Jobs and data: Here’s where the two meet
October 07, 2025
We had a rare event last week. On jobs Friday, the Bureau of Labor Statistics released no data because of the government shutdown.
With no news, market watchers and the media had time on their hands, so we spent the day fielding questions about ADP’s monthly National Employment Report.
Here’s a roundup of the questions we got and how we answered them. But first, a quick primer.
The ADP National Employment report
The ADP National Employment Report is a monthly, independent, high-frequency measure of the private-sector labor market. It’s based on the anonymized payroll data of more than 26 million U.S. employees, or 1 in 6 workers in the country.
We retooled the report’s methodology in 2022 in partnership with the Stanford Digital Economy Lab to deliver more robust and granular insights.
Last week, the National Employment Report showed a decline of 32,000 U.S. jobs in September. The goods sector lost 3,000 jobs, and services shed 28,000 jobs.
Net hiring was concentrated in large employers, those with payrolls of more than 500 people. These establishments added 33,000 jobs while small and medium-sized employers shed 40,000 and 20,000 jobs, respectively.
But there was a caveat. In September, we do a preliminary statistical rebenchmarking of the employment report, which leads to the first question.
Question 1
What is a benchmark?
Every year at this time, we align estimates in the ADP employment sample to a comprehensive population benchmark called the Quarterly Census of Employment and Wages.
The QCEW is derived from tax records filed by businesses as required by state unemployment insurance laws, in cooperation with the Bureau of Labor Statistics. This census provides the BLS and the public with the closest approximation of the size of U.S. employment.
The QCEW is made available to the public only after a five-month delay. To solve for this time lag during the year, we calculate comprehensive and representative estimates of current-month changes in QCEW employment.
Last week, we took the first step to anchor ADP data to the latest available QCEW release, which supplied data for March 2025 through March 2024 and was made public by the Bureau of Labor Statistics in September.
This first-quarter QCEW release is more granular than releases made in the other three quarters of the year. It allows us to analyze data by employer size, industry, and geography. For this reason, monthly job reports from ADP and the BLS are aligned annually to the first-quarter benchmark each year.
Question 2
If the latest QCEW data covered only March 2024 to March 2025, why did ADP’s September number change?
Great question! The National Employment Report is weighted to the population of the QCEW for the benchmark period as described above. But the QCEW has a second purpose in ADP’s methodology. We use it to inform the growth rates of the ADP client sample.
ADP data comes from businesses that have chosen to outsource payroll and human resources activities to our company. This self-selecting group tends to be made up of employers that have faster growth rates than the overall population of businesses.
We use the QCEW to tap down the bias in growth rates in the ADP sample of establishments. The National Employment Report’s lead architect, Tim Decker, explains the process here in detail.
Question 3
It appears the rebenchmarking affected only August and September. Why was there no change to prior months?
This is a preliminary mark that we announced in February 2025. We will provide the final re-benchmark for full-year 2025 in February 2026, with the January release.
Remember that both the pre- and post-benchmarking year-to-date series show a slowdown in hiring momentum.
Question 4
You report seasonally adjusted numbers for ease of month-over-month comparison. But the seasonal adjustment in September 2025 was lower than it was last year. Why?
We use weekly data and calculate weekly seasonality. But weeks don’t exactly line up with years. The seasonality of the 40th week in one year isn’t the same as the seasonality of the 40th week in the next year, because the timing of the weeks in the year depends on the day of the week the year begins.
And because we use reference weeks for our monthly values—that is, our monthly jobs number is actually the weekly value within the month that contains the 12th day of that month—our seasonal adjustment is also sensitive to the position of the reference week for a particular month, which also can change year to year. The BLS uses a similar method and also uses the week containing the 12th day for calculating its non-farm payrolls report.
In addition, we use an adaptive method to calculate seasonality, leveraging work in the 2007 paper “Seasonal adjustment of weekly time series with application to unemployment insurance claims and steel production” by William P. Cleveland and Stuart Scott. This allows the fitting to evolve over time in response to changing seasonal effects. It also better captures the complexity of weekly seasonality.
Question 5
Why did pay growth slow for job switchers?
Glad you asked. Using anonymized data, our Pay Insights report uniquely tracks the pay of about 15 million individual workers each month, then, for comparison, matches it to what each of those individuals earned 12 months prior.
For job-stayers, those workers employed by the same business in both September 2025 and September 2024, year-over-year pay growth was 4.5 percent. That pace of growth has been largely stable for the past six months.
However, for job-changers, those workers who changed employers in the past year, we saw a different pattern. Year-over-year pay growth slowed to 6.6 percent in September from 7.1 percent in August.
We’ve seen a similar seasonal drop in job-changer pay growth from August to September ever since the pandemic recovery three years ago.
This pattern is caused by summer hiring. The August cohort of matched job-changers contains more seasonal summer workers, whereas the September cohort is composed of workers who tend to keep their jobs for the longer term. September ushers in a pivot in the data from seasonal to non-seasonal employment.
The last question
What happens next?
Of all the questions we received last week, this one is the hardest to answer.
Overall, the economy is performing well, primarily because consumer spending is holding up. But by almost every employment measure available, hiring has slowed considerably since the beginning of the year. The last release of BLS employment data before the government shutdown was the Job Opening and Labor Turnover Survey for August, which showed little indication of a pickup in hiring momentum despite strong second-quarter economic growth.
The economy seems to have all the key ingredients to produce solid growth, and it could be poised to deliver stronger job gains in the coming months. But the government shutdown has deprived employers of a valuable tool for making hiring decisions: Certainty. And that missing ingredient could affect what happens next.
The week ahead
This week was already going to be light on releases. Given the shutdown, we might not see any government data at all.
Thursday: The Department of Labor last week paused publication of initial jobless claims, a proxy for layoffs. If the data is released, I’ll be watching whether their historically low levels have continued into October.
Friday: The University of Michigan gives us its first read of consumer sentiment for October. Last month showed a sentiment divide between people who own stock and those who don’t. The former feel more positive about the economy amid record-high equity prices.
