Does Averaging Yield More Accurate Local and Regional Population Forecasts?

Local and regional population forecasts inform a wide range of planning and budgeting activities, including those concerning educational provision, transport services, health facilities, electoral redistricting, and business location decisions. Unfortunately such forecasts often prove to be quite inaccurate. The aim of this paper is to evaluate a simple model for forecasting local and regional total populations in Australia which takes the average of two extrapolative methods. This is the Constant Share of Population - Variable Share of Growth (CSP-VSG) model, shown to have performed well at the local area scale in earlier research. This study extends that earlier work, making use of recently available historical local area population estimates on the current set of geographical areas. It reports on retrospective tests of the averaged model in Australia over several forecasting periods, and at three geographical scales. Forecasts are produced for three ten year forecast horizons (1991-2001, 1996-2006, and 2001-2011) and comparisons are made with simple linear extrapolation. It is shown that for all geographical scales and forecast horizons the averaged model generally produces more accurate population forecasts. The value added over linear extrapolation is greater for the smallest areas and those with the highest (positive or negative) base period growth rates. Applying the averaged model to non-metropolitan regions only results in further gains in accuracy. It is argued that the averaged model is a useful addition to the population forecaster's toolkit: it produces forecasts of respectable accuracy with low input data requirements and production costs.