Count data regression modeling: an application to spontaneous abortion

Abstract Background In India, around 20,000 women die every year due to abortion-related complications. In count data modeling, there is sometimes a prevalence of zero counts. This article is concerned with the estimation of various count regression models to predict the average number of spontaneous abortions among women in Punjab and few northern states in India. The study also assesses the factors associated with the number of spontaneous abortions. Methods This study includes 27,173 married women of Punjab obtained from the DLHS-4 survey (2012–13) to train the count models. The study predicts the average number of spontaneous abortions using various count regression models, and also identifies the determinants affecting the spontaneous abortions. Further, the best model is validated with other northern states of India using the latest data (NFHS-4, 2015–16). Results Statistical comparisons among four estimation methods reveals that the ZINB model provides the best prediction for the number of spontaneous abortions. The study suggests total children born to a woman, antenatal care (ANC) place, place of residence, woman’s education, and economic status are the most significant factors affecting the instance of spontaneous abortion. Conclusions This article offers a practical demonstration of techniques designed to handle count outcome variables. The statistical comparisons among four estimation models revealed that the ZINB model provides the best prediction for the number of spontaneous abortions, and it suggests policymakers to use this model to predict the number of spontaneous abortions. The study recommends promoting higher education among women in Punjab and other northern states of India. It also suggests that women must receive institutional antenatal care and have a limited number of children.