New Evaluations of Simple Models for Small Area Population Forecasts

Abstract

At the small area scale simple methods for forecasting total populations are often employed because of a lack of data for cohort‐component models, concerns about the reliability of these models for forecasting small population totals, and resource constraints. To date, a select number of authors have assessed the forecast accuracy of several individual, averaged, and composite models. This paper extends this stream of work by evaluating a large number of models on new datasets. The aims of the paper are to examine the performance of (a) 10 individual forecasting models (some of which are well known; others less so); (b) averages of every combination of 2, 3, 4, and 5 of the individual models (627 in total); and (c) composite models based on population size and growth rates (200,000 in total). Do averaged and composite models outperform individual models? Using new small area population datasets, forecasts from 2001 to 2031 were produced for three case study countries, Australia, New Zealand, and England & Wales. Both forecast accuracy and credibility (avoidance of negatives; degree of constraining to state populations) were assessed in 2011; for 2031, just credibility was evaluated. Of the individual models, constant share of growth (positive shares only) and constant share of population performed the best. A small proportion of averaged and composite models outperformed the best individual models in forecast accuracy. Several recommendations for the practice of small area population forecasting are made. Copyright © 2014 John Wiley & Sons, Ltd.

Continuous and consistent land use/cover change estimates using socio-ecological data

“…Population density estimates vary widely (Wilson, 2014), and given its fundamental importance to the proposed model framework, future work should aim to integrate a more dynamic product that better accounts for interannual variability and realistic representation of current and projected population density. To the authors' knowledge, this was the first attempt to make a population product dynamic.…”

Evaluation of Alternative Cohort-Component Models for Local Area Population Forecasts

“…The final forecasts are the mean of the two models' outputs. This averaged model was shown by Wilson (2014a) to produce more accurate forecasts for local areas than any single extrapolative model.…”

“…In the Principles of Forecasting, the renowned forecasting expert J Scott Armstrong recommends: ''to improve forecasting accuracy, combine forecasts derived from methods that differ substantially and draw from different sources of information'' (Armstrong 2001, p. 417). In previous work by the author, an averaged extrapolative model for total population, the Constant Share of PopulationVariable Share of Growth (CSP-VSG) model, was found to produce accurate forecasts (Wilson 2014a). Forecasts from this model were used as total population constraints on forecasts generated by the five models listed above to produce five further sets of forecasts.…”

“…The five models are applied in two forms, first without any constraints, and then with age-specific forecasts constrained to total populations projected by a simple extrapolative model which performed well in earlier research (Wilson 2014a). Retrospective forecasts were produced over a 20-year forecast horizon, 1991-2011, for 67 local government areas of New South Wales which did not undergo boundary changes over the period.…”

Sea-level rise and sub-county population projections in coastal Georgia

“…A LIN/EXP model is used to ensure that (1) long-range linear projections of decline do not project negative populations and (2) that long-range exponential projections of growth do not produce extreme values of runaway growth. Recent research suggests that a LIN/EXP model outperforms both a linear and an exponential model, respectively (Wilson 2014). Included within the regression formulas is an adjustment factor allowing for the projected and observed populations at launch year to be identical.…”