Academic journal article
By Flaim, Paul O.
Monthly Labor Review , Vol. 112, No. 8
How many new jobs since 1982? Data from two surveys differ
The growth in employment during the expansionary period that began in late 1982 has been extremely robust by any standard. The exact magnitude of the growth, however, depends on the statistical series used to gauge it. As of April 1989, the Bureau of Labor Statistics' survey of employers' payrolls--the Current Employment Statistics program--had shown an increase of about 20 million jobs since November 1982, while the survey of households--the Current Population Survey (CPS) conducted for BLS by the Bureau of the Census--showed an increase of only 18 million in the number of employed persons. There was thus a discrepancy of 2 million between the two surveys.
More recently, the employment figures from the payroll survey have been revised--or "benchmarked"--downward for the period since March 1987, (1) with the level for April 1989 being lowered by more than half a million. This has substantially narrowed the gap in growth estimates between the two employment series. However, for the period from November 1982 to April 1989, the increase in the payroll series still exceeds the growth in total employment, as measured through the household survey, by about 1.4 million.
While much of the divergence between the two series has taken place since mid-1987, their paths had begun to differ noticeably as early as 1984. Such a divergence during expansionary periods is not unprecedented. Even during the expansion of the late 1970's, the payroll survey produced substantially higher estimates of employment growth than did the household survey. Then, as now, the different behavior of the two series was cause for concern among some of the users of these numbers. (2)
Making the data more comparable
It is important to note that the two surveys do not cover quite the same universe. The employer survey counts payroll jobs in the nonfarm sector of the economy, while the household series focuses on employed persons, including those in farm work, private household work, unpaid family work, and self-employment. In addition, the two surveys differ in the way they treat dual jobholders and workers on strike or on other unpaid absences. And there are yet other definitional and methodological differences that may allow the trends in the two series to diverge significantly. (3)
For a clearer comparison of the trends in the two series--given the differences noted above--it is useful to adjust the data from the household survey to make them conform more closely to those from the less comprehensive payroll survey. To do so, we must subtract from the household series those groups of workers not covered by the payroll survey. Table 1 summarizes the changes in the data from the two surveys for the period November 1982 to April 1989 both before and after this type of adjustment. (4)
Surprisingly, the difference between the growth paths of the two series turns out to be even larger when the household data are subjected to this adjustment. While the original estimate from the household series had grown by 1.4 million less than that from the payroll series over the November 1982-April 1989 period, the adjusted series show a bigger and more rapidly expanding growth gap, which is in excess of 2 million for the same period even after the recent downward revision of the payroll data.
As indicated earlier, and as shown in chart 1, the growth disparity between the two jobs series began to develop in mid-1984. During 1985, it averaged about 1 million, but then shrank again, averaging around half a million during 1986 and the first half of 1987. Thereafter, the gap began to widen rapidly, expanding to 2.1 million by April 1989 (and to nearly 2.5 million by May 1989).
Possible reasons for a widening gap
Because the adjustments of the household data outlined above actually pull the paths of the two employment series further apart, we must look for other factors to explain the widening gap, even if we do not have the data with which to quantify their impact. …