Vuong test comparing zero-inflated negative log binomial and negative log binomial not working?

[Jill Korsiak]

Hello,
I am running a zero-inflated negative binomial model on Stata v.13.1 and have included the 'vuong' option to compare the zero-inflated model to a regular log binomial model. However, the output for the Vuong test is missing, I get the following output: Vuong test of zinb vs. standard negative binomial: z = . Pr>z = .
I'm not sure why this is happening, especially because when I include the vuong option to compare zero-inflated poisson and regular possion model, it seems to work fine.
Any insight would help, thanks!

[Joao Santos Silva]

Dear Jill,

It is a good thing that is not working because the test is invalid anyway; the same applies to the the test of the ZIP vs Poisson. Maybe someone form Stata will look into this?

Best wishes,

Joao

[Jill Korsiak]

Hello Joao,
Thank you for such a quick response. Can you elaborate on what you mean when you say the test isn't valid? Are you referring to the test itself, or how the test is internally written in Stata?
Did you have any suggestions on what else can be used to compare zero inflated negative binomial vs. normal binomial, or zero inflated poisson to regular poisson models?
Thanks so much,
Jill

[Joao Santos Silva]

Dear Jill,

Voung's test comes in different flavors and the version to use depends on the type of models being compared. The version that is easy to use is for non-nested models and that is what is implemented in Stata. However, the ZI models and they standard counterparts are overlapping models, not non-nested. For this case Vuong's test is very different (and the test statistic has a chi-square distribution, not the standard normal). Hopefully, someone from Stata will also see this and look into it. Meanwhile, you may want have a look at the -hpc- command and at this paper.

Best wishes,

Joao

[Jill Korsiak]

Great, thank you for all your help Joao, I will look into that paper.
Jill