Population Weighted Density
(PWD) is an alternative to conventional approaches to population density that is arguably better suited to some types of research in fields of social science and epidemiology. In this release WorldPop publishes what we believe may be the first set of global estimates for PWD, which we offer at national and subnational levels since 2000.
The traditional and most widely understood method for calculating an aggregate measure of human population density within any geographical region is simply to divide its total population by the total area (i.e. d = ΣP/ΣA). It has long been recognised in the field of geography and by many other scholars that this method has significant shortcomings for certain types of research, particularly in the human sciences and where the subject matter of interest may be related to the typical density levels experienced by the population, such as in epidemiology.
Population Weighted Density (PWD) — proposed by John Craig in 1984 is a family of methods that — as the name suggests — weight the density values by their corresponding population sizes in the aggregation process. We have utilised three distinct methods to generate PWD estimates: PWD-A (Arithmetic Mean), PWD-G (Geometric Mean) and PWD-M (Median).
PWD-A seems the most commonly used method for PWD estimation — this is the method adopted by the US Census — probably due to it being the most intuitive to understand. However it shares the arithmetic mean’s inherent vulnerability of being highly sensitive to outliers, which are common in population density data. Hence any statistical error in the top-end of density values also has a disproportionate effect on the PWD-A estimate. PWD-G is a more robust metric that for lognormal data distributions will tend towards the median value, and is therefore less vulnerable (though not immune) to outliers. PWD-M by contrast is largely immune to outliers, though this does comes at the cost of being largely insensitive to the region’s top-end densities.
We encourage users to explore the data as well as the theory to come to a view on which PWD method and which source resolution are most suited to their research, though for general usage we would advise using PWD-G or PWD-M.