Application of Count Models on Infant and Child Mortality in Nigeria: A Comparative Study

Background: Majority of studies on determinants of Infant and Child Mortality in Nigeria were mainly analyzed using binary models such as Logistic model, Hazard Proportional model, and Probit model which categorized infant mortality as a categorical variable. However, this study is aimed at comparing count models in identifying factors associated with Infant and Child mortality in Nigeria.

Methods: This study made use of 2008 Nigeria Demographic and Health Survey (NDHS) data. This was a stratified two-stage cluster sampling design study. The models used were Poisson model (POM), Zero-Inflated Poisson model (ZIPM), Negative Binomial model (NBIM), and Zero-Inflated Negative Binomial model (ZINBIM). Model selection criteria were Akaike Information Criterion (AIC) and Lilliefors test.

Results: Results from the model comparison for infant and child mortality showed that POM and ZINBIM had the smallest AIC values of 15167.95 and 22699.48 while the p-values for Lilliefors test for both models were 0.5553 and 0.3338 respectively, hence of best fit. Birth order, breastfeeding, parents’ education (primary/no education), toilet type (no facility, other type apart from pit latrine) and place of delivery (home) were identified to be positively associated with associated with childhood mortality (P<0.05) while Mother’s age, place of delivery (private/public hospital) and antenatal visits had negative association with childhood mortality hence experienced less child mortality.

Conclusion: Poisson model and Zero-Inflated Negative Binomial model were of best fit to model Infant and Child mortality in Nigeria respectively. Parents’ education, the practice of family planning and discouragement of mixed feeding will go a long way in reducing infant and child mortality in Nigeria.