3
$\begingroup$

For my overdispersed and zero-inflated data I have used a zero inflated negative binomial and a hurdle model, for comparison purposes.

I need to validate the model, i.e. see whether certain assumptions are fulfilled to check if the model is adequate for inference. However, I'm not able to find any information on the steps of validation of these models.

Side question: Is there a more elaborate way to compare hurdle and zero-inflated models, other than the AIC?

Does anyone have any suggestions?

$\endgroup$

1 Answer 1

1
$\begingroup$

For validating the ZINB or hurdle model, your first step could be to create a null model and compare the ZINB/ hurdle model and the null model using a likleihood ratio test. This will tell you whether your ZINB or hurdle model fits the data better than the null model, providing some validation of its utility. If so, you could proceed with methods such as cross validation and determining pseudo R2, though I don't have personal experience applying these methods to hurdle models.

$\endgroup$
2
  • $\begingroup$ That's a great suggestion! In "normal" regression models usually there is the F-test for that, what you suggested is a good alternative. Didn't occur to me. And what about visual analysis for validation? $\endgroup$
    – Bileobio
    Dec 2, 2022 at 12:20
  • $\begingroup$ @Bileobio Glad the suggestion was helpful. I haven't previously used visual analysis for validation of these model types, so I'm not positive. My go-to for most regressions is to look at marginal effect plots of individual predictors. If they look 'nice' and have reasonable confidence intervals, I feel good about the model $\endgroup$
    – SageR
    Dec 5, 2022 at 4:25

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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