[R] VGAM and vglm

Thomas Yee t.yee at auckland.ac.nz
Tue Oct 30 22:16:09 CET 2007


Hi Ted,

yes, vglm(..., family=binomialff(mv=TRUE)) does treat each response as
independent. You can see that with coef(fit, matrix=TRUE) where fit is
the object.  If you want to model two dependent binary responses then
you can try the VGAM family functions binom2.or(), loglinb2(),
binom2.rho().

 > So I'm wondering what is to be gained by using vglm instead of
 > simple glm?

Some VGAM family functions handle a multivariate response.  Whether
they are treated as independent or dependent depends on the specific
family function. Handling multivariate responses becomes important
when the modelling function is rrvglm(), cqo(), cao() etc.  Being able
to handle a multivariate response is a good thing, e.g., one can
easily plot several response curves simultaneously, as in the last
example in the vgam() help file.

 > I wonderif someone who is familiar with the details of vglm in the
 > VGAM package can assist me. I'm new to using it, and there doesn;t
 > seem much in the documentation that's relevant to the question
 > below.

Yes, the documentation
(http://www.stat.auckland.ac.nz/~yee/VGAM/download.shtml) is rather
patchy. Sorry! Things are developing there, albeit at a slow pace.

cheers

Thomas



More information about the R-help mailing list