BLS Faces a ‘Slow-Moving Train Wreck’ in Collecting Data. There Are No Easy Fixes.

Monthly payroll reports for May and June were revised sharply lower arpbd ryar** ONE-TIME USE *** A restaurant worker puts down napkins at a restaurant in San Francisco, California, US, on Tuesday, March 11, 2025. (David Paul Morris/Bloomberg)

The integrity of the U.S. government’s economic data has come under fire after the Bureau of Labor Statistics revised payroll gains for May and June sharply lower on Aug. 1. Concerns about the ability of U.S. statistical agencies to produce timely, reliable economic data have been mounting for years, and economists say the recent revisions have nothing to do with politics, contrary to comments made by President Donald Trump.

U.S. statistical agencies have grappled with lower survey response rates and budget shortfalls for more than a decade, which has hindered their collection and analysis of economic data. The Covid-19 pandemic laid bare many of these challenges, exacerbating the deterioration of survey participation and requiring agencies such as the Bureau of Labor Statistics, which compiles the monthly jobs report, to rely more frequently on estimates and models. The budget and personnel cutbacks at federal agencies this year have further added to the strain.

There is no quick or obvious fix for these longstanding problems, but nor do most economists consider the data unreliable or otherwise suspect. Still, Erica Groshen, senior economics advisor at Cornell University and a former BLS Commissioner, calls the challenges facing the Bureau a “slow-moving train wreck.” At some point, she says, “there will be a problem” with worsening U.S. economic data quality if statistical agencies aren’t empowered to develop new ways to collect and analyze data.

The issue of data quality became urgent last Friday when the BLS reported a lower-than-expected payroll gain of 73,000 in July, and revised downward by a combined 258,000 previously reported job gains for May and June.

“Today’s Jobs Numbers were RIGGED in order to make the Republicans, and ME, look bad,” Trump wrote on social media, in a post in which he also announced the firing of BLS head Erika McEntarfer, a labor economist. “Important numbers like this must be fair and accurate, they can’t be manipulated for political purposes,” Trump added, without providing evidence of manipulation.

Government economic data are a cornerstone of Social Security benefits and many pension payments, as annual payment increases are indexed to the consumer price index. Economic indicators also play a key role in the market’s ability to price bonds. If there is increased uncertainty about the accuracy of inflation or employment data, that will translate into higher nominal bond yields, says Rodney Ramcharan, professor of finance and business economics at the University of Southern California Marshall School of Business.

Fewer Survey Respondents

Much of the economic data generated in the U.S. is based, in part, on surveys of businesses and consumers. The BLS draws on the “establishment” survey of 121,000 businesses and government agencies, for example, to calculate the monthly change in payrolls.

Estimates are revised twice after the initial report, as additional responses and information become available. Later the payroll estimates undergo an annual benchmarking process to align them with more comprehensive data from administrative sources, including unemployment insurance systems.

The Bureau of Economic Analysis similarly relies on a series of three estimates to calculate quarterly growth in gross domestic product.

The quality of the estimates depends in both cases on the collection of responses. In recent years, however, businesses’ participation has lagged. The initial response rate to the July 2025 payroll survey was approximately 58%, versus an average rate of around 72% to the July survey over the past decade. Generally, by the final release, the collection rate tops 90%.

The diminishing sample size used to create the initial estimate typically means that subsequent estimates will be revised. The subtraction of 120,000 jobs from May’s initial payroll report reflected the largest change between the first and third estimate, excluding the pandemic years, since January 2009.

McEntarfer said on July 14 at a seminar in Washington, D.C., that while the BLS has seen a decline in response rates, “pretty much all of our investigation of data is that it’s not impacting any of the integrity of the published statistics. They’re still accurate.”

Larger revisions to the monthly payroll estimates tend to occur when the U.S. is close to a turning point in the economic cycle, says Gregory Daco, chief economist at EY-Parthenon. “Whether it’s an expansion or a potential downturn or recession, the revisions tend to be larger because the economy is changing direction,” he says.

Staffing and Budget Shortfalls

Beyond declining response rates, the BLS has faced staffing and budget shortfalls that have limited its ability to collect data. The Bureau issued several notices in recent months noting it was suspending collection of data tied to the consumer price index in Lincoln, Neb., Provo, Utah, and Buffalo, N.Y.

In a July 29 notice, the Bureau said it had suspended collection of price data in roughly 15% of the sample in the other 72 areas it surveys, to “align survey workload with resource levels.”

The BLS collects around 76,000 individual prices a month to calculate the CPI. When the bureau can’t track the price of a particular good—it may be out of stock during the reference period or respondents may not have answered—the agency will use a method called cell imputation, which derives an estimate using an average price in the same city or region to fill in the blanks.

Before the Covid-19 pandemic, about 2.5% of the prices reflected in the monthly CPI reading were imputed. That increased to 5.1% in November 2022, before declining to 4.2% in July 2023, the latest measure available.

In its latest notice on CPI data-collection reductions**,** the BLS seemed to indicate that the percentage of imputed price data has now climbed to at least 15%, and likely is as high as 19%, given that data collection was suspended in the three aforementioned cities, says Omair Sharif, founder and president of the research firm Inflation Insights. When the BLS imputes prices for Buffalo, for example, it now must rely on an estimate of the regional average price of an item, he says.

Discontinuing publication of some regional and other statistics diminishes the reliability of the underlying data when sliced by state and city geographies, industries, occupations, and demographics. The heavier reliance on estimates and averages also could create lags in determining when price shifts begin to emerge.

“There’s a much bigger margin of error around the data,” says Sharif. “We can’t really do anything about it, [other than] sort of letting them hire more people.”

McEntarfer said at the Washington, D.C., seminar that BLS budgets have fallen about 16% in the past 15 years, adjusted for inflation. She said the BLS has had two rounds of voluntary resignations.

Potential Fixes

The government’s statistical agencies will likely have to invest in resources and manpower to improve data collection—if they can.

While more economic data is available from private sources than in the past, much of it can’t be adapted wholesale by the statistical agencies. “One should not be under the illusion that the proliferation of private-sector ‘big data’ indicators over the past decade can substitute for high quality federal data,” Michael Feroli, chief U.S. economist at J.P. Morgan, wrote in a recent client note. “In almost all cases these indicators are benchmarked to the federal data, as private sector data are very rarely nationally representative.”

Still, the BLS replaced the collection of new-vehicle prices with transaction data from J.D. Power in April 2022, though it took two years to implement the change, after staff analyzed and tracked the data to ascertain it provided a reliable measure.

To improve data quality in the long run, the agencies will need to develop a modern survey methodology, and blend survey data with administrative data to create composite measures, Groshen says. But this would require much more funding, and more information-sharing across agencies than currently in force, she adds.

The Committee on National Statistics, which is working to develop a cohesive plan for an updated national-data infrastructure, published a report last year that laid out considerations for the use of blended data. Additionally, the American Statistical Association plans to convene a panel of statistics experts to consider what such an infrastructure would look like in practice.

Meanwhile, some agencies are taking smaller steps. The Census Bureau, in collaboration with the BLS, is currently testing the use of online data collection for the current population survey.

“In order to be able to preserve that rich landscape, we have to figure out how we are going to address declining response rates and budgets that are not keeping pace with the cost of data collection—and innovation is really how we’re gonna get there,” McEntarfer said last month.

McEntarfer won’t be around to lead the charge. Substantial challenges—and opportunities—await her successor.

Write to Megan Leonhardt at megan.leonhardt@barrons.com