inflate part in zero-inflated negative binomial

[Aysegul Dogan]

I have a question dealing with zero inflated negative binomial: How does the logistic/inflate part work in zinb? Does it take structural zeros (always zero group) as 1 for dependent variable and zeros from negative binomial as 0 for dependent variable ? ((but ı think ıt ıs ımpossıble because we do not know the groups of zeros (structural zero or zero from negative binomial )). Or does it take the dependent variable equals 1 and 0 for dependent variable(y) greater than 0 or equal to 0,respectively ( in other words whole data set)? (But ın thıs case when we do the logit for the same varıables , it gives different results according to the results of the inflate part. Do you have any knowledge or resource advice about this confusion?

Thanks for your time.
Yours sincerely.


Ayşegül Doğan

[George Ford]

There's a logit model with a 0/1 based on the DV being zero or non-zero. ZINB assumes the excess zeros are from a logit/probt and the non-zero from the NB model.

The results will depend on the what's in the inflate model since estimates from the logit model enter the NB model through the Likelihood function.

help zinb

and read the PDF file from the manual.

[Aysegul Dogan]

Thank you for your answer. It helped a lot. I thought the two known distributions(negbin and logit) were working, but actually, the likelihood process has been completely changed.
Thanks for your time.