|
Abstract:
|
For monthly labor force estimates, the U.S. Current Population Survey (CPS) utilizes a composite estimator that reduces the variance of over-the-month change by exploiting correlations in the rotating panel design. The gains in efficiency are substantial among highly correlated estimates, such as total employed persons, but come at the cost of a persistent bias due to unequal compositing coefficients and rotation group effects. In this paper, the complex variance of the CPS composite estimator is expressed as a series of geometric series, and the composite estimator’s bias is then minimized under the theoretical constraint that the variance of the over-the-month change is not increased relative to official estimates.
|
