Is there any ubiquitous (or not so much) graphical method in count response models (e.g. Poisson GLM) to diagnose conditional zero-inflation? I'm aware of statistical tests that can be used for that, but I'm specifically interested in a diagnostic visualization. Some things I've thought of:

  • Looking at marginal response distribution might give you a hint, but it's certainly not completely reliable in terms of describing zero-inflation conditioned on predictors.
  • The residuals-vs-fitted: I'm not quite sure what exactly to look at over there to diagnose this particular aspect of zero-inflation
  • The residuals-vs-true values: over here at least I know that I should be focusing on the 0-true values and what kind of errors were committed on them, but what kind of pattern should I expect there? Vast number of strongly negative residuals (due to consistent over-estimation of zeros)?

Your Answer

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

Browse other questions tagged or ask your own question.