The past decades have seen a flurry of methods that use extrapolation, smoothing, and other forms of information sharing to compensate for limited and incomplete data. In places without comprehensive vital registration or public health monitoring systems, extrapolation and information sharing techniques are particularly appealing, since there typically is simply not enough data available to produce estimates with sufficient temporal or spatial resolution to influence public health decision making. This proposal reframes uncertainty in extrapolated estimates of vital rates in terms of decision-making. A decision-making framework (i) is grounded in familiar language for policymakers and public health officials, (ii)characterizes consequential and inconsequential model decisions based on variability in outcomes, and (iii) in- corporates both extrapolation and sampling uncertainty. A cornerstone of this project is a novel collaboration with researchers and policymakers at the World Bank. Through this collaboration, we will pilot the proposed decision-making tools and conduct experiments with local and national policymakers in realistic settings.
Predictions based on machine learning models are increasingly common inputs into decision making processes across scientific domains. In this proposal I develop and evaluate strategies for making public health decisions based on predicted vital rates, particularly in places without full coverage civil registration. Results from the project will improve strategies for allocating resources for disease surveillance and health monitoring in scarce resource settings.
