The debate is heating up over which of the government's two employment series tells the right story.
Every month the Bureau of Labor Statistics (BLS) reports on total civilian employment collected by household survey and on payroll jobs collected by employer survey. The payroll count, now about 7 percent smaller than total employment, does not include the non-incorporated self employed, agricultural workers, private household and unpaid family workers, and workers absent without pay.
During the current economic expansion, which started in late 2001, total employment has grown by more than 3.9 million more than payroll jobs. That excess holds even when household employment numbers are adjusted to be consistent with the definition of payroll jobs. Of late, the gap between the two series has been widening. The BLS has yet to pinpoint the sources of the discrepancy.
Last month total employment grew by 437,000 (which passed the BLS test of statistical significance) while payroll jobs rose by only 92,000. In September, total employment grew by 123,000 more than payroll jobs. The BLS also recently reported that early next year the payroll count will be increased by an estimated 810,000 for March 2006, based on the results of its annual benchmark revision to the jobs data. This is an unusually large adjustment, about 70 percent greater than the next largest annual revision in the last decade. (That was in 2000 when payrolls were revised upward by 468,000.) Although the latest benchmark revision will close some of the gap between the total employment and the jobs data, a substantial difference will remain.
The annual benchmark revisions to the jobs data since 1996 have resulted in both upward and downward adjustments to the payroll count. In each of the five years from 1996 to 2000 the job totals were revised upward, and in four of the next five years the data were revised downward. This suggests a pattern.
If there is a systematic bias that causes upward or downward revisions to occur in runs, the latest sizable upward adjustment to the payroll count augers well for another upward revision in the jobs data for March 2007.
As former Chairman of the President's Council of Economic Advisers N. Gregory Mankiw has aptly noted, those who have used the faster-growing household employment data in recent years to assess the economy have often been seen as politically motivated. Perhaps this is not surprising considering the BLS features the payroll data in its press releases and that such high-level interpreters of the economic situation as the Federal Reserve chairman and the CEA have shown preference for the payroll job numbers over the total employment data. The media, naturally, go along.
Now economist George L. Perry enters the fray, performing an in-depth analysis of the two employment series. Mr. Perry, a senior fellow at the Brookings Institution, is a top economic analyst, co-editor of the prestigious Brookings Papers on Economic Activity, a University of Minnesota ex-professor, and a former senior economist on John F. Kennedy's Council of Economic Advisers. It would be hard, to say the least, to accuse him of rightward bias. Mr. Perry is a known straight shooter. His study of the payroll and household employment data, "Gauging Employment: Is the Professional Wisdom Wrong?" appeared in the fall 2005 volume of the Brookings Papers.
Noting that "the payroll measure is widely regarded as the more reliable" of the two employment series, Mr. Perry subjected both data sets to a number of rigorous tests, including examining how well the competing series explained movements in Treasury yields, how well they tracked changes in output and unemployment since 1994, and how the two series performed in multivariable employment models. His results convincingly upheld the validity of the household employment series, with the author concluding there is "no convincing reason to reject one series in favor of the other. ... This suggests the simple rule of thumb that users of the data should average the monthly change in the two."
Mr. Perry constructed and tested a new employment series that averaged the payroll and total employment data and found it tracked output and unemployment better than either series separately. Having found household employment data are not less reliable than the payroll data, Mr. Perry recommended that "the BLS might consider giving the employment change from the household series the same prominence as that from the payroll series in its press releases," a suggestion also made in these pages.
Averaging the two series implies employment in the current economic expansion has grown by nearly 2 million more than the current payroll data indicate. For the last two months, the rise in average employment would be 209,000 for September and 264,000 for October, both strong showings and well above the recorded rise in payroll jobs alone.
Alfred Tella is former Georgetown University research professor of economics.
Email This Page!