Abstract
Utility has recently been proposed as a criterion for evaluating forecasts because of problems that arise from evaluations that focus exclusively on accuracy, which is the normative expectation for evaluating forecasts. Utility is used to evaluate the lagged ratio-correlation method as a short-term projection tool. The utility of the lagged ratio-correlation is compared with two alternatives, exponential extrapolation and the Cohort-Component Method, using data from a case study for Washington State counties at three points in time, 1970, 1980 and 1990. Together, these three projection methods represent the range of current practice in regard to resource requirements and accuracy. Each method's utility is measured using Proportionate Reduction In Error (PRE) techniques. These show the reduction in average error across counties that occurs by using each projection method instead of the previous census as a forecast of county population, which is virtually a no-cost forecasting tool. Reductions in error are examined relative to the higher “cost” associated with each of the three projection methods. The evaluation suggests that the lagged ratio-correlation method consistently has a high level of utility for all three timepoints. It achieves reductions in error that are comparable to those achieved by the Cohort-Component Method, yet with much less resource requirements. The exponential extrapolation method is found to have high and moderate utility in two of the three timepoints.
Get full access to this article
View all access options for this article.