[R-sig-ME] Mixed additive model question

[Diaz, Eliecer R]

I am an ecologist, and this is my first time in this forum. I have data concerning plant cover (continuous variable ranging from 0-100%); I am trying to explain plant cover in relationship to the following covariables: habitat (factor: two habitats, pools and rocks:), Temperature (air temperature in the days when the plant cover was sampled among patches), patch (Factor: 12 fixed spots per habitat), and finally to 2 continuous covariables: i. elevation of the substratum and the density of grazers (average along time, so it did not change temporally) in each patch.


I found heterogeneity in the residuals versus the covariable Temperature so I try to reduce them using this following model:


M1 <-   gamm(PlantCover ~ factor(Habitat) + Elevation + Grazers + s(Temperature),

                                random =~ 1 | Patch,                                       #Patch is a factor

                                weights = varIdent(form=~ 1 | Habitat),            #Habitat is a factor

                                correlation = corAR1(form=~ 1 | Temperature) #Temp is numeric month temperature

                                method = "REML",

                                data = plants,

 family = "gaussian")


Alternatively, I have this model:


M2  <- gamm(Response ~  grazers + Elevation + s(Temperature) + Hab,

          random = list(Patch=~1),

          weights = varExp(form =~ grazers),

          method = "REML",

          data = Mac,

          correlation = corAR1(form=~ 1 | Temp),

          family = "gaussian")


These were my two best models after check fifteen versions:

Comparing only these two in terms of AIC output:

       AIC

M1: 6956

M2: 6693


Does someone see a error in the specification of the model??

M2, improves the residuals, although not totally, but this is the only I can manage to do by now. If someone has some suggestions you are welcomed!


Thanks,

Elis




	[[alternative HTML version deleted]]