Evaluating Tract-Level Intercensal Estimates of Neighborhood Demographics and Socioeconomics for U.S. Counties 2001-2009

Margaret M. Weden, RAND Corporation
Christine Peterson, RAND Corporation
Regina Shih, RAND Corporation

The American Community Survey (ACS) multiyear estimation program has greatly advanced opportunities for U.S. research on small areas such as census tracts. Challenges remain, however, for researchers studying areas smaller than the thresholds for ACS annual estimates. We evaluate intercensal estimates of tract-level demographic and socioeconomic characteristics produced via linear interpolation between the 2000 and 2010 Census and 2005-2009 ACS. Discrepancies between interpolated estimates and comparison estimates from the Population Estimates Program (PEP) and ACS are measured using the mean absolute percentage error (MAPE), mean algebraic percentage error, and percentage difference thresholds. On average 80% of the interpolated estimates of population totals were within +/- 2% of the PEP estimate, MAPE was 1.3%, and mean absolute error in gender and racial/ethnic distributions was less than 0.3 percentage-points. Error compared to the ACS was larger. Findings are discussed in relationship to the differences in estimation methodologies between the PEP and ACS.