500 – Evaluation and Research on the Current Population Survey
The Impact of Different Rotation Patterns on the Composite Estimator
Yang Cheng
U.S. Census Bureau
Khandaker Mansur
U.S. Census Bureau
One of key elements in the repeated sample design is the rotation pattern. High sampling overlap between periods reduces the sampling variance of estimates of period change and the degree of sample overlap between any two periods is determined by the rotation pattern design. Composite estimation is used in the rotation pattern design to reduce the variance of estimators. The aim of the paper is to determine an optimal rotation pattern by comparing the effect of different rotation patterns in terms of the variances, biases and mean square errors used as the performance criteria of monthly, month-to-month and year-to-year change of composite estimators. The analysis based on the CPS data on different rotation patterns indicates that the variances, biases and mean square errors of estimates for all levels under the 6-6-6 rotation pattern are less than those estimates under all other rotation patterns. These differences compared to those under the 4-8-4 are so small that they do not strongly support that the rotation pattern 6-6-6 is better than the current 4-8-4 rotation pattern. This leads that 4-8-4 rotation pattern is preferable and can be retained.