Error and Bias in under-Five Mortality Estimates Derived from Birth Histories with Small Sample Sizes

Laura A. Dwyer-Lindgren, University of Washington
Emmanuela Gakidou, University of Washington
Abraham Flaxman, University of Washington
Haidong Wang, University of Washington, Seattle

Complete and summary birth histories collected in surveys and censuses are routinely used for estimating under-5 mortality at the national level. Subnational and stratified analyses are also desirable, but the usefulness of estimates of under-5 mortality derived from small samples is unknown. We perform a data-based simulation study using Demographic and Health Survey data to quantify the expected magnitude and direction of error associated with estimates derived from birth history data with small samples. We find all methods are prone to high levels of error at small sample sizes but performance is not uniform across methods. We also perform stratified analyses to test how the performance of each method varies by level of true mortality and time prior to survey. We find that for summary birth history methods and complete birth history methods that involve considerable smoothing over time performance varies by level of true mortality and time prior to survey.