Quantifying Selection Bias from Birth History Estimates of Child Mortality
MetadataShow full item record
University of Washington Abstract Quantifying Selection Bias from Birth History Estimates of Child Mortality Matthew Coates Chair of the Supervisory Committee: Haidong Wang, Associate Professor Global Health Introduction Child mortality rates have long been used as an indicator of progress in global health and development. Their measurement in developing settings is often based on birth history data provided by mothers in surveys; however, these measures miss the mortality risk in orphans due to selection bias in only sampling mothers who are alive. Methods exist to address this bias using HIV projection models, but comprehensive estimates of the bias from nationally representative survey data do not exist. Methods I have developed an approach to estimating this selection bias, using sibling survival methods to estimate the ratio of orphan to non-orphan mortality and household survey data to estimate the prevalence of maternal orphans. In particular, I have focused on 48 Demographic and Health Surveys (DHS) in Sub-Saharan Africa and applied established methods to estimate a correction factor. Results This method has found significant, though often small bias in child mortality estimates using complete birth histories in 36 out of 48 surveys. The largest bias was found in surveys from Namibia and Zimbabwe with around a 10% increase in under-5 mortality estimates when accounting for the bias. Discussion The method shows mixed consistency with the existing methods, and several challenges exist to producing reliable estimates. Sibling survival sample sizes for the groups required by this analysis are often insufficient, and biases inherent to sibling survival methods in estimating child mortality need to be further addressed. Extensions of this method offer the potential for using nationally representative survey data to create geographically comprehensive adjustments for complete birth history estimates of child mortality due to selection bias.
- Global health