Skip to content
Prev 767 / 7419 Next

glm for ratio [0,1] data

Hi,
Venables and Ripley, commenting on the use of glm with binomial family
(MASS book, page 190):
"If  the response is a numeric vector it is assumed to hold the data in
a ratio form, y[i] = s[i]/a[i], in which case tha a[i]s must be given as
a vector of weights using the weights argument".
So, if your ratio comes from e.g. estimating cover as s[i] cells occupied
from a total of a[i] cells in a sampling grid, you still can use the
binomial glm.

I recal that another possible aproch could be betaregression (see package
betareg).

Cheers,

Marcelino

Con fecha 31/8/2009, "Peter Solymos" <solymos at ualberta.ca> escribi?: