Improving Models to Estimate Bias in Payroll Employment Estimates

George Stamas

Abstract

The Current Employment Statistics (CES) survey is a monthly payroll survey of more than 380,000 non-agricultural business establishments. It provides estimates of employment, hours, and earnings by industry for the Nation, States and Metropolitan Areas. Employment estimates are revised annually in a process called benchmarking, reconciling estimates to counts of employment available from administrative records of the Unemployment Insurance programs (UI) (BLS, 1988). Revisions made from benchmarking have been large, occasionally, even for total payroll employment at the national level. The objective of this research is to improve CES estimates to reduce benchmark revisions in a manner that can be applied in a production environment.