Heather Chamberlain (University of Southampton)
National population and housing censuses provide essential sociodemographic data on a country’s population and form the basis for many decisions concerning service provision and resource allocation. Censuses are typically conducted by a country every ten years, however they are a major undertaking with many logistical requirements. Some countries may not be able to conduct a census regularly with national coverage, for example due to insecurity or inaccessibility. This can lead to official population estimates being based on censuses from several decades previously, projected to the current year with high uncertainty and often also at coarse spatial resolution. This presentation will discuss a hybrid census approach, combining spatially incomplete national census data or population counts from enumeration surveys with geospatial datasets, derived from satellite imagery and other sources, within a statistical modelling framework. A Bayesian model is developed using gridded geospatial datasets associated with population distribution and density in areas with population counts. The predictive model can then be used to estimate population in areas with no population counts. Model outputs take the form of gridded population estimates and associated measures of uncertainty at a spatial resolution of approximately 100m, with national coverage. Examples of work with UNFPA and national statistical agencies in Afghanistan and the Democratic Republic of the Congo will be presented. The hybrid census approach has particular relevance where a population and housing census cannot be conducted with national coverage, and also has potential to improve population estimates during the inter-censal period for other countries more generally.