Confidence Intervals for Population Forecasts: A Case Study of Time Series Models for States

Stanley K. Smith, University of Florida
Jeff Tayman, San Diego Association of Governments (SANDAG)

A substantial amount of research has dealt with the use of time series models to develop confidence intervals for population forecasts. Most studies have focused solely on national-level models and few have considered the accuracy of the resulting forecasts. In this study, we take this research in a new direction by constructing time series models for several states in the United States and evaluating the resulting population forecasts. Using annual population estimates from 1900 to 2002, we develop a variety of forecasts and investigate the impact of differences in model specification, launch year, length of base period, and length of forecast horizon on the accuracy of point forecasts and the width of confidence intervals. We also evaluate the extent to which predicted confidence intervals encompass future population estimates. We conclude with several observations regarding the potential usefulness of time series models for producing state population forecasts.