Zero Inflated Poisson Regression Analysis in Maternal Death Cases on Java Island

Abstract

The basic regression model used to analyze the count data is the Poisson regression.. However, applying the Poisson regression model is unsuitable for excess zero data because it can cause overdispersion where the variance data is greater than its mean. One of the developments of the Poisson regression model can overcome this condition, Zero Inflated Poisson Regression (ZIP). In the health sector, the death of pregnant women on the Java island is an event that still rarely occurs and forms an excess zero data structure. However, the analysis of cases of maternal mortality using ZIP regression has never been studied in more depth. In this article, the maternal mortality cases in Java were modelled using ZIP regression to specify the variables that had a significant effect. The initial analysis results indicated the occurrence of overdispersion due to excess zero where there are 52% zero values in the data. The ZIP regression applied in this research provides enhancements to the Poisson regression based on the Vuong test. The results showed that the variables that had a significant effect on the maternal death cases in Java in the count model are the percentage of maternal health service coverage and the percentage of coverage of postpartum visit coverage, while in the zero-inflation model, the percentage of deliveries in health facilities and the percentage of obstetric complications treatment