Small Area Estimation of Under-5 Mortality in Zambia

Abstract

<bold>Background</bold> Under-5 mortality estimates at the district level are currently unavailable in Zambia. Estimates at this level are useful for planning and evaluative purposes but are difficult to produce due to challenges with the data available at this level. This study aims to produce estimates of under-5 mortality for each district every year from 1980 to 2010. <bold>Methods</bold> Different methods for analyzing birth history data from censuses and surveys are tested to evaluate their usefulness for estimating under-5 mortality in the context of small sample sizes using a validation framework based on data from demographic and health surveys. Data from three demographic and health surveys and three population censuses are then analyzed using the best performing methods. Five small area models are tested for synthesizing estimates from various sources and a best performing model is selected. Bootstrapping and other simulation methods are utilized for approximating uncertainty in the final estimates that is due to sampling and model error at each stage of the estimation process. <bold>Findings</bold> A moving window variant on complete birth history methods is selected as the best performing method, with summary birth history methods selected as an alternative when complete birth history data are unavailable. Of the five models tested, a natural cubic spline model with knots at 10 year intervals is found to perform best. Once both sampling and model uncertainty are accounted for, there is large uncertainty around estimates produced of under-5 mortality. Under-5 mortality levels vary considerably among districts in Zambia, as do rates of decline across the period 1980 to 2010. Nonetheless, most districts have experienced declines in mortality over this period, though rates of decline are similarly heterogeneous and most districts continue to experience high under-5 mortality.