Some blame the Public. Trump blames manipulation.
The Unresponsive Public
Wall Street Journal writer Paul Kiernan says An Unresponsive Public Is Undermining Government Economic Data
Jobs revisions that led Trump to fire head of statistics agency are due to falling survey-response rates.
Falling survey participation is an important reason the flagship jobs report released every month by the Bureau of Labor Statistics, part of the Labor Department, has undergone such big revisions recently.
This has rippled into the political sphere. On Aug. 1, President Trump fired BLS Commissioner Erika McEntarfer after a particularly large downward revision to jobs for May and June that owed partly to late responses from survey participants.
The White House and top administration officials increased their attacks on the BLS last week after the agency published an annual revision suggesting the U.S. added 911,000 fewer jobs over the 12 months through March. The BLS blamed the initial overestimate partly on response rates.
One hypothesis is known as survey fatigue: People are being asked to answer too many questionnaires. Jonathan Eggleston, a senior economist at the U.S. Census Bureau, found in a 2024 study that recent participants in that agency’s monthly and annual surveys, which are voluntary, were less likely to answer the 2020 census by mail, phone or online, without a knock on the door.
Another is the rise of cellphones with caller ID. In the days of landlines, people had to pick up the phone to know who was calling. These days, many decline to answer callers they don’t recognize.
BLS seeks to recruit about 70,000 new establishments each year. Since mid-2015, the share that agree to participate has fallen to 35% from 78%.
The share of survey participants that respond in time for the jobs report each month has fallen from an average of 74% from 2010-19 to 57% in August. Many respond by the second or third month, leading to large, and growing, revisions.
There are other hurdles to improving data quality, such as resources. Adjusted for inflation, the BLS’s budget has shrunk 13% since 2016. Trump has requested an 8% reduction in the BLS’s budget and head count for the coming fiscal year. In March, he disbanded a pair of unpaid expert committees that advised the agency on how to optimize its processes.
QCEW, BED, CES (Nonfarm Payrolls)
Response Rate Math
The CES has a response rate of 42 percent. But 42 percent of 631,000 sites surveyed is only 265,000 out of 12.2 million establishments. That’s an overall collection rate of just 2.17 percent.
But is that the problem or just a tiny slice of it?
The BLS surveys surveys suffer from sampling errors, sampling bias, nonresponse errors, and poor birth-death models to estimate jobs it presumes are being created in a healthy economy.
Larger firms are more likely to have someone handle the surveys. Businesses that are struggling don’t have time for survey nonsense.
Are the small-medium-large business sampling size weights correct?
The birth-death model is wrong at economic turns. And we have seen clear birth-death errors
In contrast, the Quarterly Census of Employment and Wages (QCEW) report constitutes 95 percent of the data. But the delay is large.
Why Does it Take Six Months to Accumulate a Quarter?
Timeline Breakdown
- 0 Months (Quarter Ends): The reference quarter (e.g., Q1: January–March) ends.
- 1–2 Months: Data Submission – Employers submit quarterly unemployment insurance tax reports (e.g., state-specific forms like DE 9/DE 9C) to state workforce agencies, typically due 30 days after quarter-end. Late submissions and incomplete data extend this period.
- 2–3 Months: Aggregation and Initial Processing – State agencies compile data and send it to the BLS. BLS integrates state data with federal employee records (UCFE), standardizing formats and performing initial quality checks.
- 3–5 Months: Review, Editing, and Validation – BLS conducts detailed data cleaning, corrects errors (e.g., mismatched employment counts), imputes missing data, and cross-validates with other sources for accuracy across national, state, county, and industry levels (6-digit NAICS codes).
- 4–5 Months (Ongoing): Confidentiality and Suppression – BLS applies statistical methods to suppress ~60% of county-level private-sector data to protect employer privacy, requiring legal and statistical reviews.
- 5–6 Months: Final Tabulation and Release – Data is aggregated, benchmarked (e.g., to revise CES estimates), and formatted for public release via press releases, databases, and files. For example, Q1 2025 data was released in September 2025.
For context, Q1 2025 data was released in September 2025.
Timely Garbage
The monthly response rate for the Job Openings and Labor Turnover (JOLTS) survey is a mere 35 percent. Does anyone believe those results?
The problem is not response rate speed. Faster garbage is still garbage.
Those following QCEW data knew major revisions were coming. Yet, the spotlight is on monthly reports.
Data Dependent Fed
The allegedly data dependent Fed is dependent on garbage.
This happens in two directions, not just one.
Nonfarm Payrolls Gone Haywire
Following the Covid recession, the BLS consistently underestimated jobs for months. Now, it’s the opposite.
In July of 2021, the BLS reported a year-over-year change of 7.5 million jobs. QCEW reported 8.7 million jobs. The BLS was low by 1.2 million jobs.
For March of 2025, the BLS reports a year-over-year change of 1.79 million jobs. QCEW reports 675,000 jobs. The BLS is now high by 1.115 million jobs.
Birth-Death Model Errors
If the problem is response rates, why is the period from 2022 to 2024 reasonable?
This points to serious flaws in the BLS Birth-Death model (the birth and death of businesses not individuals).
The birth-death model is what the BLS uses to estimate the number of jobs created by the birth and death of businesses.
The Birth-Death model is based on the BED (Business Employment Dynamics), a huge subset of QCEW.
Note the delay in BED is 6-7 months the same a QCEW.
BLS Recognition of Birth-Death Errors
Please consider CES Birth-Death Model Frequently Asked Questions (see question 9)
Since the 2020 benchmark, CES estimates have been subject to persistent and relatively large birth-death forecast errors. To help address these forecasting issues, BLS modified the model-based component of birth-death by incorporating current sample information to inform the forecasts starting with the 2024 benchmark released with the January 2025 Employment Situation. This modification was only applied from April to October 2024, known as the post-benchmark period. November 2024, December 2024, and January 2025, as well as future monthly estimates, use birth-death components calculated without this modification.
BLS releases the Quarterly Census of Employment and Wages (QCEW) monthly employment data on a quarterly basis 6 to 9 months after a given collection month. During current month birth-death processing, the forecasted portion of the birth-death model uses inputs derived from comparing the following:
- the QCEW-based employment data including business openings and closings
- the same QCEW data excluding the birth units and imputing monthly values to the death units
The difference between these two QCEW-based monthly values is the net of births and deaths. Finally, BLS creates an over-the-month change from this series, and those monthly values become the inputs to an ARIMA time series model used to create forecasts up to and including the current month.
Although the QCEW data are not available until 6 to 9 months after a given month’s reference period, the sample-based ratio of employment changes is available as soon as the monthly reference period is over. Thus, the relationship between the WLR and the QCEW-based birth-death inputs can be modeled using historical data and predicted for more recent months using the sample-based ratio. This adjustment makes the total net birth-death forecasts more accurate during both relatively stable time periods and during more volatile shifts in trend.
BLS found the modeling and forecasting to be more accurate when performed directly at the major industry sector level rather than at the detailed industry level and aggregated to total nonfarm. Therefore, forecasts were calculated at this level and raked down to detailed industries proportional to their forecast error variances. The ARIMA-based birth-death forecasts are still calculated at the detailed industry level and serve as the basis for forecasts at that level prior to raking.
This adjustment to the net birth-death forecasts was only applied to the post-benchmark months (April to October 2024). Currently, BLS will not use this method outside of the post-benchmark period as there is not sufficient time during the monthly estimation period to incorporate the additional processing needed to use this method in the current month’s estimates. However, BLS plans to continue research on the net birth-death model, including alternative models, in hopes of overcoming the operational limitations of using this adjustment in real-time estimation. However, the expected benefits in accuracy for the post-benchmark period were substantial and warranted using this approach during that period.
Three Issue Synopsis
- The monthly jobs reports are heavily influenced by QCEW and BED which is on a 6-9 month lag.
- The small sample sizes of the nonfarm payroll surveys exaggerate sampling errors, sampling bias, and nonresponse errors.
- At economic turns, the Birth-Death model is known to be flawed.
Better response rates will not fix these issues. But hey, let’s blame the public.
Meanwhile, Trump claims the BLS is doing all of this on purpose.
Related Posts
August 2, 2025: Did Trump Fire the BLS Head for Cause, Being the Messenger, or Something Else?
A case can be made for all three. But there’s a clear winner.
September 3, 2025: The Unemployment Level Is Now Greater than Job Openings
For the first time since the pandemic unemployment is above openings.
The nonfarm payroll response rate is 42.6 percent with the same issues as with JOLTS
September 4, 2025: Year-Over-Year Small Business Employment Growth Barely Above Zero
ADP reports the total YOY small business growth as +19,000.
September 5, 2025: Jobs Report Misery: Only 22,000 Gain in August, June Revised to -13,000
August was a bad month for job seekers. Here are the grim details.
September 5, 2025: Since January 2023, BLS Jobs Revisions Were Negative 24 Out of 31 Times
Witness negative revisions 77 percent of the time, with more coming.
September 9, 2025: New QCEW Data Indicates More Big Negative Revisions Coming to Job Reports
The discrepancy between jobs reports and quarterly data widens again.
