Marginal effects for hurdle negative binomial

[Selcuk Beduk]

Dear Statalisters,

I am running a hurdle negative binomial model on a counting deprivation index. I want to calculate the marginal effects (of say, log absolute income, relative income or/and social class) on the deprivation counts. I tried to use margins [margins, dydx(*) predict(equation(#2))] after the user-written command hnblogit but that does not work -I just got the same regression results.

What would you recommend me for an efficient way of calculating marginal effects (ideally for all the predictors) for a negative binomial hurdle model?

Many thanks in advance,

Selcuk

[Tom Weichle]

Hi Selcuk,

Have you considered using a Zero-Inflated Negative Binomial regression model? If you want to account for the prevalence of zero counts in the data, then the Stata command -zinb- will allow you to do so and the -margins- command can be used to calculate the marginal effects.

Tom

[Selcuk Beduk]

Hi Tom, many thanks for the suggestion. Indeed it could be an option but probably not for my data.

I preferred hurdle over zinb on a theoretical basis given that the zeros in my data (people with no deprivation) can theoretically always have a probability of a count (having some deprivation). Therefore, they are not 'structural' but 'variable' zeros. Also hurdle model conceptually makes sense for my data - crossing the hurdle of zero = crossing the deprivation threshold.

Still I did fit zinb and other count models and compared with hurdle (with vuong, aic and bic). Expectedly, hurdle fits best. Also, empirically the results of zinb and hurdle differ significantly. So, I use hurdle model.