Incorporating Spatial Effects into Hamilton-Perry Based Population Projections Improves Accuracy and Precision at the Census Tract Level

Jack Baker, University of New Mexico
Xiaomin Ruan, University of New Mexico
Adelamar Alcantara, University of New Mexico

The Hamilton-Perry method is a viable alternative for projecting population with limited data, making it attractive to applied demographers attempting to model small areas such as census tracts. Previous research has suggested that the method performs acceptably well at the tract level; however, given the inherent spatial dependency in small area datasets it stands to reason that incorporating spatial effects into the model should improve its performance. This paper presents a spatially-weighted variant of the HP method and evaluates its performance in projecting 1990 census tract populations to 2000 in terms of accuracy and bias (MAPE and MALPE). In comparison to the usual aspatial version of the model, the results of this research strongly suggest that incorporating spatial effects can reduce bias and improve accuracy substantially. The details of the method are presented and its implications for applied demography are reviewed.