Zero-Inflated Poisson (ZIP) Regression
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.
Key highlights
- 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.
Intuition
This section is available to Pro members. Upgrade to Pro
How it works
This section is available to Pro members. Upgrade to Pro
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.
Common pitfalls
This section is available to Pro members. Upgrade to Pro
Applications
This section is available to Pro members. Upgrade to Pro
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
- 1.Lambert, D. (1992). Zero-Inflated Poisson Regression, with an Application to Defects in Manufacturing. Technometrics, 34(1), 1–14.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 1). Zero-Inflated Poisson Regression. ScholarGate. https://scholargate.app/statistics/zero-inflated-poisson