Introduction

Current plans call for the American Community Survey (ACS) to replace the decennial census sample as the premier source of detailed demographic and socioeconomic information about the nation’s population and housing. In doing so it takes on an incredibly important role. It must produce reasonable, accurate, reliable, and consistent data for every area of the country from the national and state level to the tract and block group level, and do so year after year. By providing data annually, the bar has been raised for the ACS above that ever required for the census sample, which had to produce these data only once every ten years. It is the ACS data for small areas, particularly for areas that are less densely populated, that has been the concern of rural demographers, researchers, planners, and decision-makers who in the past have had to rely on the decennial census sample estimates refreshed only once a decade and reflecting a single point in time. Until the introduction of the ACS, no single current survey was producing estimates of the wide-ranging characteristics covered by the census sample, and none had samples of a size sufficient to provide statistically reliable estimates for areas with populations below 100,000.

The ACS was designed specifically to build and improve upon the solid tradition of both the census sample and the current surveys. This paper looks at the ability of the ACS to produce quality data at the county level by investigating several aspects of standard survey methods and focusing mainly on nonsampling error issues. We have confined this analysis to the 21 ACS test sites with populations of at least 65,000, the sites for which yearly survey estimates equivalent to those derived from the Census 2000 sample have been released. The sites consist of 24 of the 36 counties that have been continuously sampled and surveyed under the ACS systematic design using ACS data collection methods.