topic Zero-Inflated Poisson Regression in Statistical Procedures

Zero-Inflated Poisson Regression

Jeanne — Fri, 25 Nov 2016 10:16:11 GMT

Hello everyone,

I was wondering if someone knew about Zero-inflated poisson models, especially how to chose the independent variables for the logit model of the ZIP, and how this choice could impact the results of the ZIP.

Thank you.

Re: Zero-Inflated Poisson Regression

PGStats — Sat, 26 Nov 2016 05:02:31 GMT

Unless you are working on a an abstract problem where variables have no meaning, you should be able to describe your model from prior knowledge. For example, you should be able to hypothesize the origin of extra zeros in your data.

I prefer to start with a simple model and to refine it by looking at the fit.

Others like to put every possible effect in their original model and to simplify it by eliminating non-significant terms.

Re: Zero-Inflated Poisson Regression

Jeanne — Mon, 28 Nov 2016 08:00:34 GMT

Thank you for your reply.

I do not work on an abstract model so as you say I should find independent variables to explain the excess zeros.

Re: Zero-Inflated Poisson Regression

StatDave — Thu, 01 Dec 2016 22:09:21 GMT

Note that variable selection methods are available for zero-inflated models in PROC HPGENSELECT.