Comparer des méthodes

	Modèle à inflation de zéros	Régression de Poisson robuste
Domaine	Statistique	Statistique
Famille	Regression model	Regression model
Année d'origine	1992	2004
Auteur d'origine	Diane Lambert	Guangyong Zou
Type	Count regression with excess zeros	GLM with robust variance
Source fondatrice	Lambert, D. (1992). Zero-inflated Poisson regression, with an application to defects in manufacturing. Technometrics, 34(1), 1–14.	Zou, G. (2004). A modified Poisson regression approach to prospective studies with binary data. American Journal of Epidemiology, 159(7), 702-706.
Alias	ZIP model, ZINB model, zero-inflated Poisson, zero-inflated negative binomial	modified Poisson regression, Poisson regression with robust standard errors, log-binomial alternative, sandwich-variance Poisson
Apparentées	6	5
Résumé	A zero-inflated model is a two-component mixture regression designed for count outcomes that contain more zero values than a standard Poisson or negative binomial distribution can accommodate. One component is a binary process that generates structural zeros; the other is a count process that generates both zeros and positive counts.	Robust Poisson regression fits a Poisson log-linear model to a binary outcome but replaces the model-based variance with the empirical sandwich estimator. This yields valid standard errors and risk ratios even though Poisson variance assumptions are technically violated for binary data. The approach, popularized by Zou (2004), is widely used in epidemiology as a numerically stable alternative to log-binomial regression.