The Census Bureau initiated the American Community Survey (ACS) development program in 1994 to develop and test the feasibility of producing long form data on a yearly basis, instead of once a decade. In 2001, the Census Bureau initiated the ACS Research and Evaluation Program to answer questions about the usability and reliability of the ACS estimates. This report is an evaluation from the Research and Evaluation Program. The report compares selected quality measures for the 1999 – 2001 ACS three-year average estimates to selected quality measures for the Census 2000 long form estimates.  

The ACS data reviewed in this report were collected from 1999 to 2001 in 36 counties across the United States. The 36 counties are not a random sample of all counties in the United States. Instead these counties were selected for the ACS demonstration project to represent different types of areas, containing large cities, medium cities, and rural counties. While the estimates examined in this report are three-year average estimates, they are roughly equivalent in terms of their sampling error to the five-year estimates that the ACS will produce when the survey is fully implemented. In order to simulate the sampling error for a five-year average, a greater percentage of the individuals living in these counties in 1999-2001 were included in each year’s sample than will be included when the ACS moves to full implementation. The quality measures examined in this report, therefore, may approximate the quality measures for the five-year averages that can be expected using the fully-implemented sample size.