The Census Bureau first used statistical sampling methods in the 1937 test survey of unemployment ("Enumerative Check Census"). This test survey not only estimated the scope of unemployment in the United States during the Great Depression, but it also served as a "check" on a larger-scale, voluntary census effort attempting to measure the same thing.
The Census Bureau implemented statistical sampling in a decennial census for the first time in 1940. Sampling made it possible to ask additional detailed questions of the population without unduly increasing cost or respondent burden. Enumerators asked a random sample of the population (approximately 5 percent) a set of extra questions. The Census Bureau then used the sample to extrapolate demographic data for the entire United States.
Sampling became a fixture of the decennial censuses, with a certain percentage of the population being required to fill out longer questionnaires with more detailed questions, through the end of the twentieth century. In fact, because the American Community Survey is now the instrument used to gather long form information, the 2010 census will be the first since 1930 in which no additional questions will be asked of a sample of the population.
The success of statistical sampling in the decennial census contributed to the development of scores of recurring demographic surveys, most notably the Monthly Report on the Labor Force in 1943, expanded and renamed the Current Population Survey in 1947. The Census Bureau now conducts more than 200 economic and demographic surveys every year, using these results to produce national figures.
The Census Bureau also uses sampling and estimation techniques to measure net coverage in the decennial census. The results of a coverage measurement survey are compared to the census results for a sample of blocks nationwide. Comparing the results from these "dual systems" (the measurement survey sample and the enumeration sample), the Census Bureau uses a statistical technique called dual system estimation to estimate net over- and undercoverage in the decennial census. Coverage measurement surveys were conducted in conjunction with the 1980, 1990, and 2000 censuses. In all three censuses, the issue of using the coverage measurement survey results to statistically adjust the census counts to correct for net coverage errors was litigated. This litigation is discussed briefly in the 1980, 1990, and 2000 pages of the "Through the Decades" section.
Additionally, the Census Bureau uses sampling techniques in the planning, development, and implementation of its statistical programs and operations to ensure their quality and efficacy. For instance, most quality assurance operations are sample-based. Sampling techniques are also used in the pre-implementation evaluation of new programs and surveys.
The Census Bureau used statistical sampling in the 1937 Survey of Unemployment.
Photo courtesy of the Social Security Administration.
Information and suggestions for further reading on statistical sampling can be found in A Bibliography of Selected Statistical Methods and Development Related to Census 2000.