Bayesian Modified Zero Hurdle and Zero-Inflated Models with Application to Under-Five Mortality in Nigeria

Abstract

Despite remarkable progress that has been made both locally and internationally to reducing underfive death mostly in developing countries, under-five death has continue to dominate and generate global discuss in many conferences, seminars and workshops on how to reduce its impact. The increase is largely due to lack of hygiene, poverty, HIV/AIDS, unhealthy lifestyle and corruption which has limited access to critical health infrastructure. In this paper we proposed a modified zeroinflated geometric (MZIG) and modified hurdle geometric (MHG) models to capture risk factors affecting under-five mortality in Nigeria.. The models assumed whether death occurred in the household in the first part with binomial logit link function and that the number of deaths in the households follows a truncated geometric distribution with log link function. Count models based on Standard Error, (SE) mean absolute bias (MAB), Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC) and Monte Carlo Standard deviation (MCSD) from simulation studies at different sample sizes and zero-inflation parameters were estimated. The MZIG and MHG models performed better than other standard count models and zero-inflated and hurdle regression models at some sample points. The models were applied to analyze under-five mortality data of households from NDHS with about eighty-eight percent zero observations.