Hierarchical multiple linear regression : a comparative analysis of classical and Bayesian approach.
Gabela, Mzwandile Lindokuhle.
MetadataShow full item record
Low birth weight is a problem in Africa due to its contribution to high infant mortality. Most studies on low birth weight have neglected the use of Bayesian methods in analysis of medical data. This study aims to investigate the risk factors of low birth weight in Malawi. Malawi is a country in the sub-Saharan region which is characterized by infant and child mortality of 12%. Inferences made in this study are only based on Malawi demographic and health survey data for the three years 2000, 2004 and 2010. The year 2010 Malawi Demographic and Health Survey data is used to make classical and Bayesian analysis for the study. The years 2000 and 2004 data are used to set up the prior information for Bayesian approach to the analysis. Data will be analysed using descriptive and inferential statistics. Hierarchy is taken into consideration since some of the risk factors are known to be hierarchical. The hierarchical multivariate linear regression analysis is done in a comparative procedure of Classical and Bayesian approach. The study shows that the age of the mother, birth order number of a child, region, work during pregnancy and HIV status of the mother are significant determinants of birth weight. In the comparison of classical and Bayesian approach it was found that all the variables that were significant in the Bayesian approach were also significant in the classical approach and the opposite was not true. The use of the Bayesian prior in the analysis gave more realistic results on factors of weight at birth.