331 – American Community Survey (ACS) Sample Design and Statistical Issues
Pilot Study on Combining Direct Estimates of Income and Poverty from the American Community Survey with Predictions from a Model
Richard Griffin
U.S. Census Bureau
Small area estimation using models that borrow strength from relationships between variables across geographic areas has become increasingly popular. Typically, these approaches combine direct estimates with model estimates. The American Community Survey (ACS) produces direct five-year estimates at the census tract level for income and poverty. An improvement in the accuracy of these estimates as measured by the estimated sampling error is desired. This pilot study will compare the mean squared error of three potential model-based estimation methods with direct five-year estimates of income and poverty at the tract level. The goal is to make a preliminary assessment on the potential gain in accuracy from using these model-based estimates. For maximum improvement, these models require administrative data correlated with income or poverty. If one of these methods could produce significant improvement in accuracy of the estimates, we want to investigate the development of an application for data users to use publically available administrative data to produce model-based estimates to combine with the published ACS direct estimates in order to improve accuracy.
