Hi All,
I am using the glm.nb function from the MASS package (current version)
to fit and compare GLMs with negative binomial error distributions. My
question is: what is the appropriate method to use in the anova function
to compare models? If only one fitted object like
The re-estimated version is the principled one -- the other is a
computational shortcut.
However, what you model might choose depends on why you are doing model
selection, and also if there is a subject-specific reason to consider the
terms in this order.
And if you look in MASS (the book)
2 matches
Mail list logo