One day last month, 818,000 jobs vanished from the U.S. economy. Or did they? It would be more accurate to say that the government revised its official estimate of jobs based on new information. While noteworthy, this preliminary revision shouldn’t undermine its credibility.
What happened is that the Bureau of Labor Statistics released a statement that 818,000 fewer nonfarm payroll jobs (or 0.5% of the total) were likely created from March 2023 to March 2024 than it had previously announced. This led some partisans to charge that the bureau had been “cooking the books” for political purposes.
In fact, revisions are a normal part of obtaining an accurate picture of the nearly $30 trillion U.S. economy with a labor force of about 170 million people. If the U.S. wants timely statistics—and to limit the burden on survey participants—then revisions are a necessary byproduct.
The BLS’s first estimates of monthly payrolls are from a largely voluntary survey of 119,000 businesses and government agencies and come out in just three weeks. This speed comes at a cost. Only about 60% of businesses this year responded in time for the first estimate. The response rate rose to 88% by the third month, but the gap helps explain the early payroll revisions.
In many other countries, as well as the European Union, most labor-force statistics are quarterly, not monthly. Having the statistics early, even if they are preliminary, benefits U.S. decision-makers.
Revisions also reflect a commitment to high-quality statistics. The reason for these latest revisions is because the BLS was able to use tax records from state unemployment insurance agencies to provide a more comprehensive count of employment, rather than relying solely on a survey of businesses. That is far less costly and more practical than running a monthly, mandatory census of all businesses. Blending survey and administrative data is an advantage, not a shortcoming.
That said, revisions do impose certain responsibilities on those using the data. It is easy to see how “datapoint-driven” decisions—an extreme version of “data driven” decisions where the exact number is the focus—could lead policymakers astray. Looking at averages across months, instead of a single month in isolation, can help reduce the dependence on one estimate. It’s also better to focus on the broad contours of the data instead of a specific number. Applying that approach to the monthly payroll estimates—both the latest issue and the revisions of previous ones—shows that the basic story of a cooling labor market is mostly unchanged.
Fed Chair Jerome Powell’s statement last month that the bank does not “seek or welcome further cooling in labor market conditions” fits this approach. It also expands the analysis to more than payroll gains. In characterizing the labor market as “cooling,” Powell also discussed the unemployment rate, layoffs, hiring, job vacancies, quits and wages. Each estimate is subject to revision, but all are unlikely to be off in the same direction.
Another approach is to try to anticipate the revisions using other data. Tomaz Cajner and other Fed economists have shown that data from the payroll processor ADP helped forecast revisions to the official payroll estimates. It’s a helpful reminder that official statistics are not the final word. Drawing on other sources, including from the private sector, can improve overall estimates.
Of course, revisions can be disruptive, and the BLS should work to improve or at least maintain the precision of its initial estimates. One challenge for data quality in general is declining survey response rates. As I wrote last fall, participation in various business and household surveys has fallen dramatically over the past decade.
The BLS highlighted one project to improve participation in its latest budget request: developing a web-based collection tool for the Consumer Price Index housing survey. The survey is the data source for tenants’ rent and owners’ equivalent rent, currently the largest contributor to CPI inflation. Participation can matter: An imputation from a very small number of responses in the New York City area in May likely contributed to a bump in U.S. owners’ equivalent rent inflation. It reversed the next month, but it masked the progress that the Fed was eagerly watching for in shelter inflation.
Data revisions should not raise questions about the integrity of the BLS. Still, BLS made some mistakes—ones that it acknowledges—in communicating the latest revision. Due to a malfunctioning public webpage and internal miscommunication, analysts at a handful of financial services firms received the revision before everyone else by calling BLS. It was the third instance this year of data sharing that did not follow standard procedures.
With a hyper-data-driven Fed, any data from the BLS on employment and inflation are market-moving, and the BLS cannot give any advantage to particular firms or individuals. Answering questions from data users is an important service, but the BLS should share its expertise publicly in the data release or on its website.
The integrity of the BLS is crucial not just to ensure public trust in its data, but in shaping people’s willingness to participate in surveys (which in turn helps improve the quality of its data). At the same time, it’s entirely appropriate, even necessary, to be aware of the limitations of government data. Official statistics are not the truth, but rather our best estimates of the truth — and we should always be working to improve not only how we collect and disseminate them, but also how we use them.
Claudia Sahm is the chief economist at New Century Advisors and a former Federal Reserve economist. She is the creator of the Sahm rule, a recession indicator.
This article was provided by Bloomberg News.