Each month, the Bureau of Labor Statistics (BLS) publishes labor force estimates for the U.S. resident population and a variety of its demographic subgroups, e.g., teenagers, Hispanics. Published figures include estimated numbers of persons employed, unemployed, and not in the labor force, as well as relevant rates such as unemployment rates. These statistics are computed using data from the Current Population Survey (CPS), a monthly household survey the Census Bureau conducts for the BLS.
The CPS sample is a two-stage probability sample of housing units, covering the entire U.S. Each new sample unit remains in the sample for four months, leaves the sample for eight months, and then re-enters for another four months. One quarter of the sample is new (or re-entering) each month, while half of each month's sample comes from the sample for the same calendar month one year earlier. This "four-eight-four" sample rotation scheme results in positive correlation between CPS estimates from different months, improving measures of change over time. The positive correlation is further increased by composite estimation. Composite estimation is the last in a series of estimation steps performed on CPS data, prior to seasonal adjustment. Unlike weighting techniques, composite estimation does not affect CPS micro data; composite estimates are computed using estimated totals from the various rotation groups—groups of respondents who enter the sample together. Since the composite estimates incorporate information from several months' data, users cannot compute composite estimates from only one month's micro data.
In this paper, we present a method of computing composite weights for the CPS micro data weights that incorporate the effect of composite estimation. Data users would compute composite estimates by simply adding these weights, using only one month's CPS data. This method, suggested by Fuller (1990), also allows us to tailor the composite estimator by varying coefficients to the correlation structures of major labor force categories, thus improving reliability. Section 2 provides a brief overview of current CPS estimation procedures, including composite estimation. In Section 3, we describe the process of selecting compositing coefficients for different labor force categories. Section 4 contains results of an empirical study of two variants of Fuller's composite weighting method, as applied to CPS data.