US Labour Reports: Payrolls, the Household Survey, Wages, JOLTS, Claims and Revisions

What you will learn

• How the establishment survey (payrolls) and the household survey (unemployment rate) differ.
• How revisions, the annual benchmark and the birth–death model change the meaning of the headline.
• What average hourly earnings, JOLTS and weekly claims add.
• How to read a report in the first minute after release.

Explanation

Two surveys in one report. The Employment Situation report from the Bureau of Labor Statistics (BLS), usually published on the first Friday of the month at 08:30 Eastern Time, combines two independent surveys. The establishment survey asks employers how many people were on their payroll in the pay period including the 12th of the month; it produces nonfarm payrolls, hours and average hourly earnings (AHE). The household survey interviews households and produces the unemployment rate and the participation rate (labour force divided by the civilian population aged 16 and over). A person with two jobs counts twice in payrolls but once in the household survey; the self-employed appear only in the household survey. The household sample is much smaller, so its monthly employment change is far noisier: the BLS cites a confidence interval of several hundred thousand for the household count against a little under plus or minus 140,000 for payrolls (check the current technical note). The unemployment rate is more reliable than the household employment level because errors in numerator and denominator partly cancel out.

Revisions. Each payroll figure is revised in the following two reports as late responses arrive, so the relevant number is the headline plus the net revision to the previous two months. Once a year the BLS benchmarks the level to near-complete counts from unemployment insurance tax records (the Quarterly Census of Employment and Wages): a preliminary benchmark in late summer, the final one in the January report published in February. The March 2025 benchmark was unusually large: about −911,000 jobs preliminarily and −898,000 in the final figure, around 0.6 percent of employment; the March 2026 preliminary benchmark was only about −79,000. A few tenths of a percent is normal.

The birth–death model. The survey cannot capture firms that have just opened or closed, so the BLS adds a modelled estimate of net jobs from business births minus deaths. The model extrapolates recent history, which works in steady times and fails at turning points, when it adds jobs that do not exist, as the 2025 benchmark illustrated.

Wages. AHE is affected by composition: if low-wage workers lose jobs, average earnings rise mechanically. The quarterly Employment Cost Index is cleaner but less timely.

JOLTS and claims. The Job Openings and Labor Turnover Survey (JOLTS) arrives five to six weeks late with openings, hires, quits and layoffs; watch openings per unemployed person and the quits rate. Initial jobless claims come every Thursday for the previous week, the timeliest indicator but volatile around holidays, so use the four-week average. Continuing claims show how hard it is to find a job.

Schedules can break. During the 43-day US federal shutdown in autumn 2025 the October household survey was never collected, so there is no official unemployment rate for that month. Always check the current BLS calendar.

Worked example

Hypothetical example: payrolls +150,000 against a consensus of +160,000; the previous two months revised down by a combined 30,000; 6.8 million unemployed in a labour force of 168.0 million and a civilian population of 270.0 million. AHE rises from $35.00 to $35.12 ($33.80 a year ago). Weekly claims: 215,000, 222,000, 230,000, 237,000. JOLTS openings: 7.2 million.

1. Net new information: 150,000 − 30,000 = +120,000, a miss of 40,000 against consensus once revisions are included.
2. Unemployment rate: 6.8 / 168.0 = 4.05 percent. Participation rate: 168.0 / 270.0 = 62.2 percent.
3. AHE m/m: 35.12 / 35.00 − 1 = 0.34 percent; y/y: 35.12 / 33.80 − 1 = 3.91 percent.
4. Four-week average of claims: (215 + 222 + 230 + 237) / 4 = 226,000, rising steadily.
5. Openings per unemployed: 7.2 / 6.8 = 1.06, near balance.

Reading: a miss that looks larger after revisions, rising claims and a balanced openings ratio point to gradual cooling, not collapse; wage growth near 3.9 percent is the one line still warm.

Common mistake

Practice activity

On paper: payrolls +90,000 against +110,000 expected; prior two months revised up by 45,000. Unemployed 7.3 million; labour force 169.0 million. AHE $36.10 this month, $36.00 last month, $34.90 a year ago. Calculate the revision-adjusted figure versus consensus, the unemployment rate and both wage growth rates, and write a short reading.

Key takeaways

• Payrolls come from the establishment survey, the unemployment rate from the household survey; they have different error bands.
• Always add the net revision to the previous two months before judging the headline; the yearly benchmark can move the level by several hundred thousand.
• The birth–death model adds modelled jobs that tend to overstate hiring at turning points.
• Use the four-week average of claims and the JOLTS openings-per-unemployed ratio to cross-check the monthly report.

Next step

Next we study the purchasing managers' indices, the fastest broad business surveys available, and learn how a diffusion index is calculated.