Exploring the impact of dasymetric refinement on spatiotemporal small area estimates

Abstract

Comparing demographic small area estimates across multiple time periods is hindered by boundary changes in census enumeration units. Areal interpolation can resolve temporal incompatibilities, but underlying assumptions of uniform population density within units is sometimes flawed and results in distorted estimates. Dasymetric modeling refines spatial precision by limiting areal interpolation to the most likely residential areas. Here, a systematic examination of the impacts of dasymetric refinement on temporal interpolation accuracy compares errors that emerge as a consequence of differing time spans. This paper compares the accuracy of three commonly utilized methods of areal interpolation for temporal analysis of population data over the 1990–2010 decades. It examines whether multi-temporal dasymetric refinement prior to areal interpolation improves the accuracy of small area estimates, comparing two different demographic contexts. Data sets include tract-level demography exhibiting dramatic growth (Las Vegas, Nevada), and relative stability (Pittsburgh, Pennsylvania). Areal interpolation with and without the dasymetric refinement is validated using block level data. The dasymetrically refined target density weighting (TDW) provides the overall best performance for the 2000 source data and the expectation maximization (EM) method gives the overall best performance for the 1990 source data; effects of refinement are more prominent in areas of faster population change.