0

I have been trying to explore the residuals associated with zero inflated models for I x J contingency table. I can figure out deviance residuals for ZINB model through zinbwave package. But I don't understand why there is no deviance residual for ZIP model? What is the underlying concept behind deviance for ZIP model?

Please clarify my doubts.

Ferdi
  • 4,882
  • 7
  • 42
  • 62
Sri Priya
  • 67
  • 7
  • Likelihood ratio and score tests are not applicable to non-nested models. Use the Vuong test or AIC/BIC. – Digio Sep 06 '18 at 12:24
  • @Digio Thank you very much. Could you explain why it is like this? – Sri Priya Sep 06 '18 at 12:28
  • This has been answered [here](https://stats.stackexchange.com/questions/70391/why-cant-likelihood-ratio-tests-be-used-for-non-nested-models). – Digio Sep 06 '18 at 12:33
  • Since the log-likelihood is well-defined for the ZIP, the deviance and the deviance residuals will be correspondingly well defined. I can't think of any particular reason that the deviance residuals wouldn't be defined for ZIP ... – Ben Bolker Sep 06 '18 at 15:03
  • @ Ben Bolker As you said, I tried calculating deviance residual manually using the predicted value using the formula d = 2 {y log (y/mu) - y + mu}, but I am getting infinite value. I think we couldn't get deviance residual from ZIP. Any suggestions?? – Sri Priya Sep 07 '18 at 06:15

0 Answers0