A New Approach For Developing Probabilistic Intervals Around Population Forecasts: A Subnational Example Using Washington State Counties

BACKGROUNDPopulation forecasts produced by governments at all levels are used in the public sector, the private sector, and by researchers. They have been primarily produced using deterministic methods.OBJECTIVEWe propose to show how a method for producing measures of uncertainty can be applied to existing subnational population forecasts while meeting several important criteria, including the concept of utility. We will assess the efficacy of our method by: (1) examining the change in uncertainty intervals it produces by population size and growth rate; and (2) comparing the width and temporal change of the uncertainty intervals it produces to the width and temporal change of uncertainty intervals produced by a Bayesian approach.METHODSOur approach follows the logic of the Espenshade-Tayman method for producing confidence intervals in conjunction with ARIMA equations to construct a probabilistic interval around the total populations forecasted from the Cohort Component Method, the typical approach used by demographers.RESULTSPopulation size and growth rate are related to the width of the forecast intervals, with size being the stronger predictor, and the intervals from the proposed method are not dissimilar to those produced by a Bayesian approach.CONCLUSIONSThis approach is well-suited to generating probabilistic subnational population forecasts in the United States and elsewhere where these forecasts are routinely produced. It has a higher level of utility, is simpler, and is more accessible to those tasked with producing measures of uncertainty around population forecasts.