This paper looks at a number of fertility indicators including levels, patterns, and trends using 1999 to 2008 Nigeria Demographic Health survey datasets. In a typical regression situation where the dependence of an outcome variable is to be modelled on several independent variables, a better way to achieve a more parsimonious model is to consider screening of the independent variables. In this paper, we adopted a modelling technique that permits screening and selection of the determinants of fertility using a Bayesian stepwise geoadditive regression model.

Often, effects of continuous covariates are modelled by assuming a linear dependence of the outcome variable on the predictor. Assumption of linear dependence is often too rigid in many realistically complex situations. Therefore, classical parametric regression models for analysing fertility data have severe problems with estimating small area effects and simultaneously adjusting for other covariates, in particular when the effects of some covariates are non-linear or time-varying. Therefore, flexible geoadditive approaches are needed which allow one to incorporate small area spatial effects, non-linear or time-varying effects of covariates and usual linear effects in a joint model.

Variable selection for complex regression models has been an area of research that recently attracts attention. In realistically complex models, the decision as to which variables to include in a model, inclusion of continuous covariates as linear or nonlinear, etc., was difficult to make. Therefore, we adopt a Poisson stepwise geoadditive regression approach through a Bayesian perspective for screening and selection of the variables.

Model that excludes cluster and spatial random effects best fits the data. Education, ideal number of children desire, husband's desire for more children were found to be significantly associated with fertility. Findings from this analysis will enhance effective policy formulation towards attaining the Millennium Development Goals on reduction of child and maternal mortality rates.

Keywords: Variable selection and model diagnostics; Parsimonious models; Millennium development goals; Geoadditive regression

Biography: Samson B Adebayo is an Associate Director and Head of Research and Evaluation Division of the Society for Family Health, a leading public health NGO in Nigeria that uses evidence-informed decision making through social marketing approaches. Samson has verse background in academic having lectured at University level in Nigeria and UK. He specializes in Bayesian semiparametric modelling techniques using Markov chain Monte Carlo and restricted maximum likelihood techniques. He has served on a number of committees on national surveys in Nigeria on public health: maternal and child health, malaria, family planning and reproductive health, HIV & AIDS, Demographic and Health Survey, etc.