Jessica Godwin
Published: 2021
Total Pages: 0
Get eBook
Child mortality is an important metric used in quantifying and monitoring the health of a population's children. Moreover, child mortality can be a key indicator of the overall health of a population, and is often used to quantify mortality at other ages. Over the past several decades, there have been huge global reductions in child mortality. However, child mortality remains large in many low and middle income countries (LMICs). The United Nations' Sustainable Development Goals (SDGs) call for a reduction of child mortality in the period 2015--2030. In particular, SDG 3.2 continues the global initiative to improve chid mortality outcomes by calling for an end to preventable child deaths and reaching a target under-five mortality rate (U5MR) of 25 deaths before age 5 per 1000 births and 12 deaths per 1000 births in the first month of life by the year 2030. Many methods exist for estimation of national child mortality measures, but there is a growing desire for more and better methods for subnational estimation of child mortality. Improved methods for subnational estimation of U5MR can allow for a better understanding of the geographic differences of trends in U5MR and more targeted intervention for child mortality reduction. Though there are many ways to think about geographic variability within a country, in this thesis we focus on the administrative divisions of a country that have political and infrastructural meaning. The national level of a country is referred to as Admin-0, the coarsest subnational administrative division is referred to as Admin-1, and the next coarsest as Admin-2. Many sources of child mortality data collect information for which Admin-1 estimates are reasonable, but this leads to small sample sizes at the Admin-2 level. The focus of this thesis is on subnational child mortality estimation at the Admin-2 level. In this thesis, we provide a review of demographic and statistical methods for age-specific period child mortality, synthesize notation across fields, and develop two methods for subnational estimation of U5MR at the Admin-2 data in LMICs. We make use of child mortality data from household surveys throughout this thesis and discuss in detail how an individual's mortality information is used in various existing demographic and statistical methods for mortality estimation. The two methods we develop for subnational child mortality estimation in LMICs at the Admin-2 level take different approaches to accounting for the method of data collection in the household surveys and one method incorporates census data. We apply the method developed to incorporate census data separately to the countries of Kenya and Malawi. We find mixed evidence of improvements in estimation with the incorporation of census data, and note the method's shortcomings when small sample sizes result in many Admin-2 areas with no observed deaths. The limitations of this method, motivate the need for a method that can be applied more broadly to countries with sparser coverage of household survey data. We develop a method and reproducible, replicable pipeline for data acquisition, cleaning, estimation, and visualization. We apply this method to Admin-1 and Admin-2 levels in 22 LMICs, fitting separate models for each country and administrative division. The country-specific modeling and steps in the pipeline allow us to address the unique context of child mortality in each country if the generic base model is inappropriate, such as in countries with generalized HIV/AIDs epidemics, the genocide events in Rwanda, and Cyclone Nargis in Myanmar. The estimates for 22 LMICs have been published in collaboration with UNICEF and the UN Inter-agency Group for Child Mortality Estimation, have undergone review in consultations with country representatives, and are available at \url{https://childmortality.org}. This thesis improves on the current understanding in the processing and use of child mortality data in the literature and develops two new methods for estimation of child mortality in LMICs at the Admin-2 level. In particular, the development of a clearly-defined and replicable pipeline that allows for easy adaptation to address the unique mortality estimation needs of individual countries fills a previous gap in existing methodology. However, there are areas for future research and methodological improvement. Throughout the thesis we make note of ongoing work to improve existing aspects of the developed methods and make clear the issues that remain unaddressed.