Multilevel Multiple Imputation of Missing Birth Weights in Developing Countries: Analysing Neonatal and Post-Neonatal Mortality

Andrew R. Channon, University of Southampton

Many infants in developing countries are not weighed at birth, resulting in many infants not having a recorded birth weight in retrospective surveys such as Demographic and Health Surveys. Models studying the determinants of infant mortality will therefore be biased if only using infants with a recorded birth weight are used. Using a recently developed method, termed multilevel multiple imputation, birth weights can be imputed which takes account of the hierarchical organisation of the survey data. This paper assesses this method by imputing birth weights for three countries, Cambodia, Kazakhstan and Malawi. The results indicate that the imputed birth weights are accurate with only minimal error. Furthermore, models of the determinants of neonatal and post-neonatal mortality using the imputed data produce expected parameter estimates. It can therefore be concluded that multilevel multiple imputation is a good method to use when using birth weight to model infant health outcomes.