[R] Question about glm using R

[Rubén Roa-Ureta]

Jordi Garcia wrote:
> Good morning,
>
> I am using R to try to model the proportion of burned area in 
> Portugal. The dependent variable is the proportion. The family used is 
> binomial and the epsilon would be binary.
>
> I am not able to find the package to be used when the proportion (%) 
> has to be used in glm. Could someone help me?
>
> I am using normal commands of glm.. for example:
>
> glm_5<- glm(formula=p~Precipitation, family=binomial(link=logit), 
> data=dados)
>
> where p is the proportion of burned area, but this error message 
> apperars:
>
> Warning message:
> non-integer #successes in a binomial glm! in: eval(expr, envir, enclos)
>
> That is why I think I am not using the proper glm package.
>
> Thank you very much in advance.
>
> Jordi
Jordi,
Your statistical model is wrong. The binomial family if four counts data 
(counts of successes given n trials), not for proportions. To model 
proportions, your family is the Beta family. I've modeled proportion 
response variables with function betareg of package betareg. If you want 
my example applications I can send you code and data off list.
Reference: Ferrari and Cribari-Neto. 2004. Beta regression for modelling 
rates and proportions. Journal of Applied Statistics 31:799-815.
Rubén