Bayesian conditional negative binomial autoregressive model: a case study of stunting on Java Island in 2021

Abstract

Sustainable Development Goals (SDGs) include stunting as one of its worldwide objectives, which requires an average annual reduction in the stunting rate of about 2.7%. Therefore, mapping stunting cases along with the underlying risk factors is crucial in Indonesia, especially on the island of Java, which has the largest population. The result can then be taken into consideration for policymaking to address stunting cases, particularly in Java and in Indonesia in general. This research aims to analyze stunting cases on the island of Java using the best models among GLM (Generalized Linear Model), GLMM (Generalised Linear Mixed Models), ICAR (Intrinsic Autoregressive), and CAR BYM (Besag York Mollie) with a negative binomial distribution. The data used in this study were obtained from the 2021 Health Profile of all provinces in Java, with predictors including the percentage of infants with exclusive breastfeeding, the percentage of complete basic immunization, the percentage of families with access to adequate sanitation, and the percentage of poverty rates. The study's findings reveal that the CAR BYM model has the best forecasts since it has the lowest DIC (Deviance Information Criterion) and MAD (Mean Absolute Deviance) values. The significant predictors are the percentage of families with access to adequate sanitation (X₃) and the percentage of poverty rates (X₄). A relative risk greater than one exists in 31.1% of Java's districts and cities, the majority of which are found in West Java and Central Java. Bogor district/city has the highest relative risk (RR) of stunting (RR = 5.833), followed by Bandung district/city (RR = 3.721), Tegal (RR = 3.291), and Brebes (RR = 2.252). Meanwhile, areas with low risk are found in districts and cities in the Special Region of Yogyakarta (RR = 0.610).