I am estimating a hurdle model, which consists of two data-generating processes: one for the zero/positive relationship, and a different one for the positive numbers above zero. A binary logit model is often used for the first part, and a zero-truncated poisson or negative binomial model for the second.

In R, the hurdle() command in the pscl package jointly estimates the models together, and returns a single AIC and log-likelihood value. I need to incorporate some random effects, and therefore am using the glmmADMB() command in the package of the same name. This package does not have a single hurdle model command, so I must estimate them separately (first binary, then poisson).

My question: how should one calculate the joint log-likelihood of two separately estimated models? I assume this is possible, given that hurdle seems to do just this.


So it turns out the solution is trivial, and is simply the sum of the two sub-models.

| cite | improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.