[R-sig-ME] GLM in an observatory study.

[Benjamin Cinget]

Good Morning everyone,

I am interested in whether plant pathogens species show any distribution across various agricultural factors and, in particular, whether there are any interactions among these factors impacting the pathogen abundances. I am working with different farmers, and consequently I did not choose the experimental design. Consequently, I have a very unbalanced structure in my data because the farms have different number of fields and, of course, different agricultural practices (Survey study).

I considered that the response variable is the species abundances (�Abun�) and three explanatory variables, all categorical: �Fun� is a three-level factor for fungicide treatments of the farm (one treatment by farm), �Cult� is a nine-level factor for cultivar (only one planted in one field), �Stime� is a three-level factor for three sampling times for a same field,  and there are twelve pathogen species observed.

Basically, because I learned like that:

First - I built the following full model with a Poisson correction error (because I am working with abundances):

model. -> glmer(Abun ~ Fun * Cult * Stime * Spe + (Stime|Farm/Field), family = Poisson)

where �(Stime|Farm/Field)� term is to consider the repeated measurements of nested fields in farms as random effects.

Second -  I performed by deviance analysis, step by step, by suppressing the interactions involving the �Spe� variable to test their significant, and by beginning by the highest interaction (Fun:Cult:Stime:Spe, three-way interaction).

My questions are :

  1.  Because my glm classes are far away (2005), is that still a good approach? I do not want to build a perfect model,  but just to estimate if my three parameters (and interactions) are involved in the species distributions. If not have you some up to date suggestions, please ?



  1.  I am not sure for the �(Stime|Farm/Field)� because �Stime� is categorical, do I have to dissociate the two random effects ?

Thank you very much for your help,

Benjamin Cinget


	[[alternative HTML version deleted]]