BINOMIAL-MIXTURE MODELS ON UNDER-FIVE MORTALITY IN NIGERIA

Abstract

The reduction of under-five mortality (U5M) in Nigeria remains a challenge. Many studies have looked into factors associated with U5M by using single or standard models such as binary logistic regression, cox-proportional hazards, regression and probit models for categorical response variable. To accurately estimate the impact of these factors on U5M, a mixture model of this type of binary response variable, is needed. This study was aimed at addressing this situation. The data from Nigeria Demographic and Health Survey 2018 (NDHS 2018) were analyzed among mothers aged 15 – 49 years. The response variable was “died” or “alive” for each child birth. The Binomial Mixture models (Logit, Probit, Cloglog and Zero Inflated Binomial models) was used. The Akaike Information Criterion (AIC) and Residual Deviance (RED) were the criteria used to select the model. The lowest AIC and RED indicated the best model. The models had the following AIC and RED: (52129; 33103), (52170; 33144), (52114; 33088) and (50854; 29865) for Logit, Probit, Cloglog, and Zero Inflated Binomial Models respectively. The best model was the Zero Inflated Binomial Model because it had the lowest AIC and RED. The risk factors associated with U5M in Nigeria are living in the North- East and North West, Parents Education, Religion, wealth index, Types of toilet and Breast-feeding.Zero Inflated Binomial model was the best model for measuring factors associated with U5M in Nigeria. The risks factors identified in this study should be given serious attention, especially the education of both parents. Also, health education, be facilitated on importance of exclusive breast-feeding.