Space - The final frontier - Autocorrelation and small-area income forecasting models

Regional planning agencies often project future income at the neighborhood level to determine future needs for transportation, jobs, social services, and amenities. Despite a decade of methodological advances regarding the importance of spatial autocorrelation in altering or reducing the reliability of regression estimates, few forecasters have tried to include such spatial relationships in their neighborhood-level projections. Part of the reason is that controlling for spatial autocorrelation is complex and can require expensive software. The authors use a free and user-friendly software package to estimate spatial effects in forecasting and then show how to develop and utilize proxy variables that can mitigate such autocorrelation. They illustrate how the failure to include such controls in Southern California could lead to overestimates of income in poor areas and hence a tendency to underserve needy areas.