Bayesian and Frequentist Comparison: An Application to Low Birth Weight Babies in Ghana

Abstract

The aim of this study is to evaluate the association between maternal factors and birth weight among babies by using and comparing frequentist and Bayesian methods’ results from an epidemiologist or public health point of view. Low birth weight babies, defined by WHO as babies born at term who weigh less than 2.5 kg is an important indicator of reproductive health and general health status of any Population. The incidence of low birth weight is quite high in the sub region which has a public health concern. Our study was based on data from 2011 Multiple Indicator Cluster Survey conducted by Ghana Statistical Service. A total sample size of 10,963 women within the reproductive age were selected throughout the entire country for the survey. The results from the frequentist and the Bayesian models show that, the two approaches can yield similar results using same data set. However, there are factors that the Bayesian technique can unfold which might not be the case using the frequentist model. We were able to show with our data set that the Bayesian method may have a lot of benefits than the frequentist method. However, in order to narrow the credible intervals, there is the need to bring in informative priors so as to be able to well formulate the null and the alternative hypotheses. However, one can use the Markov Chain Monte Carlo, when using no priors to predict reliable results. Comparing the two approaches with respect to our data set, we can infer (from Table 4) that using Bayesian model provides better estimates in predicting low birth weight among babies in Ghana. We note however that to better understand the phenomenon under study the two methods could be used together. Our findings further revealed that low birth weight is not only a public health problem but also a socio-cultural issue.