Abstract

Zero-inflated bivariate count responses frequently occur in medical, environmental, ecological, social and transportation studies. These bivariate count responses recorded from a group of independent subjects or from a specific time or place may suffer the presence of excessive zeros, so constitute complexity while modelling. Although the responses under the umbrella of zero-inflated clustered and longitudinal count setup have been studied extensively in the past two decades, the literature is rather limited for analyzing zero-heavy bivariate count data. In this paper, we propose a marginal-conditional approach based bivariate zero-inflated Poisson-Poisson regression model to overcome the complexity imposed due to excessive zeros in bivariate count responses. The estimation and test procedures are developed using the likelihood function evolved from the marginal and conditional approaches. The proposed approach will be illustrated with the application to two traffic accidents and fatalities data collected from Virginia, USA and Groningen, The Netherlands.

Keywords: Correlated data; Poisson-poisson; Likelihood approach; Zero-heavy; Generalized linear model