2
$\begingroup$

I am using STATA to analyze count data (weekly disease counts), and I am trying to pick the best test between Poisson, negative binomial and zero-inflated Poisson / negative binomial but I am not sure how to do this. My outcome variable has a variance that is much larger than the mean and contains a large number of zeros.

I've read a couple of articles mentioning that I should look at the predicted / actual residuals, but I don't know how to do it or what I should be looking for to pick the right test.

$\endgroup$
  • $\begingroup$ What do you actually want to test or rather what kind of model do you want to fit ? A GLM or GLMM, for example ? It sounds like you have zero inflated data, so a ZIP or ZINB might be appropriate. Can you post histogram of the data ? $\endgroup$ – Robert Long Jun 11 '16 at 21:57
  • $\begingroup$ Thank you for your reply! I have continuous and categorical predictors for the disease outcome and my outcome variable is over dispersed: $\endgroup$ – MargotC Jun 11 '16 at 22:03
  • $\begingroup$ Is it overdispersed and zero-inflated ? $\endgroup$ – Robert Long Jun 11 '16 at 22:10
  • $\begingroup$ Thanks again, one last question: when I use zip in Stata, how do I know which variable I should use for the inflate() section of the code? Thanks! $\endgroup$ – MargotC Jun 11 '16 at 22:12
  • $\begingroup$ inflate() is used to specify one or more predictors of the excess zeros. It is optional. If my answer was useful, you can upvote it :) $\endgroup$ – Robert Long Jun 11 '16 at 22:29
2
$\begingroup$

You can start by fitting a zero-inflated poisson model and a zero inflated negative binomial model. The latter is particularly suited if your outcome data are overdispersed as well as zero inflated.

In Stata you can use zip and zinb, for example:

use http://www.stata-press.com/data/r10/fish, clear
zip count child camper, inflate(persons) vuong

webuse fish, clear
zinb count child i.camper, inflate(persons) vuong zip

You can compare models using AIC or BIC and by looking at their predictions.

$\endgroup$

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.