How to Read the US Jobs Report - And Why Its Two Headline Numbers Disagree

Once a month, usually on a Friday morning, the US Bureau of Labor Statistics publishes a document that reliably produces more market movement per page than anything else on the economic calendar.

It is also routinely reported in a way that obscures what it actually contains. The report is not one measurement. It is two.

Two surveys, not one

The Employment Situation report combines two entirely separate statistical exercises.

The establishment survey contacts businesses and government agencies - roughly 119,000 of them, covering around 622,000 individual worksites - and asks how many people were on the payroll. This produces the nonfarm payrolls figure, the headline number of jobs added or lost.

The household survey contacts households and asks the people in them whether they are working, looking for work, or neither. This produces the unemployment rate, along with the participation rate and the employment-population ratio.

Different samples. Different questions. Different units of measurement.

Why they disagree

This is the single most useful thing to understand about the report, because it explains most of the apparent contradictions.

The establishment survey counts jobs. The household survey counts people. Someone holding two part-time jobs appears twice in payrolls and once in the household data.

They cover different workers. The payroll survey excludes the self-employed, agricultural workers and unpaid family workers. The household survey includes all of them. A month in which self-employment expands while companies trim staff can show weak payrolls and a steady unemployment rate simultaneously.

They have different sampling errors. The household survey has a considerably smaller sample and is therefore noisier month to month.

So when a headline says the jobs report was "mixed", it often means the two surveys disagreed. That is not a flaw in the data. It is two instruments measuring adjacent but distinct things.

The revisions problem

The payroll number published on the day is a first estimate based on incomplete survey responses.

It is revised in each of the next two monthly reports as late responses come in, and again in an annual benchmark revision that reconciles the survey against actual tax records.

Those revisions are not cosmetic. There have been months where the revision exceeded the original reported change, which means the initial figure that moved markets by billions was later shown to have described a different situation entirely.

The practical consequence: the three-month average is more informative than any single month, and a headline that looks like a turning point frequently is not one.

What is underneath the headline

The two numbers that lead coverage are the least detailed part of the release.

Average hourly earnings is often the market-moving line rather than the job count, because it feeds directly into the inflation outlook. Strong hiring with flat wages is a very different signal from weak hiring with accelerating wages.

The participation rate tells you whether the unemployment rate fell because people found work or because they stopped looking. A falling unemployment rate driven by people leaving the labour force is not good news dressed as good news - it is bad news dressed as good news.

Sector detail shows where the change happened. Hiring concentrated in leisure and hospitality has different implications from hiring in professional services or manufacturing.

Average weekly hours frequently turns before headcount does. Employers cut hours before they cut people, so a decline here can lead a slowdown in payrolls.

How markets actually trade it

The mechanical part is that markets already carry a forecast. Economists publish estimates, and those estimates are embedded in prices before the release. What moves the market is the surprise - the distance between the actual figure and the consensus.

This is why a genuinely healthy jobs number can knock equities down. If the number implies a tighter policy path than was priced, the discount rate applied to future earnings rises, the dollar tends to firm, and both effects push asset prices lower regardless of how good the news is for employment.

The transmission runs through rate expectations, which is also why the report reaches currencies and gold within seconds of publication, not just equities.

It is also why the reaction sometimes reverses within an hour. The first move is algorithmic, keyed to the headline. The second move follows once the detail - wages, participation, revisions - has been read.

A reading order that works

1. Payrolls against consensus. The surprise, not the level.
2. Revisions to the previous two months. These often matter more than the new figure.
3. Average hourly earnings, month-over-month and year-over-year.
4. The unemployment rate, with participation. Did it move for a good reason?
5. Weekly hours. An early indicator of direction.
6. The three-month average. The trend, once the noise is stripped out.

The bottom line

The jobs report is two surveys stapled together, published as a first estimate, and revised twice afterwards. Reported as a single number, it is misleading. Read properly, it is one of the most informative documents released all month.

The BLS publishes its schedule a year in advance, so you always know when it is coming. Reading one release carefully, all the way through, is worth more than following the headline for a year.

This article is educational and is not financial advice. Economic data is subject to revision and past patterns do not indicate future results.