continuous variable = Distance to heterospecific source plant
> 
> Distance should be nested within Subpopulation which in turn should be
> nested within Population.
> The number of replicates per subpopulation varies between 8 and 36.
> The number of subpopulations per population varies between 4 and 9.
> 
> I haven't figured out how to do nesting, but guessing that nesting is
> denoted with brackets I guess the syntax should look something like
> this (below). Could you please help me to correct this syntax so that
> it becomes useful in R?
> 
> model<-glm(Herbivory~Distance(Subpopulation(Population)),
family=binomial)
> 
> 
You probably need a GLMM (generalized linear mixed model), which is
a little bit of a can of worms.  If so, you will need the "glmer"
function
inside the "lmer" package.

  I'm not entirely clear about your experimental design: I understand
that subpopulations are nested within populations, but it's not clear
whether covariates (distances to heterospecific plants) differ within
subpopulations or populations.

  If they don't differ with subpopulations, I would (strongly) recommend
aggregating the
values within subpopulations and analyzing proportions as a regression
analysis:
see Murtaugh, Paul A. ?SIMPLICITY AND COMPLEXITY IN ECOLOGICAL
 DATA ANALYSIS.? Ecology 88, no. 1 (2007): 56-62.

  If they do, then your design is

model<-glmer(Herbivory~Distance+(1|Population/Subpopulation),
family=binomial)

  See also:

https://stat.ethz.ch/pipermail/r-sig-mixed-models/2009q2/002320.html
https://stat.ethz.ch/pipermail/r-sig-mixed-models/2009q2/002335.html
-- 
View this message in context:
http://www.nabble.com/Hierarchical-glm-with-binomial-family-tp23742335p23743418.html
Sent from the R help mailing list archive at Nabble.com.