Current Population Survey (CPS)

The Current Population Survey (CPS) is a nationally representative large-sample survey of households in the United States, conducted by the U.S. Census Bureau and cosponsored by the Bureau of Labor Statistics. The survey's chief purpose is to provide monthly labor force data, including estimates of employment and unemployment. The survey is also a rich source of data widely used by social scientists seeking descriptive population statistics about the United States. The CPS [Page 174]consists of a core monthly survey and special topic supplements. Each month's core survey includes demographic and employment questions. Periodic supplements cover a variety of additional topics including income, poverty, and health insurance (each March), school enrollment (each October), voting and voter registration (in November of even-numbered years), tobacco use, computer and Internet use, occupational mobility and job tenure, and other topics. Many survey methodologists and statisticians rely upon the CPS estimates as a benchmark to test the accuracy of other surveys and as a source of population statistics that form the basis for survey weights.

The CPS originated as the Sample Survey of Unemployment, administered by the Work Projects Administration in 1940. Responsibility for the survey was transferred to the Census Bureau in 1942, and revisions over the following years led the CPS to assume many of its current characteristics during the 1950s. A decades-long span of comparable measurements is available for many key operational measures. However, substantial changes were made to the CPS in 1994, including the introduction of computer-aided personal interviewing (CAPI) and computer-aided telephone interviewing (CATI) techniques.

The CPS sample consists of approximately 60,000 households each month. The survey respondent, or “reference person,” provides information about each household member. Households remain in the sample for a period of 16 months and are surveyed during the first 4 months and the last 4 months of this period, with an 8-month intervening period during which they are not interviewed. One eighth of the sample is replaced with fresh sample each month, so during any given month's survey, one eighth of the sample is being interviewed for the first time, one eighth for the second time, and so on. This sample design is intended to promote continuity in month-to-month and year-to-year comparisons of estimates. In 2 consecutive months, six eighths of the sample is the same. In the same month in 2 consecutive years, half of the sample is the same. The first and last interviews are usually conducted by CAPI, and most intervening interviews are conducted by CATI.

Data collection takes place during the week containing the 19th day of the month, and questions refer to the week containing the 12th day of the month.

Response rates on the Current Population Survey have been very high. The unweighted response rate for the core monthly survey has been 90 to 93% in recent years. Response rates on the supplements are typically above 90% of those who completed the basic monthly survey, or 80 to 90% overall.

Like nearly all sample surveys of the general population, the CPS uses complex sampling procedures rather than simple random sampling. In the CPS sampling procedure, the United States is first divided geographically into approximately 2,000 primary sampling units (PSUs), which are grouped into approximately 800 strata. One PSU is chosen from within each stratum, with a probability proportional to the population of the PSU. This design dramatically reduces the cost of data collection, particularly by limiting the areas within which interviewers must travel. With this design, CPS sampling errors are somewhat larger than they would be under the impractical alternative of simple random sampling. This means that the classical approaches to hypothesis testing and the estimation of sampling error and confidence intervals (which assume simple random sampling) are not appropriate for CPS data, as these procedures would generally overstate the precision of the estimates and lead researchers to erroneously conclude that the difference between two estimates is statistically significant when it is not.

...