Empirical Prediction Intervals for County Population Forecasts

Area of Science:

• Demography
• Statistical Modeling
• Geospatial Analysis

Background:

• Population forecasts inherently contain uncertainty, particularly for long-term projections and small or volatile populations.
• Existing methods to address forecast uncertainty include presenting projection ranges or developing statistical prediction intervals.
• Prediction intervals can be derived from stochastic models, empirical analysis of past errors, or a combination of approaches.

Purpose of the Study:

• To develop and evaluate prediction intervals for U.S. county population forecasts based on empirical analysis of historical forecast errors.
• To assess the accuracy of empirically-derived prediction intervals in quantifying forecast uncertainty.
• To provide tools for users of small-area population forecasts to better understand and manage uncertainty.

Main Methods:

• Utilized decennial U.S. census data from 1900 to 2000.
• Applied trend extrapolation techniques to generate county population forecasts.
• Calculated forecast errors by comparing projected populations with subsequent census counts.
• Constructed empirical prediction intervals using the distribution of historical forecast errors.

Main Results:

• Empirically-based prediction intervals demonstrated reasonable accuracy in predicting the precision of population forecasts.
• The developed intervals offered limited information regarding the directional bias (overestimation or underestimation) of the forecasts.
• The study validates the utility of empirical methods for quantifying forecast uncertainty at the county level.

Conclusions:

• Empirically-derived prediction intervals are a valuable tool for assessing the precision of small-area population forecasts.
• These intervals aid users in evaluating forecast uncertainty and improving future planning.
• Further research may be needed to address the directional bias in population forecasts.