Paper presented at the annual meeting of the Southern Demographic Association, Hilton Head, South Carolina This paper evaluates summary measures of population projection accuracy and bias for a large sample of counties and county equivalents in the continental United States over the period 1900–2000. The analysis has two primary purposes. The first is to investigate the relationship between accuracy and bias and the length of the projection horizon and base period. The second is to compare different trend extrapolation techniques with respect to their forecasting performance, both in the aggregate and by county size and growth rates. The study finds that the length of the base period has only a limited impact on accuracy and bias; that errors grow about linearly with increases in the projection horizon; that most projection methods provide comparable results for shorter projection horizons; that accuracy and bias vary by population size and growth rate; and that averages generally perform very well, equaling or exceeding the performance of individual techniques. The study confirms many of the findings of the earlier projection evaluation literature. By using a significantly enlarged dataset both with respect to space and time, it conclusions strengthen those previous studies and provide guidance regarding the production and interpretation of small area population projections.
Assessing the Accuracy of Trend Extrapolation Methods for Population Projections: The Long View
Thursday, October 14, 2004