[R] Vuong test

[Marc Schwartz]

On Jul 13, 2012, at 12:35 AM, sanchez ana wrote:

> Dear All,
> 
> I am using the function vuong from pscl package to compare 2 non nested models NB1
> (negative binomial I ) and Zero-inflated model.
> 
> 
> NB1 <-  glm(, , family = quasipoisson), it is an
> object of class: "glm" "lm"
> zinb <-
> zeroinfl( dist = "negbin") is an object of class: "zeroinfl"
>  
> when applying vuong
> function I get the following:
> vuong(NB1, zinb)
> Error en predprob.glm(m1) : 
>   your object of class glm is unsupported by
> predprob.glmyour object of class lm is unsupported by predprob.glm
> 
> 
> Any help will be really appreciated.



Either you did not properly copy your code above or something is amiss in your understanding.

NB1 is NOT a negative binomial model, but is a quasipoisson model. The latter is not fit using maximum likelihood, hence the Vuong test, which is a likelihood ratio based test, fails.

If you want a negative binomial model, you should be using glm.nb() in V&R's MASS package:

require(MASS)
NB1 <- glm.nb(...)

Then you can use the Vuong test from pscl.

If you have not, you might want to read:

  vignette("countreg")

Regards,

Marc Schwartz