Statistical Methods for Human Rights and Child Mortality Estimation

Abstract

This dissertation addresses statistical methodology commonly used in human rights research and child mortality estimation. We first consider two related problems, record linkage and multiple-systems estimation, typically used to estimate the number of civilian casualties in the wake of a conflict when probability surveys are not available, and then consider the problem of estimating child mortality over time in a country that has experienced conflict. In Chapter 2, we propose a novel Bayesian approach for record linkage in the general setting where there may be any number of files, with arbitrary patterns of duplication across files. In Chapter 3, we present a re-framing of multiple-systems estimation which places identifying assumptions front and center in the multiple-systems estimation workflow, and examine how common models fit into this framing. In Chapter 4, we develop spatial and temporal smoothing models which incorporate knowledge of expected shocks in child mortality, such as the timing of a conflict, leading to estimates of child mortality which are not oversmoothed. Finally, we conclude with discussion of future work in Chapter 5.