Comparison of Statistical Models Used for Assessing Factors Associated with Infant Mortality in Nigeria

Background: Infant mortality is a public health concern especially in developing countries, particularly Nigeria. Different models had been used independently to identify factors associated with infant mortality. Some of the used models sometimes violate the underlying assumption for the models. This study was designed to compare the models that have been previously used and identify the appropriate model using standard model selection criteria to analyse risk factors for infant mortality in Nigeria. Methods: The study utilised 2008 Nigeria Demographic and Health Survey (NDHS) data with a sample size of 7107. The NDHS was a stratified two-stage cluster design where a questionnaire was used to collect data on the birth history of women aged 15-49 years. The models employed for this study were: Logit, Probit and Clog-log. The model selection criteria were Akaike Information Criterion (AIC), Residual Deviance and Vuong test. The model with the smallest criteria was considered to be the best fit. Results: The results showed that Infant Mortality in Nigeria can be appropriately modelled by Clog-log model. The models and corresponding AIC values were: Logit (6171.1), Probit (6212.6) and Clog-log (6126.6). The residual deviance included: Logit (6135.1), Probit (6176.6) and Clog-log (6090.6). Clog-log had the smallest AIC and residual deviance values; hence, it was of the best fit. Home delivery and delivery by professionals had negative significant associations with infant mortality while women's education (primary/no education) and birth order had positive significant association, (p < 0.05). Conclusion: The best model for infant mortality evaluation in Nigeria was Clog-log. Generally, improved women’s education would significantly reduce Infant Mortality in Nigeria.