The annual revision is benchmarked to state tax records. From the BLS:
Establishment survey benchmarking is done on an annual basis to a population derived primarily from the administrative file of employees covered by unemployment insurance (UI). The time required to complete the revision process—from the full collection of the UI population data to publication of the revised industry estimates—is about 10 months. The benchmark adjustment procedure replaces the March sample-based employment estimates with UI-based population counts for March. The benchmark therefore determines the final employment levels ...Using the preliminary benchmark estimate, this means that payroll employment in March 2011 was 192,000 higher than originally estimated. In February 2012, the payroll numbers will be revised up to reflect this estimate. The number is then "wedged back" to the previous revision (March 2010).
Click on graph for larger image.This graph shows the impact of the preliminary benchmark revision on job losses in percentage terms from the start of the employment recession.
The red line on the graph is the current estimate, and the dotted line shows the impact of estimated coming benchmark revision. This puts the current payroll employment about 6.7 million jobs below the pre-recession peak in December 2007. Still very ugly.
For details on the benchmark revision process, see from the BLS: Benchmark Article and annual benchmark revision for the new preliminary estimate.
