BAYESIAN SPATIAL JOINT MODELING OF ZERO-INFLATED COUNT DATA WITH APPLICATION TO ANTENATAL CARE UTILIZATION IN NIGERIA

Abstract

Antenatal care (ANC) utilization is lower in most developing countries despite policy innovations brought about by the Millennium Development Goals and the Sustainable Development Goals, aimed at improving access globally. Though there are slight improvements in utilization in Nigeria, a large proportion of women still make no attendance leading to the problem of large proportion of zero values in the data obtained. Analyzing such data with classical regression models is not the best because the models can provide biased inference due to excess zero values, overdispersion, spatial and random effects in the dataset. This study used 2018 Nigeria Demographic and Health Survey dataset, containing a sample of 21,447 women of reproductive age (15-49 years). Joint modeling for zero inflated count data was used to analyze the binary and count parts of the ANC visit response data under Bayesian framework. The spatial components were modeled using Besag, BYM and BYM2 in R-INLA. Results show that most of the socioeconomic and demographic variables have high significant influence on ANC utilization in Nigeria. There appears to be a lower likelihood of attendance in Adamawa, Bauchi, Borno, Gombe, Yobe, Kogi, Enugu, Anambra, Bayelsa, Imo, Lagos, Ogun, Oyo, and Osun states. Findings suggest that more attention should be given to the women living in the states with lower likelihood of attendance, through increase in the provision of health services, educating them on the importance of ANC and making the antenatal care easily accessible.