The QCEW nonfarm payroll gap has widened enough to deserve more scrutiny than the monthly headline number typically receives. Year-over-year, the U.S. Bureau of Labor Statistics Current Employment Statistics survey shows nonfarm payrolls up 211,000 jobs through March, while the Quarterly Census of Employment and Wages, drawing on mandatory returns from 12.2 million establishments, shows just 86,000. Both figures are a long way below the levels of early 2024, when the equivalent comparison ran above two million, or early 2025, when it still exceeded one million.
The consensus interpretation of recent payrolls data leans bullish: the trend is recovering, and August’s report of 162,000 jobs added was widely read as confirmation. The numbers underneath that read are less comfortable.
Why the QCEW Nonfarm Payroll Gap Matters More Than Usual
The structural difference between the two measures is well understood. The CES surveys roughly 629,000 worksites, with a response rate that had fallen to 43.3 percent as of January 2026, against the QCEW’s near-universal coverage of 12.2 million establishments at approximately 95 percent. Discrepancies between a near-census and a low-response sample are not merely predictable; they are guaranteed. The BLS attempts to correct this through annual benchmark revisions, in which CES is reset to align with the QCEW.
The preliminary benchmark revision for March 2026, released in July, was a reduction of 79,000 total nonfarm jobs, or -0.1 percent. For total private employment specifically, the revision was -178,000, also -0.1 percent. After the very large revisions of preceding years, this looks tame, and parts of the coverage treated it accordingly.
The sector detail, however, is less neutral. According to the BLS preliminary benchmark release, retail trade carried a revision of -154,600 jobs (-1 percent), a meaningful downward correction concentrated in a single sector. Transportation and warehousing moved in the opposite direction, with a positive revision of 135,100 jobs (2 percent). The net aggregate may look modest, but the underlying composition tells a more varied story, with the retail contraction largely offset by transport gains rather than being broadly distributed strength.
Sampling Bias and the Nonresponse Problem
The more awkward question is whether the CES, at 43.3 percent response, is now systematically misrepresenting the labour market rather than merely lagging it. When a firm does not return a survey, the BLS imputes from similar reporters by industry, size and region. That method works if nonrespondents are a random cross-section of the universe. There are reasons to doubt they are.
Smaller, newer firms, and those more reliant on immigrant labour, are plausibly more reluctant to engage with any government form in the current environment. If those firms are disproportionately underrepresented in the CES sample, the survey’s remaining respondents skew towards larger, older, more stable operations, which would have been more resilient through 2025 and 2026. The birth-death model, which adjusts for firm creation and closure, would compound the problem: it would be slow to register the disappearance of firms that stopped reporting, and would miss the formation of new ones that never started.
The author of the original analysis is explicit that this cannot be confirmed yet: ‘I suspect this is happening, but there is no way to prove it, now. Proof comes later.’ The QCEW is the mechanism by which that proof eventually arrives, but it lags by four to seven months from the reference period. With the most recent QCEW data running only through March, the picture for the summer months will not be visible for some time.
Seasonally adjusted QCEW comparisons add another layer of difficulty. The BLS does not publish seasonally adjusted QCEW data, making month-over-month comparisons unreliable. John ‘Jake’ Bush at Piper Sandler has produced seasonally adjusted QCEW series, and analysis using his methodology alongside BLS seasonal factors shows that adjustment factors diverge materially between April and September, meaning the supposedly safe comparison window is narrow and the summer months are precisely where the uncertainty is greatest right now.
The BLS’s own Business Employment Dynamics report, which underpins the birth-death model, provides another eventual checkpoint. The October 28 release will show how accurate the birth-death assumptions were for March. But even that, as the underlying analysis drily notes, ‘is as good as it gets for BED.’ Understanding what actually happened in June will have to wait until December. Between now and then, the BLS continues to publish monthly nonfarm payroll reports built on the same stale model.
The consensus may be overweighting the recent uptrend in payrolls as a genuine inflection. The QCEW nonfarm payroll gap, the collapse in CES response rates, and the sector-level divergence in benchmark revisions collectively suggest the stronger read deserves more scepticism than it is receiving. On 28 October, the birth-death accuracy test for March will offer the next concrete data point.
