Population projection accuracy: The impacts of sociodemographics, accessibility, land use, and neighbour characteristics

Abstract

Population projection is essential to governments, businesses, and research communities for many purposes. Although projection performance is often evaluated, we know very little about what factors affect projection accuracy. It is important to understand these factors in order to utilize the projections knowledgeably. This study fills this gap in the literature by comprehensively investigating the possible factors associated with population projection accuracy in 2010 for the continental US counties. The results indicate that the counties whose populations are more predictable tend to be desirable places-places with abundant employment opportunities, reliable public transportation infrastructure, easy access to work, and/or high land development potential; their neighboring counties tend to have a well-educated population and a higher income level. Also, projection accuracy is highly spatially associated. The findings provide important insights for population projection users to understand the characteristics of counties and their neighboring counties associated with their projection accuracy.

Keywords: bias; driving factors; neighboring counties’ characteristics; population projection; precision; projection accuracy.

Fig. 1. Distributions of (a) absolute percentage errors and (b) percentage errors

Notes: The values of APEs are non-negative. Large values of APEs indicate lower precision. A positive PE indicates that the 2010 population has been overprojected; a negative PE indicates that the 2010 population has been underprojected.

Fig. 2. Local clusters of spatial association of (a) absolute percentage errors and (b) percentage errors

Note: Low-low clusters of APEs highlight counties with low APEs that are surrounded by counties with low APEs; the projections for these counties have high precision. High-high clusters of APEs highlight counties with high APEs that are surrounded by counties with high APEs; the projections for these counties have low precision. Low-low clusters of PEs highlight counties with low PEs that are surrounded by counties with low PEs; these counties are underprojected and have high downward projection bias. High-high clusters of PEs highlight counties with high PEs that are surrounded by counties with high PEs; these counties are overprojected and have high upward projection bias. Low-high outliers indicate counties with low APEs (or PEs) that are surrounded by counties with high APEs (or PEs). High-low outliers indicate counties with high APEs (or PEs) that are surrounded by counties with low APEs (or PEs).