331 – American Community Survey (ACS) Sample Design and Statistical Issues
Results of Monitoring the American Community Survey Using Statistical Process Control Methodologies
Steven Hefter
U.S. Census Bureau and Federal Housing Finance Agency
Jessica Schwan
U.S. Census Bureau
To detect potential errors in response data from the American Community Survey (ACS) in near real-time, we have developed an automated statistical process control (SPC) system. For each year from 2005 to 2010, the ACS collected data about roughly two million housing units, 4.5 million people in the household population, and 150,000 people in group quarters facilities. Beginning in June of 2011, the ACS housing unit sample was increased by 22 percent. We expect to see approximately the same increase in the number of interviews. Several SPC methodologies are being used to investigate responses, using traditional Shewhart charts as well as basic statistical tests for differences between proportions. We are also using these techniques to monitor any impact the realignment of our Regional Office (RO) structure may have on data quality. This paper describes the details of the methodology, and a summary of results to date including preliminary results from the assessment of the realignment of the RO structure. We also discuss the inherent challenges and obstacles faced when applying traditional process control methods to a large-scale, multi-mode, demographic survey.