How the jobs numbers work, answered in eleven questions

Picture a Friday morning, 8:30 Eastern, the first Friday of the month or close to it. A government agency posts a few tables, and within ninety seconds every cable chyron in America has a single number on it. One party calls it proof that the country is thriving. The other calls it proof that the country is being fleeced. Both are looking at the same table, and I will bet that neither of them, in most cases, could tell you what the number actually counts.

That is a problem for the rest of us, because the jobs report is the closest thing we have to a monthly report card on the paycheck economy, and a report card is only useful if you know what the grades mean. So here is the whole thing in questions and answers. No current figures, no predictions, no scoring points for either side. Just the machinery, explained the way you would explain it to a neighbour at the fence. The agency is the Bureau of Labor Statistics, part of the Department of Labor, and its monthly release is called the Employment Situation. Everything below comes from how the BLS describes its own methods, and you can check every definition on its website.

The basic definitions

1. Who counts as "unemployed"?

Not everybody without a job. This is the first thing people get wrong. To be counted as unemployed you must meet three tests: you had no job in the survey week, you were available to take one, and you actively looked for work at some point in the previous four weeks. The exception is someone on temporary layoff who expects to be called back, who does not have to be looking.

Someone who stopped searching months ago is not unemployed by this definition. Neither is a retiree, a full-time student, or a parent who left the workforce to raise children. They are "not in the labor force." Whether that definition is the right one is a fair argument. But it is the definition, it has been applied the same way for decades, and it is why the headline rate can fall for two very different reasons: people found jobs, or people stopped looking.

2. And who counts as "employed"?

Anyone who did at least one hour of paid work in the survey week, plus anyone who worked fifteen or more unpaid hours in a family business, plus people who had a job but were off that week for vacation, illness or weather. One hour. A man who mows lawns for a single paid hour is employed in the statistics. That is not a trick. It is a design choice, because the survey wants to measure whether the economy is putting people to work at all, not whether the work is any good.

The survey week, by the way, is the one that includes the 12th of the month. Keep that date in your head, because it comes back when we talk about the second survey.

3. What does the unemployment rate actually divide by what?

Unemployed people divided by the labor force, where the labor force is the employed plus the unemployed. Here is a hypothetical with round numbers, labelled as one: a small country with 100 adults old enough to count, of whom 60 have jobs, 4 are looking for work and 36 are retired, studying or at home. The labor force is 60 plus 4, which is 64. The unemployment rate is 4 divided by 64, which is 6.25 percent. The 36 who are outside the labor force appear nowhere in that fraction.

Now suppose two of the four job seekers give up and drop out. The unemployed are now 2, the labor force is 62, and the rate is 2 divided by 62, or about 3.2 percent. Nobody got a job, and the rate fell by half. That is the weakness of the headline number in one paragraph, and it is why serious people look at more than one measure.

The broader measures

4. What are U-3 and U-6?

The BLS publishes six alternative measures of labor underutilization, labelled U-1 through U-6. U-3 is the official unemployment rate, the one in the headlines: the unemployed as a share of the civilian labor force. U-6 is the broadest. It adds the people who are "marginally attached" to the labor force, meaning they want a job and looked at some point in the past year but not in the last four weeks, and it adds people working part-time who want full-time work and cannot get it. The BLS calls those workers part-time for economic reasons.

Within the marginally attached sit the "discouraged workers," who are not looking specifically because they believe no job is available for them. U-6 is always higher than U-3, because it counts more people. Neither is "the real number." They answer different questions. U-3 asks how many are actively searching. U-6 asks how much slack is in the labour market altogether. If someone on either side tells you only one of them is honest, they are selling something.

5. What is labor force participation, and why does it matter?

It is the labor force divided by the civilian non-institutional population aged sixteen and over. The non-institutional part leaves out people in prisons, nursing homes and the like, and the civilian part leaves out the armed forces. It tells you what share of adults are working or looking.

Participation moves for reasons that have nothing to do with the economy's mood. An older population means more retirees, and retirees leave the labor force by definition. More young people in school lowers it. So many analysts prefer to look at prime-age workers, usually defined as ages 25 to 54, where retirement and schooling matter far less. A related gauge is the employment-population ratio, the share of all those adults who are actually employed. It cannot be flattered by people giving up, because it does not have a "looking" test at all. When you hear a claim about the jobs picture, ask which of these three the speaker is using.

The two surveys and their revisions

6. Why are there two surveys, and why do they disagree?

The headline unemployment rate comes from the household survey, formally the Current Population Survey, conducted by the Census Bureau for the BLS. Interviewers contact a sample of households, tens of thousands of them, and ask about each adult's activity in the survey week. The headline job-count figure, the "jobs added" number, comes from a different survey of employers, the Current Employment Statistics program, which asks well over a hundred thousand businesses and government agencies how many people are on their payrolls for the pay period that includes the 12th.

They disagree because they measure different things. Count them as follows:

1. The household survey counts people. The employer survey counts jobs. A person with two jobs is one person in the first survey and two payroll entries in the second.
2. The household survey includes the self-employed, farm workers and unpaid family workers. The employer survey, which counts payrolls, mostly does not.
3. The household survey is a sample of families, so its month-to-month swings carry a larger margin of error. The BLS publishes that margin in the technical note to each release, and a change of a tenth or two in the unemployment rate is often within it.
4. They draw on different samples, so even identical concepts would produce slightly different numbers.

Over many months the two tend to tell a similar story. In any single month they can point in different directions, and a partisan can always pick the one that flatters. A careful reader looks at both.

7. Why does the "jobs added" number get revised?

Because the first estimate is made before all the employers have reported. The BLS publishes an initial figure, then revises it in each of the next two monthly reports as late responses arrive. Months later, the numbers are benchmarked against a near-complete count taken from state unemployment-insurance tax records, which cover almost every employer, and the estimates are adjusted to match. This annual benchmark revision can move the level of employment noticeably.

There is also an estimate for the businesses that opened or closed since the sample was drawn. The BLS uses a statistical model to account for them, often called the birth-death model, because a brand-new company is not yet in anybody's list. It is a sensible fix and a known source of error. The practical lesson: treat the first print of any month as a first draft. Quoting it as gospel, in either direction, is a habit of people who need a headline.

8. What is seasonal adjustment, and is it cooking the books?

It is not. Hiring has regular rhythms. Stores take on seasonal help in the fall and let it go in January. Construction slows in northern winters. Schools empty out in June and teenagers flood into summer jobs. If the BLS published only the raw counts, every December would look like a boom and every January like a disaster, and you could not tell a real change from the calendar.

Seasonal adjustment uses years of past data to estimate the normal pattern for each month and removes it, so that a reported change reflects something other than the usual rhythm. The raw, unadjusted figures are published too, in the same release tables, for anyone who wants to check. The adjustment factors are themselves revised once a year. Can the method misfire when the pattern changes, as when the pandemic scrambled everything? Yes. That is a limitation to keep in mind, not a conspiracy.

Two common mix-ups

9. Is the unemployment rate the same as the number of people collecting benefits?

No, and this is worth a question of its own. The unemployment rate comes from the household survey, which asks people about their situation. It does not count benefit recipients, and plenty of unemployed people are not collecting anything. Some never qualified, some have run out, and some never applied.

The benefit data come from a separate source: the weekly report from the Department of Labor on initial and continued claims for state unemployment insurance. It is a different series answering a different question, namely how many people are filing. Analysts watch it because it arrives weekly, which makes it timely. But it is not the unemployment rate, and the two should never be swapped in an argument.

10. Can a politician or a president change the numbers?

The structure is built to make that hard. The data are collected and tabulated by career statisticians, using methods that are documented in public and applied the same way regardless of who sits in the White House. The Census Bureau does the household interviews. The commissioner of labor statistics is a presidential appointee confirmed by the Senate, but the survey design, the definitions and the release schedule are matters of long-standing practice, and the technical notes are public. Anyone who thinks a number is wrong can compare it to the unemployment-insurance tax records, to private payroll data, and to the raw tables.

That is the strength of a system with published methods: you can criticise it, and you can check it. Skepticism of Washington is a conservative virtue. Inventing fraud where there is a documented method is something else, and it hurts us, because it lets the other side wave away our real criticisms of how the numbers are used. If you think a particular measure misleads, argue about the definition. That argument is legitimate, and it is the one in Question 4.

How to read the report

11. So what is the right way to read the report?

Here is a short routine, whichever party is in power:

1. Read the trend over several months, not one month. A single print is a first draft.
2. Check both surveys. If they disagree, say so out loud.
3. Look at participation and the employment-population ratio alongside the unemployment rate, so a falling rate cannot hide people leaving the workforce.
4. Compare U-3 with U-6 to see the slack the headline leaves out.
5. Remember that the report counts jobs, not wages, not hours, not the cost of living. A job is the start of the question, not the answer to it. That is why this site also gives you How the CPI is built and why it never matches your grocery receipt and why your raise feels smaller than the number your boss pays out.

The report tells you whether people are working. It does not tell you whether work is paying, whether it is the work they wanted, or whether the dollar they earn buys what it did last year. For that you need prices, and you need to understand what the central bank does to them, which is the story in The Federal Reserve in plain English and the levers it really holds. And if the unemployment rate has you thinking about the rules that keep people out of jobs in the first place, the license you need to braid hair and the jobs it quietly kills is the post to read next.

The bottom line

• The unemployment rate counts only people with no job who are actively looking. It can fall because people found work or because they stopped searching, so never read it alone.
• Two surveys, households and employers, measure different things. Expect them to disagree in a given month, and trust neither's first print.
• Check U-6, participation and the employment-population ratio before you accept anyone's victory lap or doom story.
• The method is public and documented. Argue about definitions; they are fair game. Do not argue that the arithmetic is secretly fake.
• The jobs report tells you who is working. To know whether your family is getting ahead, put it next to prices and wages, and ask what the Fed is doing to both.