One of the big pieces of news in the September jobs report released by the Labor Department yesterday was that the Bureau of Labor Statistics' (BLS) preliminary benchmark revisions showed that job loss had been 824,000 greater through March of 2009 than had been previously reported. This is a really big deal. It means that job loss averaged almost 70,000 more than originally reported each month over the year from March 2008 to March 2009. The monthly jobs numbers are based on a survey of 160,000 businesses and governmental agencies. While this is a large sample, it necessarily excludes newly created firms that cannot be included in the sample. This would lead the sample to have a downward bias in measuring job creation. On the other side, there are a large number of firms that go out business each month. Many of them may not feel the need to send back their survey to the BLS as they clean out their office. The non-response by firms shutting down could lead to an upward bias in measuring job growth. To correct for both these problems, the BLS includes an imputation every month for jobs created in new firms and jobs lost in firms that have gone out business. This imputation is based on a "birth/death" model that estimates the number of jobs missed by the survey based on recent economic growth and other variables. While the model produces reasonably reliable estimates in normal times, it has been notoriously bad in missing turning points, underestimating job growth during upturns and overestimating job growth during downturns. Economists who follow the data closely noted that the monthly imputations from this model had been running at about the same level or even slightly higher during this downturn than in the corresponding months of the prior year. (The imputations are not seasonally adjusted so it is necessary to measure January against January, etc.) Since it was absurd to imagine that new firms were creating as many jobs in the six months when the economy was in a free fall (October, 2008-March, 2009) as in the corresponding months of the prior year, it was virtually certain that the imputation was leading to a substantial overstatement of job growth. This suspicion has now been confirmed with the BLS benchmark revision. This revision is based on data from state unemployment insurance records. These data are almost a complete census of payroll employment since more than 99 percent of employees are covered by state unemployment insurance. As a result of the release of preliminary data on this benchmark revision, we now know that the job loss in this downturn has been far more severe than initially reported, unless of course we rely on the Washington Post for our news. (The final data, which will be incorporated into the establishment numbers in the January jobs report, always are close to the preliminary data.)
--Dean Baker
