Statistical Methods for Official Statistics and Mortality Estimation

Abstract

This dissertation develops statistical methodology for small area estimation, the analysis of official statistics, and mortality estimation. All methods are developed with the statistical challenges faced in data from low- and middle-income countries in mind. We first consider a problem in small area estimation known as benchmarking, where subnational estimates must agree in some sense with a national estimate or estimates. We then consider two related problems addressing data challenges specific to low- and middle-income country child mortality data, where data comes from a survey, is interval censored, and may need to be smoothed across time to ensure reasonable and precise estimates in situations with little data. The outline of the dissertation is as follows. In Chapter 2, we propose a benchmarking method for subnational estimates of a proportion in a setting where a national estimate is available with uncertainty. In Chapter 3 we develop a pseudo-likelihood approach to mortality estimation that allows us to obtain continuous mortality curves for children under the age of 5 across time from interval-censored survey data. In Chapter 4 we consider different temporal smoothing models for the direct estimates produced in Chapter 3, and develop a multivariate random walk prior to simultaneously smooth multiple, correlated summaries across time. Finally, we conclude with discussion of future work in Chapter 5.