Semiparametric estimation of a zero-inflated Poisson regression model with missing covariates

Abstract

Zero-inflated Poisson (ZIP) regression models have been widely used to study the effects of covariates in count data sets that have many zeros. However, often some covariates involved in ZIP regression modeling have missing values. Assuming that the selection probability is known or unknown and estimated via a non-parametric method, we propose the inverse probability weighting (IPW) method to estimate the parameters of the ZIP regression model with covariates missing at random. The asymptotic properties of the proposed estimators are studied in detail under certain regularity conditions. Both theoretical analysis and simulation results show that the semiparametric IPW estimator is more efficient than the true weight IPW estimator. The practical use of the proposed methodology is illustrated with data from a motorcycle survey of traffic regulations conducted in 2007 in Taiwan by the Ministry of Transportation and Communication.