Reading the Jobs Report: Payrolls, Participation and Slack

Two surveys, one labour market, and they disagree

The report is two surveys with different designs, and understanding why they disagree is most of the skill. The establishment survey asks about 120,000 businesses how many people are on their payrolls. It is large, it is anchored by tax records, and its headline is the monthly payroll change — but it counts jobs rather than people, so one person with two jobs is counted twice, and it excludes the self-employed, farm workers and private household employees entirely. The household survey asks about 60,000 households what each adult did last week. It is smaller and noisier, but it counts people, and from its levels come the unemployment rate, the participation rate and the employment-to-population ratio. The unemployment rate is a ratio, not a survey answer: the number of unemployed divided by the labour force, where the labour force is everyone employed or actively looking. That structure produces the confusing move everyone meets eventually — the rate can fall because people left the labour force rather than because they found work, and it can rise in a booming month because discouraged workers re-entered and started looking. A rate that falls while participation falls is not good news, and a rate that rises while employment surges is frequently the best news in the report. So never read the rate without the participation rate and the level of employment beside it. The third thing to understand is that the payroll number is an estimate with a distribution, not a measurement. Each month’s print carries a confidence interval of roughly a hundred thousand jobs either way at the ninety percent level, the first two months are revised twice, and once a year the whole level is re-anchored to administrative tax records in a benchmark revision that has historically moved the level by around half a percent. That is why a market reaction to a 20,000-job miss is a reaction to the surprise rather than to the economy, and why the three-month average exists: it is the shortest window that contains the noise. The same report, read twice — Payrolls: +145,000: within a ±100,000 confidence interval — inside the noise · Prior two months revised +38,000: the revision is over a quarter of the headline ← · Average weekly hours: 34.4 → 34.2: fewer hours per worker, so total hours grew less than headcount · Household survey employment: +61,000: below the payroll gain, so the two surveys are not telling one story Headlines quote the payroll change and the unemployment rate and stop. Those are the two least reliable numbers in the report: one is an estimate with a wide interval, the other a ratio that moves with its denominator. The information is in the cross-check.

Wages, participation, and how much slack is left

Once the surveys are reconciled, the report answers one economic question: how much unused capacity is left in the labour market? Three series do that work. The participation rate — the labour force as a share of the adult population — says how many people are available at all; a participation rate still below its pre-pandemic trend means there are potential workers outside the labour force, which loosens the market even with a low unemployment rate. The unemployment rate that includes discouraged workers and those employed part-time for economic reasons — the broadest measure — says how much hidden slack the headline rate is missing. And the employment-to-population ratio is the level check that does not move with the labour force at all. Wages are the evidence that the slack has actually been used up, and they are also an input to the inflation the reaction function cares about, which makes this the place where the labour report and the CPI report meet. Average hourly earnings are a blunt instrument: the composition of the workforce moves them — a month that adds a lot of low-paid hospitality jobs mechanically lowers the average — so a single print is weak evidence and the three- and six-month annualised rates are better. Watch the direction of the revision and the wage series together: a payroll beat with decelerating wages is a labour market adding workers, while a payroll miss with accelerating wages is a market that is still tight enough to pay up. A simple real-time rule is worth carrying: take the three-month average of the unemployment rate and compare it with its own lowest level over the past twelve months. That gap has never reached half a percentage point in the run-up to a recession without one following, because a labour market deteriorates gradually before it deteriorates abruptly — hours fall, then hiring freezes, then separations begin. The rule is not a forecast, it is an alarm, and its value is that it only fires when the deterioration is already broad rather than in a single noisy month. • Participation below trend means potential workers outside the labour force — that is slack. • The broadest unemployment measure is where the hidden slack shows up first. • Wages are the proof that slack has been used, and an input to the inflation print. • Average hourly earnings move with composition, so use the three- and six-month annualised pace. • A 0.5pp rise in the three-month average unemployment rate from its twelve-month low has never been benign.

The part of payrolls that is a model

The establishment survey does not count every firm; it samples them. Two adjustments turn that sample into a national number, and both are estimates. **Seasonal adjustment** removes the regular pattern — retail hiring before the holidays, education hiring in September — using factors estimated from prior years, and firms whose timing shifts from year to year will be over- or under-counted as a result. **The birth-death model** estimates net job creation from firms that did not exist when the sample was drawn, because those firms cannot be surveyed, and its accuracy depends on new businesses behaving like the ones they are modelled on. The consequence is a first print that is noisy and a revision process that is large. Payroll revisions routinely move the headline by tens of thousands, and the annual benchmark revision — which re-anchors the series to actual tax records — has at times run into the hundreds of thousands of jobs. None of this is manipulation; it is the arithmetic of measuring a moving population with a lagged sample. But it means the market’s reaction to the first number is a reaction to an estimate with a wide error band. How to use it: treat the first print as the direction of travel and the revisions as the level; watch the three-month average rather than any single month; and when the household survey and the establishment survey disagree, take the disagreement seriously rather than picking the one that supports a view. The household survey is smaller and noisier, but the two measure genuinely different things — jobs and workers, including self-employment — and the gap between them is information. A single month of payrolls is a sample. A trend is a story about three to six of them, and the levels themselves are revised for years after the fact.

The indicators beyond the headline

Payrolls are the most watched number in macro and among the most revised, which makes a labour-market read built only on the headline a read of one noisy series. The rest of the report, and the reports that follow it, answer the questions the headline cannot: who is working, how confident they are, and whether the market is cooling through vacancies or through job losses. The household survey supplies the measures that adjust for composition. The **prime-age employment ratio** — the share of people aged roughly twenty-five to fifty-four who are employed — removes most of the demographic noise that makes the participation rate so hard to read, because that group is neither retiring nor in school. The **U-6** measure widens unemployment to include people who want work but have stopped looking and those working part-time involuntarily, and it therefore moves earlier in a deterioration than the headline rate. Both are slower-moving and both are less sensitive to a population estimate that gets revised. Then the job-openings survey, whose structure is worth knowing before its headline is quoted. **Openings** measure demand and are noisy and heavily revised; **quits** measure workers’ confidence, because people leave jobs they believe they can replace; **hires** are the flow that actually employs somebody; and **layoffs** are the slow-moving part that stays quiet until it does not. The relationship between openings and unemployment is the **Beveridge curve**, and its value in the post-pandemic period was precisely that it answered the central question — can the labour market cool by vacancies falling rather than by unemployment rising? A rising unemployment rate with falling openings and stable layoffs is a cooling; a rising rate with rising layoffs is the other thing entirely. Two mechanical series deserve a place in the routine. **Average weekly hours** is a leading indicator in the ordinary sense: employers cut shifts before they cut headcount, so a decline in hours across a few months anticipates a weaker payroll number. And aggregate hours — hours times employment — multiplied by productivity gives the labour input into output growth, which is the bridge from this report to the growth figures discussed earlier in the subject. The practical conclusion is the same one this subject reaches everywhere: use a small composite rather than a favourite series. The prime-age employment ratio for the structural read, the quits rate for confidence, the three-month trend in payrolls for momentum, and the hours series for the earliest warning. And treat a disagreement between the two surveys as information in its own right: the household survey includes the self-employed and the establishment survey does not, so a divergence often means the change is happening in the part of the labour market only one of them can see. • The prime-age employment ratio and U-6 correct for composition and move earlier. • Quits measure confidence, hires measure employment, layoffs are the slow part. • The Beveridge curve answers whether cooling comes from vacancies or from job losses. • Average weekly hours falls before headcount does; a composite beats a favourite series. The one number that has repeatedly anticipated weakness is the change in weekly hours, because a firm that expects to keep its staff cuts their shifts first. It is unglamorous, it is revised less than payrolls, and it is in the same release.

Sources

• The Employment Situation: methodology for both surveysU.S. Bureau of Labor Statistics, “Employment Situation” technical notes and the CES/CPS handbooks
• Revisions to the payroll estimate and their sizeBLS revision policy and the annual benchmark process
• Real-time indicators of recession and the Sahm ruleClaudia Sahm, “Direct Stimulus Payments to Individuals” (2019); FRED labour-market series
• Wage growth, participation and the remaining slackStandard labour-market literature on the employment-to-population ratio and the unemployment gap