How the ACS gets it wrong: The story of the American Community Survey and a small, inner city neighborhood

In 2010, the long form Decennial Census was not used for the first time in several decades, replaced instead by the American Community Survey (ACS). While the ACS has been collecting data for more than a decade, the first census tract results were released in 2010. The ACS collects its data based on a rolling sample. It is purported to have many advantages over the Decennial Census, including frequently updated data, less overall cost, and better coverage. We have found, however, that the ACS can produce highly inaccurate data, in part because of reliance on extremely small samples in sparsely populated neighborhoods. We explore these dynamics through a community survey which replicates ACS questions while employing the old Decennial Census long form sampling strategy of requesting information from one in six households in a single inner city census tract in Nashville, Tennessee. Our results show that the ACS has grossly underestimated the total population and the number of people living in poverty in the neighborhood, among other variables of interest. Through an examination of the number of census tracts with very small sample sizes and a comparison of recent ACS results with the 2010 Decennial Census, we show that the problems identified in this single tract occur throughout the country. In addition, we consider how statistical profiles of neighborhoods derived from the ACS and other data sources enable wide ranges of action in those neighborhoods, action which can be harmful given inaccurate and/or decontextualized data.