Space-Time Modeling of Under-Five Mortality in India: A Small-Area Analysis of the National Family Health Survey (306516)*Miranda Fix, University of Washington
Jon Wakefield, University of Washington
Keywords: under-five mortality rate, small-area estimation, space-time smoothing
The under-five mortality rate (U5MR) is a key health indicator in low and medium income countries. In many such countries, there is limited vital registration, and estimates of U5MR are based on survey and census data. Although U5MR has been decreasing in recent years, there is considerable variability in subnational estimates, and spatial disaggregation is highlighted in the Sustainable Development Goals. In this study, we use the National Family Health Survey (NFHS) of India to produce yearly U5MR estimates in each of the Admin-2 areas of India. In each Admin-1 region, direct weighted estimates of the U5MR are calculated for each of the constituent Admin-2 regions, across five-year periods, accounting for the complex survey design. We then fit a Bayesian space-time smoothing model to these direct estimates. This step alleviates issues of data sparsity by sharing information between neighbors in space and time, and produces estimates with smaller variance and higher signal-to-noise ratio. By modeling each Admin-1 region separately we allow the space-time smoothing to differ across the country. The method is implemented via integrated nested Laplace approximation (INLA).