A Jobless Recovery with More People Working?
As determined by the National Bureau of Economic Research, the most recent recession ended in November 2001. By November 2003, however, the number of nonagricultural jobs- payroll employment-had fallen by about 809,000, according to the Bureau of Labor Statistics (BLS). On the other hand, also according to the BLS, the total number of people working-household employment-had risen by more than 2.3 million over the same period.
How can the two measures of employment indicate such wildly different pictures of the labor market? If the number of people working has been rising, can we really say that it has been a "jobless recovery"? To address these questions, one needs to understand how the different types of labor-market data are collected and what each of the two employment measures is designed to convey.
Under the Hood
Every month, in its Employment Situation report, the BLS announces the results of two surveys designed to measure the state of the U.S. labor market. The first of these, the Current Population Survey (CPS), is based on a sample of about 60,000 households and is used to construct data on the labor-market status of individuals, i.e., whether they are employed, unemployed or out of the labor force. The second survey, the Current Employment Statistics (CES) survey, is based on the payroll reports of a sample of more than 390,000 establishments employing nearly 50 million nonfarm wage and salary workers. The most important data based on the CES survey are the number of wage and salary employees, average hours and average earnings.
The major difference between the two surveys is that the household survey covers more employment categories than the payroll survey does. These categories include the self-employed, farm workers, private-household workers and unpaid workers in family-operated businesses. Another difference is in the ways that the surveys handle multiple-job holders. Because the household survey simply notes whether someone is employed or not, a person with two jobs will be counted just once. On the other hand, because the payroll survey counts jobs, both of this person's jobs will be counted. A third difference between the surveys is in how they handle unpaid absences from jobs. In the household survey, a person with a job but who is temporarily absent without pay-because of illness, vacation, strike, etc.-is counted as employed. In the payroll survey, though, this person is not counted as an employee.
The table presents some numbers that try to reconcile some of the differences between the two employment series. The first step is to broaden the payroll employment series to include job categories that are covered by the household survey but not by the payroll survey. Specifically, add the 874,000 increase in the total number of workers in four household survey categories: agricultural employment, nonagricultural self-employment, nonagricultural unpaid family work and private-household work.
The next step is to account for the 129,000 decrease in the number of workers who were on unpaid absences from their jobs and were, therefore, not counted as being on an establishment's payroll despite having a job. Finally, to remove the double counting of people from the payroll employment number, subtract the 137,000 increase in the number of multiple-job holders.
Although these adjustments add just over 600,000 to the change in payroll employment, the result is still a net job loss of 201,000 between the end of the recession and November 2003. More glaring, though, is that these adjustments make only a dent in the discrepancy between the payroll and household employment numbers. What had been a discrepancy of more than 3.1 million employees before the adjustments is still a discrepancy of more than 2.5 million after the adjustments. In other words, after adjusting the payroll employment number to make it more compatible with the household measure, only about 19 percent of the discrepancy between the changes in the two employment measures is closed.
Scaling by Population
One explanation for the remaining discrepancy arises from the ways in which the survey results are scaled up to represent the experience of the entire population. For example, the household survey reveals the employment situation of the 60,000 or so households that were surveyed. The BLS then extrapolates from these households the employment situation across the more than 100 million households in the country. To do this, the BLS needs to have estimates of the size of the relevant subset of the population, namely, the civilian noninstitutional population that is 16 or older. If this population control overstates actual population, then employment will be overstated, and vice versa.
In the late 1990s, the discrepancy between the payroll and household employment series was the opposite of what has occurred more recently. Then, people were concerned with the rapid growth of payroll employment relative to household employment. In a paper published by the New York Fed, economists Chinhui Juhn and Simon Potter argued that the widening gap between the payroll and household surveys in the 1990s was probably due to an underestimate of the working-age population. This assessment appears to have been correct: Since then, two sizable upward revisions to the population controls narrowed the gap significantly.
A second paper, presented to the Federal Economic Statistics Advisory Committee in October, provides evidence that the recent widening of the gap between the two series is probably due to overestimates of population growth.1 The authors argued that the divergence between the two series in the 1990s arose, in part, because the strong economy had attracted an influx of illegal immigration. But economic growth during the recovery and expansion has been weaker than normal. Accordingly, the recent discrepancy in the employment series might reflect the opposite bias in the population controls. In essence, current population controls assume a rate of growth in illegal immigration that was consistent with the growth of the economy in the period 1994 to 2000, but that is inconsistent with the more recent slow-growth period.
In fact, the household employment numbers for January 2004, which are part of the Employment Situation that was released Feb. 6, use revised estimates of the civilian noninstitutional population. Although the BLS did not revise official estimates for previous periods, it did provide unofficial revisions for the December 2003 data, which suggest that the population control had been overstated. If this were adjusted for, household employment for December would be reduced by 409,000. For the sake of simplicity, assume that this is also the overestimate of November's household employment. The change in household employment reported in the table would then be reduced by 409,000, as would the discrepancy after the adjustments made in the table.
Because payroll employment is derived from a sample of nonfarm establishments in the economy, it is also subject to substantial revisions over time. In the present context, the concern is that current methods do not keep up with the relatively rapid growth of establishments during recovery periods. If so, then the scaling up from the sample will underestimate the actual number of employees. This was a problem during the recovery following the 1990-91 recession, during which payroll employment data throughout 1992 were indicating a net loss of jobs. Following frequent revisions, however, payroll data now show that about 1.4 million jobs were added during 1992.2 Unfortunately, it may take some time before we can assess the steps that the BLS has taken since then to correct the undercounting of firms during recovery.
Are We Any Closer?
By reconciling the coverage of the payroll and household employment series, we were able to explain about 19 percent of the discrepancy in the change in employment in the two years following the end of the recession. We explained an additional 13 percent or so of the discrepancy by approximating the revision to the population controls for the household series. Two possible explanations for the remaining discrepancy are: (1) continuing overestimates of the population controls, which might be revised again in the future; and (2) an undercounting of the growth in the number of new establishments, although the BLS has taken steps to avoid the problems of the past.
Can the Employment Numbers Be Reconciled?
The two official employment series-payroll and household-tell very different stories about the so-called jobless recovery. Some of the discrepancy is due to the differences in the types of jobs covered by the surveys used to create the series. For instance, one series does not consider agricultural workers or the self-employed. Even after adjusting for these and other differences, the discrepancy remains large.
|Change in employment
November 2001 to
|A||Payroll employment (CES survey)||−809|
|B||Household employment (CPS)
|CPS job categories not covered by CES survey
Nonagricultural unpaid family workers
|F||Adjusted payroll employment (A+C+D-E)||−201|
|Discrepancy before adjustments (B-A)||3,139|
|Discrepancy after adjustments (B-F)||2,531|
- Nardone et al. (2003). [back to text]
- This is the change in payroll employment between January 1992 and January 1993. [back to text]
Juhn, Chinhui and Potter, Simon. "Explaining the Recent Divergence in Payroll and Household Employment Growth." Federal Reserve Bank of New York Current Issues in Economics and Finance, Vol. 5, December 1999, pp. 1-6.
Nardone, Thomas; Bowler, Mary; Kropf, Jurgen; Kirkland, Katie; and Wetrogan, Signe. "Examining the Discrepancy in Employment Growth between the CPS and the CES." Paper presented to the Federal Economic Statistics Advisory Committee, Oct. 17, 2003.