Michael Harper (Flowminder Foundation)
Bayesian approaches are being increasingly used within spatial modelling, as they allow uncertainty to be faithfully represented in model outputs. These methods are particularly applicable for population modelling, where spatially disaggregated population data is required in the absence of national population and housing census data, as limitations in input datasets can result in large amounts of localised uncertainty in model outputs. However, the ability to use these full model results is hindered by both the quantity of data produced and the computational intensity required to analyse this data. As a result, aggregated results are typically provided to end users, which will produce a smaller and easier to use dataset, but at the expense of flexibility to interact with uncertainty estimates. This presentation will highlight the benefits of using population estimates including uncertainty, and showcases the use of an R package under development to support this analysis. This package is designed to allow both higher level visualisation of results and detailed analysis to be conducted against bayesian model outputs, providing a range of options for both decision-makers and data analysts alike. Within the context of Nigeria, the presentation will demonstrate the value of uncertainty estimates for some example use-cases, including vaccination programmes and infrastructure planning.