What the Unemployment Rate Leaves Out, From U-6 to Discouraged Workers
By MHB Admin ·
On the first Friday of most months, the US Bureau of Labor Statistics (BLS) releases a single number that shapes political debate, financial markets and Federal Reserve decisions: the unemployment rate. In its report for September 2026, released on 2 October, the figure was 4.2%. By historical standards, that is low.
Yet a low unemployment rate can coexist with millions of people who want more work than they have, people who have given up looking, and people who have left the labour market altogether. None of them appear in the headline rate. That is not a flaw or a conspiracy. The official rate measures one specific thing, carefully defined, and it is easy to read it as measuring something broader.
This article explains how US unemployment is measured, what the official rate counts and excludes, and what the alternative measures published every month show. It also covers why the two main monthly employment numbers often seem to disagree.
How the government decides who is unemployed
The unemployment rate comes from the Current Population Survey, a monthly household survey conducted for the BLS by the Census Bureau. It covers a sample of about 60,000 eligible households. Interviewers ask people aged 16 and over about their activity during a specific reference week, generally the week that includes the 12th of the month.
Based on the answers, each person is placed into one of three categories.
Employed people did any work for pay or profit during the reference week, even a single hour. Also counted are people who worked 15 hours or more without pay in a family business, and people temporarily absent from a job they have, for example because of illness or vacation.
Unemployed people had no job during the reference week, were available to work, and had actively looked for work at some point in the previous four weeks. People on temporary layoff who expect to be recalled are also counted as unemployed, even if they are not looking.
Not in the labour force covers everyone else: retirees, students, people caring for family members, people with disabilities that prevent work, and people who want a job but have not looked recently.
The labour force is the employed plus the unemployed. The official unemployment rate is the number of unemployed divided by the labour force.
Two features of these definitions explain most of what the headline rate leaves out. First, anyone with even a few hours of paid work counts as employed, however much more work they want. Second, anyone who has not actively searched in the past four weeks is excluded from the labour force entirely, however much they want a job.
The six measures the BLS publishes every month
The BLS does not publish just one measure of unemployment. Each month it releases six measures of labour underutilisation, labelled U-1 to U-6, in table A-15 of the Employment Situation report. The official rate is U-3. The others are narrower or broader.
html
<table>
<thead>
<tr>
<th>Measure</th>
<th>What it counts</th>
<th>September 2025</th>
<th>September 2026</th>
</tr>
</thead>
<tbody>
<tr>
<td>U-1</td>
<td>People unemployed 15 weeks or longer, as a percent of the labour force</td>
<td>1.8%</td>
<td>1.8%</td>
</tr>
<tr>
<td>U-2</td>
<td>Job losers and people who completed temporary jobs, as a percent of the labour force</td>
<td>2.1%</td>
<td>1.9%</td>
</tr>
<tr>
<td>U-3 (official rate)</td>
<td>All unemployed people, as a percent of the labour force</td>
<td>4.4%</td>
<td>4.2%</td>
</tr>
<tr>
<td>U-4</td>
<td>U-3 plus discouraged workers</td>
<td>4.7%</td>
<td>4.4%</td>
</tr>
<tr>
<td>U-5</td>
<td>U-4 plus all other marginally attached workers</td>
<td>5.4%</td>
<td>5.0%</td>
</tr>
<tr>
<td>U-6</td>
<td>U-5 plus people working part time for economic reasons</td>
<td>8.1%</td>
<td>7.6%</td>
</tr>
</tbody>
</table>All figures are seasonally adjusted. For U-4 to U-6, the denominator is also widened to include the additional groups, so the rates are not simple sums.
The difference between U-3 and U-6 is the most important. In September 2026, the official rate was 4.2%, while the broadest measure was 7.6%. The gap is made up of three groups the official rate does not count as unemployed.
Discouraged and marginally attached workers
Marginally attached workers are people who are not working and have not looked for work in the past four weeks, but who say they want a job, are available to take one, and have looked at some point in the past year. In September 2026 there were about 1.5 million of them.
Discouraged workers are a subset of the marginally attached. They have not looked recently specifically because they believe no jobs are available, they lack the necessary qualifications, or they face discrimination. In September 2026, there were about 414,000 discouraged workers.
These groups are excluded from the official unemployment rate because they do not meet the active-search requirement. The rationale is that the official rate should measure people actively competing for jobs. The consequence is that when people give up looking, the official unemployment rate can fall even though their situation has not improved. Economists therefore look at broader measures alongside U-3 to judge whether falling unemployment reflects people finding jobs or people leaving the search.
Part-time for economic reasons
The largest group in the gap between U-3 and U-6 is people who are working but not as much as they want. The BLS counts as part time for economic reasons those who usually work fewer than 35 hours a week but want and are available for full-time work. They are working part time either because their hours have been cut or because they could only find part-time work.
In September 2026, there were about 4.5 million such workers. In the official statistics they are employed, which they are. But they are also underemployed in a sense that matters for household incomes. A worker whose hours have been cut from 40 to 20 a week has lost a large share of their pay without appearing in any unemployment figure.
The BLS distinguishes them from people who work part time by choice, such as students or people with caring responsibilities. Those people are not counted as underemployed.
What even U-6 does not capture
U-6 is broader, but it still leaves out several forms of labour-market difficulty.
Underemployment by skill or pay. A university graduate working full time in a job that does not require a degree, or a skilled worker who has taken a lower-paid job after a layoff, is fully employed in every BLS measure. Some research organisations estimate this kind of mismatch, but it is not part of the official underutilisation measures.
People who want work but have not looked in a year. Marginally attached workers must have looked for work at some point in the previous 12 months. Someone who wants a job but has not searched for longer is simply outside the labour force.
Low pay and instability. Unemployment measures count whether people work, not whether the work pays enough to live on or provides predictable hours. Fluctuating schedules, variable earnings and multiple jobs are invisible in the unemployment figures.
Quality of self-employment and gig work. A person doing a few hours of app-based delivery work in the reference week is employed, regardless of earnings.
Participation, the other half of the picture
Because the unemployment rate depends on who is counted in the labour force, economists also track the labour force participation rate: the share of the civilian population aged 16 and over that is either employed or unemployed. In September 2026 it was 61.8%. The employment-population ratio, the share of the population that is employed, was 59.2%.
These measures help distinguish between two very different situations. If unemployment falls because people find jobs, the employment-population ratio rises. If it falls because people stop looking, participation falls and the employment-population ratio may not improve.
The overall participation rate has to be interpreted with care, because it is strongly affected by ageing. US participation peaked at 67.3% in early 2000 and has generally been lower since, partly because the large baby-boom generation has been retiring. Retirement reduces participation without indicating a weak labour market. For that reason, many economists prefer the participation rate or employment-population ratio for people aged 25 to 54, the so-called prime working age. Those groups are less affected by retirement and schooling, which makes them a cleaner measure of whether people who might be expected to work are actually working.
Why the two monthly job numbers often disagree
The Employment Situation report contains two surveys, and they frequently tell different stories.
html
<table>
<thead>
<tr>
<th>Feature</th>
<th>Household survey (Current Population Survey)</th>
<th>Establishment survey (Current Employment Statistics)</th>
</tr>
</thead>
<tbody>
<tr>
<td>Who is surveyed</td>
<td>About 60,000 households</td>
<td>About 119,000 businesses and government agencies, covering about 622,000 worksites</td>
</tr>
<tr>
<td>What it counts</td>
<td>People</td>
<td>Jobs on payrolls</td>
</tr>
<tr>
<td>Main headline</td>
<td>Unemployment rate, participation, employment-population ratio</td>
<td>Change in nonfarm payroll employment</td>
</tr>
<tr>
<td>Self-employed and gig workers</td>
<td>Included</td>
<td>Excluded</td>
</tr>
<tr>
<td>People with more than one job</td>
<td>Counted once</td>
<td>Each job counted separately</td>
</tr>
<tr>
<td>Agricultural workers</td>
<td>Included</td>
<td>Excluded</td>
</tr>
</tbody>
</table>The headline "jobs added" figure, which was 29,000 in September 2026, comes from the establishment survey. It counts jobs on payrolls, not people. A worker who takes a second job adds a job to the payroll count without changing the number of employed people. A self-employed contractor appears in the household survey but not in the payroll count.
The two surveys also have different sampling errors. The household survey's sample is smaller relative to the population it represents, so its monthly changes are noisier. Both are revised. Payroll estimates are revised in each of the following two months as more employer reports arrive, and once a year they are benchmarked against near-complete records from state unemployment insurance tax filings. Those benchmark revisions can be large. In September 2025, the BLS's preliminary benchmark estimate indicated that payroll employment for March 2025 had been overstated by about 911,000 jobs.
Measurement itself can also be interrupted. During the federal government shutdown in autumn 2025, data collection was disrupted, and the BLS did not publish a household-survey unemployment rate for October 2025. Gaps of this kind are rare, but they show how much the monthly picture depends on continuous survey operations.
Why the long-term unemployed matter
Another figure in the monthly report adds important context: the duration of unemployment. In September 2026, about 1.9 million people had been unemployed for 27 weeks or more, accounting for 27.1% of all unemployed people.
Long-term unemployment matters for several reasons. Research has found that people unemployed for many months are less likely to find work than the short-term unemployed. That may be because employers view long gaps negatively or because skills and networks erode over time. Long-term unemployment is also associated with exhausting unemployment insurance benefits, which in most states last up to 26 weeks under regular programmes. And people who remain unemployed long enough may eventually stop looking, moving from the unemployed to the marginally attached or out of the labour force entirely. At that point they disappear from U-3.
A labour market can have a low overall unemployment rate while a substantial share of those who are unemployed have been looking for a very long time.
Wages are a separate question
Unemployment measures say nothing directly about pay. The same report includes average hourly earnings, which rose to $37.81 for private nonfarm employees in September 2026, up 3.0% over the previous 12 months.
That figure is nominal. Whether workers gained purchasing power depends on inflation over the same period, measured separately by the Consumer Price Index. Average hourly earnings is also an average that can be pulled up or down by changes in the mix of jobs. When lower-paid workers lose jobs disproportionately, as happened in spring 2020, average earnings can rise even though no individual worker received a raise.
Reading the monthly report more fully
The headline rate is useful, consistent and comparable over long periods. It is a reasonable first indicator of labour-market slack. But it answers a narrow question: of the people actively looking for work, what share have not found it?
A fuller reading of any month's report looks at several numbers together. U-6 and the number of people working part time for economic reasons show underemployment among the employed. Marginally attached and discouraged workers show people who want jobs but have stopped searching. Participation and employment-population ratios, preferably for prime-age workers, show whether people are moving into or out of work. The duration of unemployment shows whether job seekers are finding work quickly. And payroll and household figures should be read with their differences in mind, along with the knowledge that both will be revised.
Historical context
The gap between the official rate and the broader measures tends to widen in downturns and narrow in strong labour markets. During the 2007–2009 recession, the official unemployment rate peaked at 10.0% in October 2009, while U-6 rose to around 17%. During the pandemic shock in April 2020, the official rate reached 14.8% and U-6 reached 22.9%.
By September 2026, the official rate stood at 4.2% and U-6 at 7.6%. Both were below their year-earlier levels of 4.4% and 8.1%. The gap between them was about 3.4 percentage points, much narrower than in the downturns above.
Those comparisons show that the broader measures are not hidden figures that contradict the official rate. They move in the same general direction and are published in the same report. What they add is a more complete picture of who is struggling in the labour market and how. That is information the single headline number, by design, does not provide.
