I apologise in advance if this sounds like a stupid question. I am used to GLM with continuous data.
I've run a hurdle model in R (pscl:hurdle) with a negative binomial distribution for the "count" component.
With GLM models usually we report whether the model is significant, a measure of variance accounted for, how well it fits the data etc.
I'm struggling to understand what the equivalent statistics are for hurdle models.
I've managed to use Log likelihood tests (and Vuong) to establish that this model is better than ZINB, NB, ZIP, Poisson. But I'm struggling to figure out how I would test if the model actually accurately predicts the data (or to what degree it does).
Is there an equivalent of plotting predicted vs actual values for these models? And would determining if a model is significant be a simple case of performing an LL test against a version of the model with no predictors in it?
