Negative Binomial and count data

[Laila]

Dear expert

(Im using Eviews 5, sorry im a bit out of date )

I am trying to estimate an equation using count data (number of violent incidents) and thought it would be best to use Negative Binomial QLM as the data is overdispersed and Poisson cant be used. The options are confusing me, what is the criteria for choosing the Optimization Alogrithim (defaul is quadratic hill climbing) ? Also, how do I specify the hurdle model (NB), cant find that and the zero inflated option? After I select the NB (QML), the default value for fixed variance parameter is 1, could you explain this?

Thank you very much for any help you can provide. Id be grateful

Laila

[EViews Glenn]

The optimization algorithm is just that, the method used to optimize the likelihood. There is no real criterion. The only issue is that the method does tie into the method used for computing coefficient covariances. The quadratic hill climbing uses the estimated Hessian.

The QML/NB is an estimate for a fixed value of the overdispersion parameter. If you want to estimate the full likelihood, you need to use the ML/NB.

Hurdle and ZIP models aren't supported natively.

[Laila]

OK, I got it, thanks a ton