An Evaluation of Design-based Properties of Different Composite Estimators
share
For the last decades, the US Census Bureau has been using the AK composite estimation method for generating employment level and rate estimates. In this paper, we devise an evaluation study to compare the AK estimator with different competitors using the Current Population Survey (CPS) data and a sample design that mimics the CPS design. To this end, we first expand the list of potential competitors to the AK estimator by considering the class of linear estimators. The AK estimator is obtained from the first class when the variances and covariances are estimated under rather strong stationary assumptions on the variances and covariances. The AK estimator and other estimators obtained from the optimum estimator in either class when the unknown variances and covariances are substituted by their natural estimators did not produce good results in our evaluation study. In the real data analysis, the AK estimates are constantly below the survey-weighted estimates, indicating potential bias in the estimator. Better estimators of the variances and covariances are needed. We did not entertain this problem in this paper. A different approach is to use a variant of the regression composite estimator used by Statistics Canada. This estimator does not require estimation of variances and covariances of the month-in-sample estimators and is less sensitive to the rotation group bias. Our study demonstrates that there is a great potential for improving the estimation of levels and month to month changes in the unemployment rates by using the regression composite estimator.
READ FULL TEXT
