By Mike Horrigan, President, Upjohn Institute
August 13, 2025
The latest Employment Situation report from the Bureau of Labor Statistics showed a slowing labor market with employment growth of only 73,000 jobs in July, which, in the parlance of data geeks, is a statistically insignificant change from June. Even more eye-catching were the revisions to the monthly changes first reported for May and June. May’s initial report of a 144,000 gain in payroll employment was revised down to 19,000. And June’s initial 147,000 now comes in at 14,000. While much attention has focused on the subsequent and unjustified firing of the Bureau of Labor Statistics, or BLS, Commissioner Erika McEntarfer, with a charge that the revisions were manipulated, one should not lose sight of what the revisions are saying about the direction of the U.S. economy. This short piece aims to do exactly that.
But first, why are payroll employment estimates revised? There are two principal reasons. First, the Current Employment Survey of payroll employment has a sample size of 121,000 employers that report employment and wages each month for approximately 631,000 locations and it takes time to get firms to report their data for a given month. BLS allows firms three months to report, with approximately 70% of firms reporting in the first collection month (or first closing), about 90% by the second closing, and about 95% by the third. Revisions incorporate the slower-arriving data to yield more-reliable estimates.
Second, the payroll employment data, like many other series, are seasonally adjusted. In the case of payroll employment, seasonal adjustment removes the influence of predictable seasonal patterns and events to show the true underlying cyclical and trend patterns of employment change. Seasonal events can include changes due to major holidays and school schedules, among others. Think hiring employees in advance of Christmas – was it the normal seasonal bump in hiring or was it unusually large or small this year? So, as they do every month following long-standing procedures, when BLS receives the additional reports from firms in the second and third closing about a prior month’s employment, they redo their calculation of seasonal employment based on the arrival of this new information.
So, we have two separate ways in which the payroll employment numbers for any month can change, the arrival of additional reports over two subsequent data collection periods, and the use of concurrent seasonal adjustment. The question that remains: what factors were in play behind the very large first-to-third closing of payroll numbers for May (144,000 to 19,000) and the similarly large first-to-second closing of payroll numbers for June (147,000 to 14,000)?
A potential answer is that the U.S. economy is beginning a period of much slower and possibly negative employment growth. In the past, when the economy has undergone contractionary shifts in the labor market, we often see large negative revisions in the payroll estimates. For example, in September 2008, at the start of the Great Recession, the first closing showed a decline in payroll employment of 159,000. The second closing came in at -284,000 and the third closing showed a decline of 401,000. In fact, throughout the Great Recession, there were many months in which the subsequent closings showed a much weaker labor market than initially reported. Employers under stress tend to have other priorities beside quickly returning government surveys. When the economy’s weakening, that tends to produce downward revisions.
This is not to say that we are in a recession. However, employment change has been either very slow or negative in most industries since 2023, with only two notable exceptions, Health Care and Social Assistance. The strength we have been observing in payroll numbers has been concentrated in a relatively few industries--not widespread--signaling an overall slowdown as compared to the more robust patterns of industry employment gains in 2021-2022. Therefore, it is not surprising that the additional reports for both May and June showed a consistent pattern of negative revisions across most industries.
It is not clear if the U.S. economy is at a crossroads, and as my time as a BLS staff member taught me, data for one or even two months do not a trend make. It would take several months of data to provide definitive evidence either toward or away from that conclusion. However, in my view, after 33 years of federal service at the BLS, one fact does emerge as unshakeable: the numbers provided by BLS are of high quality and without a trace of political influence.
Could that high quality be improved? Of course, the BLS and all of the statistical agencies have a proud tradition of introducing innovations to their data collection, modeling, and estimation processes. For example, the Current Employment Survey uses electronic data collection that allows large firms with multiple locations to embed automated reporting from their own data systems and allows faster and more accurate reporting of potentially thousands of worksites. At the same time, the statistical agencies have been suffering from declining real budgets for years, which has slowed the ability of agencies to continue introducing innovations to their data collection and other operations.
But most importantly, instead of focusing on politically motivated accusations of data manipulation, now is the time to take a careful look at the direction of the U.S. economy, of our economic policies, and their potential impacts on all of us. For that, we really need credible economic statistics, which only politically independent agencies can produce.