Improving the Race Edit in the Consumer Expenditure Survey

• Activity Number: 442
• Type: Contributed
• Date/Time: Tuesday, August 6, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Government Statistics Section
• Author(s): Barry P. Steinberg*+ and Sharon Krieger
• Companies: United States Bureau of Labor Statistics and U.S. Bureau of Labor Statistics
• Keywords: Coverage Rate ; Nonrespondent ; Race ; Imputation

Abstract

In recent years there has been concern about the declining coverage rate of the Black/African American population in the Consumer Expenditure Survey (CE). This demographic group seemed to be undercounted in the survey and the problem was growing over time. Analysis of the data found the biggest source of the undercount was among nonrespondents - people who were unable to be contacted or who refused to participate in the survey. CE has various ways of getting the race of nonrespondents, but as a last resort about 20% of them have to be randomly assigned to a category. Research showed this particular edit process was systematically placing too few nonrespondents into the Black/African American category. This paper describes an improvement CE recently made to the random assignment process. The probabilities used to assign nonrespondents to a race category were based on the racial distribution of respondents, but research showed that respondents and nonrespondents had different distributions. Simply changing the source of data used to generate the racial probabilities from the survey's respondents to the most recent Decennial Census improved the results.