Redesign of Current Population Survey Raking to Control Totals

Martha Duff, Edwin L. Robison, Harland Shoemaker, and Brandon Schneider


Weighting for the monthly Current Population Survey (CPS) includes a process called composite weighting. It is found that for highly correlated items, a lower-variance estimate for the current month can be obtained by also using data from previous months, suitably adjusted with an estimate of change. A composite estimate of this type has been in use for the CPS for decades, but a weighting procedure that could reproduce composite estimates from just a single monthly data file was not implemented until 1997. The procedure now in place successively computes ratio adjustments to weights so that estimates are forced to match three sets of composite estimates of employment, unemployment, and not-in-labor-force: 1.) by state, 2.) by ethnicity x gender x age, and 3.) by race x gender x age. A new process to be implemented in January 2003 retains the three-way iteration, but is designed based on a reevaluation of convergence properties and interaction.