Zero-inflated Upper Truncated PoissonRegression Model: An application to Miscarriage Data

Abstract

Modeling count data is important in many application fields. In recent years, the count data regression models are often used. The poisson regression is most commonly used. However, this model assumes the equidispersion of the data. When there is the excess of zeros in the observed data, the Zero-inflated Poisson regression model fits better. In many situations, the dependent variable is restricted to be observed on some range. In this case, truncated distributions are commonly used by researchers. In this paper, Zero-inflated Upper Truncated Poisson regression model is proposed and implemented to miscarriage data to determine the factors that affect the miscarriage counts. Zero-inflated Upper Truncated Poisson regression model is compared with Poisson regression and Zero-inflated Poisson regression models according to the model selection criteria. Empirical findings show that the Zero-inflated Upper Truncated Poisson regression outperforms to model the number of miscarriage and the most important factor of miscarriage counts is obtained as the number of previous pregnancies.