Simpler versus more complex population forecasting methods for regions and local areas: evidence from Australia

Tom Wilson, University of Queensland

If one is concerned with projecting just the total populations of regions and local areas, which type of model tends to provide the most accurate forecast: simple mathematical extrapolations or more complex methods such as cohort-component models? The question is important because the resources required to produce simple extrapolative projections are far less than those needed for complex methods. Demographers have debated the issue for many years, with many studies concluding that there is little difference between simpler and more complex methods in total population forecast accuracy. Most previous research, however, is based on work from the United States. This paper broadens the evidence base by presenting findings from Australia. Retrospective population forecasts of total populations were produced for local councils in Queensland (or aggregations of councils where boundary changes have been problematic). Four groups of simpler model were evaluated: (i) extrapolative models: (ii) various share models which take a share of a larger region’s projected population or growth; (iii) averaging methods, calculated as the mean of several simple methods; and (iv) composite methods, which take averages of a few simple methods selected on the basis of the local area’s demographic characteristics. Two types of more complex model were also assessed: the shortcut Hamilton-Perry cohort component model, and a cohort-component model incorporating directional migration. Forecasts were evaluated against Estimated Resident Populations produced by the Australian Bureau of Statistics. Reflecting the varied way in which forecast errors affect different users, several complementary measures of error and bias were used. In addition, the error in predicting the distribution of population across the State’s local councils was assessed. The results not only contribute to the wider debate on simpler versus more complex forecasting methods, but also have important implications for the way in which local council population forecasts are produced by State governments.

Presented in Session 40: Modelling demographic change in the context of uncertainty

´