Every jobs report contains two employment numbers produced by two different surveys, and they regularly disagree — sometimes by hundreds of thousands of jobs in the same month. This is not an error. It is what happens when you measure the same economy by calling businesses and by calling households, and the gap between them is often the most informative part of the release.
This article is for informational and educational purposes only. It is not financial advice, investment advice, or a recommendation to buy, sell, or hold any security, cryptocurrency, or financial product. Always verify data with official sources before making financial decisions.
Short answer: one report, two surveys
Short answer: The Bureau of Labor Statistics runs two separate monthly surveys. The establishment survey (formally the Current Employment Statistics programme, or CES) calls businesses and produces the nonfarm payrolls headline. The household survey (the Current Population Survey, or CPS) calls households and produces the unemployment rate, the participation rate and its own employment count. They measure different things, use wildly different sample sizes, and are published in the same document — which is why readers assume they should match.
| Establishment survey (CES) | Household survey (CPS) | |
|---|---|---|
| Who is asked | Employers | Households |
| Approximate sample | ~119,000 businesses and agencies | ~60,000 households |
| Headline it produces | Nonfarm payrolls change | Unemployment rate, participation rate |
| Counts a person twice if they hold two jobs | Yes | No |
| Includes the self-employed and farm workers | No | Yes |
| Revised in later months | Yes, routinely | No monthly revision |
Who each survey actually counts
The establishment survey counts jobs on payrolls. If you work two jobs, you appear twice. If you are self-employed, run an unincorporated business, work on a farm or work unpaid in a family business, you do not appear at all. The unit of measurement is a filled position at a surveyed employer.
The household survey counts employed people. One person, one count, regardless of how many jobs they hold. It includes the self-employed, farm labour, domestic workers and unpaid family workers. The unit of measurement is a person aged 16 or over who did any paid work in the reference week.
Once that distinction is clear, most of the apparent contradictions between the two resolve themselves.
Sample size vs coverage: the trade-off nobody mentions
The establishment survey covers roughly 119,000 businesses representing hundreds of thousands of individual worksites — an enormous sample by survey standards. Its monthly sampling error is correspondingly small, which is exactly why markets trade the payrolls number and largely ignore household employment.
The household survey covers about 60,000 households. That is a robust sample for estimating a rate across a population of hundreds of millions, but it is a small sample for estimating a monthly change in a level. The month-to-month swings in household employment routinely run into the hundreds of thousands and are frequently statistical noise rather than signal.
The trade-off is real: the establishment survey buys precision by narrowing what it measures, and the household survey buys breadth by accepting volatility. Neither is the better survey in the abstract. They answer different questions.
Where the two diverge
Four mechanisms generate most of the gap, and each one points in a predictable direction.
- Multiple jobholders. When people take second jobs, payrolls rise faster than household employment. A rising multiple-jobholder rate alongside a widening gap is a recognisable pattern in a softening labour market.
- Self-employment. Growth in gig, contract and unincorporated work shows up in the household survey and is invisible to payrolls. This has been a persistent source of divergence.
- Agricultural and unpaid family work. Excluded from payrolls by definition; included in the household count.
- Population controls. The household survey’s levels are benchmarked to Census population estimates, and when those controls are updated — typically in January — the level can jump without any underlying change in employment. The BLS flags these breaks explicitly.
Which number the unemployment rate comes from
The unemployment rate comes entirely from the household survey. So does the labour force participation rate, the employment-population ratio, and the U-6 underemployment measure.
This is the single most common misreading of the jobs report. When a commentator says payrolls were strong but the unemployment rate rose, they are not describing a paradox — they are describing two independent surveys, one of which said employers added positions and the other of which said more people were looking for work without finding it. Both can be true simultaneously, because the unemployment rate has nothing to do with the payrolls figure.
The rate is a ratio of unemployed people to the labour force, and it can rise for a good reason (people re-entering the workforce and searching) or a bad one (people losing jobs). The participation rate tells you which.
How to read a month where payrolls rise and household employment falls
This happens often. A disciplined reading runs in this order.
Start by checking the revisions to the prior two months of payrolls. The establishment survey is revised twice, and large downward revisions have repeatedly changed the character of a report after the fact — a headline beat sitting on top of heavy prior-month cuts is a weaker report than it first appears. Next, look at the participation rate to interpret the unemployment rate’s direction. Then check the multiple-jobholder rate, which is the cleanest single explanation for a persistent gap between the two surveys.
Finally, resist reading a single month at all. Both series are noisy; the household series especially so. The three-month moving average of payrolls and the trend in the unemployment rate carry far more information than any individual print, and the BLS itself publishes confidence intervals precisely because month-to-month changes frequently fail to reach statistical significance.
Mini glossary: CPS, CES, birth-death model
- CPS (Current Population Survey). The household survey. Produces the unemployment rate. Roughly 60,000 households.
- CES (Current Employment Statistics). The establishment survey. Produces nonfarm payrolls. Roughly 119,000 businesses and agencies.
- Birth-death model. A statistical adjustment inside the establishment survey estimating net job creation from business openings and closings that the sample has not yet captured. It is an estimate, it is published, and it is a recurring source of error at economic turning points — it tends to overstate job creation as a downturn begins and understate it as recovery starts.
- Benchmark revision. An annual reconciliation of the payrolls series against near-universal unemployment insurance tax records. It can revise the level of employment by hundreds of thousands.
- Population controls. Annual updates to the household survey’s population benchmarks, which can shift levels without reflecting any real change in employment.
What this article does not conclude
This is a guide to how the two surveys are built, not a claim that either one is currently telling the truer story. Analysts who prefer the household survey during downturns and the establishment survey during expansions are usually selecting the series that confirms their prior, and the historical record does not support a stable rule about which leads the other.
Both series are published by the BLS with documented methodology, standard errors and revision histories. Where a specific month’s figures matter to a decision, read the release itself rather than a summary of it.
