Abstract

The American Community Survey (ACS) began full implementation in 2005. Estimation for ACS, as typical for other large Census Bureau surveys, uses a complex series of ratio estimation and other adjustments to the weights. Working originally from ACS data for 1999- 2001 in 36 test counties, previous research suggested that multiyear tract-level estimates could be improved by imbedding a step of model-assisted estimation, specifically generalized regression estimation (GREG), in the current ACS estimation. In particular, the GREG step incorporates administrative record data. Most data sets produced for the ACS Multiyear Estimates Study incorporate the GREG step, but alternative sets of 3- and 5-year estimates without the GREG step were produced for purposes of comparison. The paper will describe new refinements in the estimation approach and its potential future role in ACS estimation.