Nicholas L. Haltom, Vanessa D. Mitchell, and Ellis W. Tallman
Economic Review, Vol. 90, No. 2, 2005
Download the full text of this article 
During the recovery from the 2001 recession, the business press and economic analysts used payroll employment data released monthly by the U.S. Bureau of Labor Statistics (BLS) as evidence of protracted weakness in the labor market. But using these monthly releases for this type of analysis can be premature and potentially misleading. The initial BLS releases can differ substantially from payroll employment data that are revised to incorporate information from less timely but more complete sources.
This article highlights the historical revisions to the aggregate nonfarm payroll employment series. Examining both monthly survey-based revisions and the more extensive annual benchmark revisions, the authors focus specifically on how the sequence of data revisions modifies payroll employment estimates from their initial release. The graphs in the article display the magnitude and direction of each revision from the initial estimate for a particular month to its currently published value, demonstrating that the largest portion of enduring change for the estimates occurs in the benchmark revisions.
The authors then investigate empirically whether these revisions contain information that can be exploited to anticipate future revisions. The analysis shows that previous benchmark data revisions are useful for explaining the variation in subsequent payroll employment benchmark data. Such information, the authors note, could prove useful for further research aimed at modeling better real-time estimates of employment conditions.
July 2005