Zero-Inflated Poisson (ZIP) Regression
Zero-Inflated Poisson Regression (ZIP) · Also known as: ZIP regression, zero-inflated count model, Sıfır-Şişirilmiş Poisson Regresyonu (ZIP)
Zero-Inflated Poisson regression is a two-component model for count data that contains more zeros than an ordinary Poisson model can explain. Introduced by Diane Lambert in 1992, it combines a logistic model for the zero-generating mechanism with a Poisson model for the genuine counting process.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use ZIP when the dependent variable is a count and shows excess zeros beyond what a Poisson model predicts, typically with at least about 50 observations. It is appropriate when there is a plausible distinct mechanism generating structural zeros separate from the counting process, and the zero-inflation and count components can be driven by their own covariates. If the non-zero counts are also overdispersed, prefer the zero-inflated negative binomial extension.
Strengths & limitations
- Explicitly separates structural zeros from sampling zeros, so it fits count data with excess zeros far better than ordinary Poisson.
- Yields two interpretable sets of effects at once: incidence-rate ratios for the count process and odds ratios for the structural-zero process.
- The zero-inflation and count components can use different covariates, matching the substantive theory of how zeros arise.
- The basic ZIP still assumes the count component is Poisson, so it does not handle overdispersion in the positive counts — that calls for zero-inflated negative binomial.
- Requires a reasonable sample (about 50+) because two linked models are estimated simultaneously.
- Interpretation is harder than a single-equation model, and identifying which covariates belong in the zero versus the count part requires judgement.
Frequently asked
How is ZIP different from ordinary Poisson regression?
Poisson assumes a single process generates all counts, including zeros, and tends to under-predict zeros when they are abundant. ZIP adds a separate logistic component for structural zeros, so it can match an inflated number of zeros that Poisson cannot.
How do I know if I actually need a zero-inflated model?
Compare the fitted ZIP to a plain Poisson model with the Vuong test (a z above about 1.96 favours ZIP). If the test is not significant, the simpler Poisson model is preferable.
When should I switch to zero-inflated negative binomial (ZINB)?
When the non-zero counts are overdispersed — their variance clearly exceeds their mean — in addition to having excess zeros. ZINB replaces the Poisson count component with a negative binomial one to absorb that extra variability.
How do I interpret the coefficients?
Exponentiated count-model coefficients are incidence-rate ratios: the multiplicative change in expected count per unit of the predictor. Exponentiated zero-model coefficients are odds ratios for an observation being a structural (always-zero) case.
Sources
- Lambert, D. (1992). Zero-Inflated Poisson Regression, with an Application to Defects in Manufacturing. Technometrics, 34(1), 1–14. DOI: 10.2307/1269547 ↗
How to cite this page
ScholarGate. (2026, June 1). Zero-Inflated Poisson Regression (ZIP). ScholarGate. https://scholargate.app/en/statistics/zero-inflated-poisson
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Logistic RegressionResearch Statistics↔ compare
- Negative Binomial RegressionEconometrics↔ compare
- Poisson RegressionEconometrics↔ compare
- Zero-Inflated Negative Binomial RegressionStatistics↔ compare