[R] Re: adjacent category model in ordinal regression

Claire Chabanet #3354 chabanet at dijon.inra.fr
Tue Nov 23 11:38:53 CET 2004


I suspect that vglm would do the job for the logit link.  Try :

     pneumo$let = log(pneumo$exposure.time)
     vglm(cbind(normal, mild, severe) ~ let, multinomial, pneumo)

     vglm(cbind(normal, mild, severe) ~ let, acat, pneumo)

This is an adjacent category model with logit link, although the
help page speaks about log link (log of odds = logit). 
But I think the model is, in fact, the same as the previous one.
(same deviance ...). It seems that only the parametrisation is different.

Hope this help. 

>   I want to analyze some multinomial data. And the
> response has a natural ordinal sturcture. I want to
> fit a adjacent category model to the data by logit,
> probit and complementary log-log link functions. I
> found a package "VGAM"
> (www.stat.auckland.ac.nz/~yee/VGAM/ ) whose function
> "acat" can fit the adjacent category model but it only
> has "log" and "identity" link functions. Does anybody
> know there is a package or function can do the
> analysis I want?

Claire Chabanet

More information about the R-help mailing list