Measuring Uncertainty in Population Forecasts by Age

David A. Swanson, University of California, Riverside
Jeff Tayman, University of California, San Diego

Three basic approaches have been used to assess population forecast uncertainty: (1) judgment and personal opinion; (2) a range of projections based on alternative scenarios; and (3) statistical forecast intervals. In terms of the latter, there are two complementary approaches: (1) model-based intervals; and (2) empirically-based intervals. We evaluate a model-based approach in this paper, but enhance it by using it the information in historical data, a feature found in the empirically-based approach. We describe and test in this paper a regression-based approach for developing 66% forecast intervals for 16 age-group forecasts made using the Hamilton-Perry Method. We use a sample of four states (one from each census region in the United States) with nine ex post facto tests, one for each census from 1930 to 2010, which yields 576 observations. The four states and the nine test points provide a wide range of characteristics in regard to population size, growth, and age-composition, factors that affect forecast accuracy. The tests reveal that the 66% intervals contain the census age-groups in 397 of the 576 observations (69 percent). We discuss the results, including intervals by age group, and make some observations regarding the limitations of our study. We conclude that the results are encouraging, however, and offer suggestions for further work.

  See paper

Presented in Session 141: Cohort-Component Forecasts…without the Components