Multilevel modelling of determinants of contraceptive method choice among women in South Africa

Abstract

Multilevel models take into account various degrees of aggregation in the data. This study aims to bring together multilevel models from both frequentist and Bayesian perspectives in identifying determinants of contraceptive choices. The study uses the data from the 2016 South African Demographic and Health Survey (SADHS). To analyse the dataset, a multinomial logistic regression model has been used, model parameters were estimated in SPSS for frequentist models. The Bayesian analyses with non informative priors were strengthened by the use of the state of the art Hamiltonian Monte Carlo algorithm (HMC), as implemented in the RStan package in the R statistical software. The Bayesian nal model was selected based on Watanabe{Akaike information criterion (WAIC), which has been shown to outperform conventional information-criterion such as DIC. The results established that an individual woman's choice of contraception is a function of both individual characteristics and community e ects. In bivariate analysis, injections showed a continued dominance as a preferred choice in SA. Community level education was the most useful determinant of contraceptive choices. Thus, this study recommends that Empowering woman through education, will have a positive e ect on overall contraceptive prevalence.