Skip to content
Prev 2862 / 7420 Next

R-sig-ecology Digest, Vol 48, Issue 8

You need to dig into the zeroinfl code and figure out how to set the 
parameter k to a fixed value. In fact...the code uses log(k)....where k 
is the parameter in the variance term of the NB. Set the k of the 
submodels to the k of the full model. Then calculate the likelihood 
values of the submodels and apply the likelihood ratio test.

The drop1 function for a NB GLM does this all automatically...but for 
zeroinfl you have to do it manually.
A detailed example plus all R code is given in: Zero Inflated Models and 
Generalized Linear Mixed Models with R. Zuur, Saveliev, Ieno. (2012). It 
comes out in 2 weeks and can be ordered from 
http://www.highstat.com/book4.htm


Alain