Dollars and Jens
Friday, September 17, 2004
I noted a while ago that this past weekend would be the snapshot date for the last employment report to be released before the election; that is, the September employment report, to be released in October, is supposed to capture the level of U.S. employment on September 12, which was this past Sunday. Further job gains will presumably do little to help President Bush.
I'm now going to rip off a large chunk of the economic report of the President for 2004 on the difference between the surveys:
Everyone who works is either employed by a firm or is self-employed. Therefore, to count the total number of workers, one could ask each person whether he or she is employed, or one could ask each firm how many workers it employs. The Bureau of Labor Statistics, the agency responsible for tracking employment, uses both approaches. When the BLS asks individuals about their employment status, the results are summarized in the household survey of employment. When the BLS asks firms, it produces the establishment survey of employment.Note that well over 1 million jobs have been created in the past year; thus payroll employment is higher than it was in November of 2001. Regardless of which survey you want to lean on, it seems a bit unfair to blame the President for the heavy job losses that took place in his first ten months in office, at the beginning of which the economy was just leaving what is now widely acknowledged as an unsustainble bubble period. The unemployment rate is lower than it was in 1996, when President Clinton was reelected. It seems a stretch to call even the employment data "weak".
Though both surveys ask about employment, they have some important differences that can cause their results to diverge. For example, the establishment survey obtains data from about 160,000 businesses and government agencies that represent about 400,000 worksites and employ over 40 million workers. The sample covers about one-third of all nonfarm payroll jobs in America. The household survey, in contrast, collects data from about 60,000 households, thereby directly covering fewer than 100,000 workers. The establishment survey’s larger base of respondents means the calculated margin of error of its estimates is significantly smaller than that associated with the household survey estimates. In addition, the establishment survey is revised annually to match complete payroll records from the universe of establishments participating in state unemployment insurance programs, while the household survey is not.
Furthermore, definitional differences affect the scope of employment measured by the surveys. The establishment survey estimate represents the number of payroll jobs, or the number of jobs for which firms pay compensation, while the household survey estimate represents the number of employed persons. Because some people hold more than one job, the total number of payroll jobs can exceed the total number of employed persons. On the other hand, the household survey includes employees working in the agricultural sector, the unincorporated self-employed, unpaid family workers, workers in private households, and workers on unpaid leave from their jobs. The establishment survey excludes all of these categories because they are not reported on the nonfarm business payrolls that provide the source data for the survey.
These differences and other factors create a gap between the household and establishment surveys’ employment estimates, though they tend to display similar long-term trends. The average gap since 1990 has been about 6 percent, or 8 million workers.
While long-term trends in the two surveys are similar, over shorter periods of time their trends have sometimes diverged. This has been the case since late 2001, when employment from the two surveys has trended in opposite directions. For the first time in the two series’ histories, one showed a large and sustained decrease in employment while the other showed a large and sustained increase. In particular, the establishment survey reported a decline in employment of over 1.0 million from the end of the recession in November 2001 to August 2003, while the household survey reported an increase of over 1.4 million. In every month of 2003, the establishment survey showed employment below the November 2001 level, while the household survey showed it above this level. Such a sustained string of divergence is unprecedented.
One possible explanation is that the establishment survey misses some new firms and therefore may underestimate employment at the start of an economic expansion. Past revisions to the establishment survey offer some support for this theory. For the recent data, however, this theory can explain at most the divergence since March 2003, because establishment survey data up to that point appear consistent with unemployment insurance records that cover all establishments. Another possible explanation is that the household survey results are overstated because of the way in which the survey results are extrapolated to represent the entire population. Specifically, information from the 2000 Census, together with estimates of how the population is changing over time, are used to determine how many actual U.S. households correspond to each household in the sample. If, for example, immigration has been unexpectedly low because of tighter border controls and the weaker labor market over the past few years, the estimated number of U.S. households corresponding to each household in the sample may be overstated. As a result, the estimates of total employment (and other aggregates based on the population estimates) from the household survey could be too high.
Both surveys contain valuable information about current economic developments, but, as with all economic statistics, the data from both surveys are imperfect. The Bureau of Labor Statistics has stated that the establishment survey is generally the more reliable indicator of current trends in employment. Still, the explanation for why these two surveys’ results have diverged so markedly over the last few years, and what this might indicate about the economic recovery, remains a puzzle.