[R-sig-ME] GLMM or GLM?

[Gabriel Baud-Bovy]

On 4/23/2013 10:14 PM, joana martelo wrote:
> Dear list
>
>   
>
> I'm trying to model the relationship between prey capture success by fish
> and fish size, density and velocity, but I’m not sure which statistical
> method to use – GLMM or GLM? Capture success was collected during trials
> that consisted in sending a prey every x min, which could be captured or not
> by individual fish. A total of 20 prey was send in each trial.
>
>   
Are density and velocity fish characteristics ? see below.
>
> Method A: GLMM
>
>   
>
> I used fish.id as a random effect and the behavior of fish (0 no capture, 1-
> capture) as the response variable. One of my models looks like this:
>
>   
>
> Model1<-glmer(capture~fish size + density
> velocity+(1|fish.id.),family=binomial,data=cap)
>
>   
>
>   
>
> Method B: GLM
>
>   
>
> The response variable was the proportion of prey captured by each fish and
> one of my models:
>
>   
>
> Model1<-glm(prop.capture~fish size + density
> velocity,family=binomial,data=cap)
>
>   

>
>   
>
> I used model selection using Akaike weights to examine the performance of
> each model.  Results were similar with both methods, but I think I lose a
> bit of biological information if I use A: 1) I can't model average which may
> mean a loss of information when I have many interpretable models as I have
> with both approaches, 2) I lose the "fish size" effect which is an important
> bit of biological information (i.e. larger fish have higher capture success
> rates). However, A might be more appropriate for my type of data, even
> though I’m not interested in variation within fish but among fish ….
>
1. Fitting ungrouped or grouped (proportions) data should give the same 
results if
done correctly. You should not loose any information.

For glm, a typical response is a two-column matrix with the columns 
giving the numbers of successes and failures. If
you use proportions, you should also use the weights argument to 
indicate the number of cases.

2. To me, the first model seems more correct if  the model includes 
within-subject/fish
factors (density or velocity ?)

3. If you have only between-subject variables, then the GLM model is 
correct
(see 10.3 in Zuur et al., 2009, Mixed Effects Models and Extensions in 
Ecology with R).

My 2 cents.

Gabriel

>   
>
> Can anyone help?
>
>   
>
> Thanks in advance!
>
>   
>
>   
>
> Joana Martelo, PhD Student
>
> Centro de Biologia Ambiental
>
> Departamento de Biologia Animal
>
> Faculdade de Ciências, Edificio C2,5ºPiso,Sala 2.5.15B
>
> 1749-016 Lisboa, Portugal
>
>   <http://ffishgul.fc.ul.pt/> http://ffishgul.fc.ul.pt
>
> Por favor pense na sua responsabilidade ambiental antes de imprimir este
> email
>
>   
>
>   
>
>   
>
>   
>
>
> 	[[alternative HTML version deleted]]
>


-- 
---------------------------------------------------------------------
Gabriel Baud-Bovy               tel.: (+39) 02 2643 4839 (office)
UHSR University                       (+39) 02 2643 3429 (laboratory)
via Olgettina, 58                     (+39) 02 2643 4891 (secretary)
20132 Milan, Italy               fax: (+39) 02 2643 4892