My dissertation focused on a new way to estimate adult mortality rates in poor countries where there are currently no reliable data on deaths. This problem is important because, as long as we can’t figure out how to measure adult mortality in the world’s poorest countries, we can’t measure life expectancy, understand the full impact of the HIV/AIDS epidemic, or investigate socioeconomic differences in health. In my dissertation, I develop a new method for estimating adult death rates from a household survey. The new strategy – which builds on existing techniques from demography, sociology, and anthropology – is based on the idea that we can ask survey respondents to tell us about deaths in their personal networks to learn about deaths in the general population. I apply the new method in a large household survey conducted in 27 Brazilian cities (see below). I also apply the method in a nationally-representative household survey from Rwanda, which is a country where adult death rates aren’t known, and new techniques are needed to estimate them (see below). Although my dissertation focuses on adult mortality, the new technique could be applied to estimate many other quantities that are currently hard to measure.
BrazilMy collaborators and I designed and implemented the largest network scale-up survey ever conducted, with n=25,000 respondents in 27 Brazilian cities. The survey allows me to apply the new method I develop in my dissertation to estimate adult death rates and out-migration rates in each city. For deaths and out-migration, there are gold-standard rates we can validate the new method against, meaning that we will have 27 different tests of the new method's accuracy for two different quantities.
RwandaMy collaborators and I designed and implementated a large, nationally-representative household survey from Rwanda (n=5,000). We collected the data needed to produce adult death rate estimates using the new method. (Adult death rates aren't known in Rwanda, and new techniques like the one my dissertation is based on are needed to estimate them.) We also conducted a survey experiment that showed that there may be a trade-off between the quantity and quality of information collected in a scale-up study. Our experimental design can serve as a template for future scale-up studies that seek to produce substantive estimates while simultaneously using an experiment to learn more about the methodology used to produce scale-up estimates.