A fraction has two halves
The unemployment rate is not a count of people without work. It is the number of unemployed people divided by the labour force, and the international definition of "unemployed" is narrower than it sounds: without work, available to start, and actively looking. Someone who stops looking does not become unemployed. They leave the labour force, which means they leave the bottom of the fraction as well as the top.
That is why a falling rate needs reading twice. It falls when unemployed people find work, which is the reading everyone reaches for. It also falls when unemployed people stop looking, because that removes one from the numerator and one from the denominator, and taking one from a small number changes a fraction more than taking one from a large one.
Between May and July 2026 the number of unemployed Americans fell by 391,000. That part is straightforward and it is the reason the published rate moved. The question this piece is about is what the other half of the fraction was doing at the same time.
What the same survey says happened
The rate and the employment count come from the same instrument - a monthly survey of households, which asks people what they did in a particular week. So the two numbers below cannot be reconciled by saying they measure different populations. They measure the same population, in the same interviews, in the same months.
The number of people with a job fell by 594,000. The labour force - the employed and the unemployed added together - fell by 984,000. And the count of people outside the labour force altogether rose by 1,213,000. The employed fell by more than the unemployed did, which is the whole story in one sentence.
| What is counted | May | July | Change |
|---|---|---|---|
| Unemployed - without work, available, and actively looking | 7,307 | 6,916 | -391 |
| Employed - did any paid work in the reference week | 162,771 | 162,177 | -594 |
| Labour force - the employed and the unemployed together | 170,078 | 169,094 | -984 |
| Outside the labour force - neither working nor looking | 104,976 | 106,189 | +1,213 |
| Participation rate - the labour force as a share of the population | 61.8% | 61.4% | -0.4 pp |
| Employment-population ratio - the employed as a share of the population | 59.2% | 58.9% | -0.3 pp |
The ratio that cannot do this
There is a second ratio in the same release, and it settles the question without needing anyone to take our word for it. The employment-population ratio divides the number of employed people by the population, not by the labour force. The population does not shrink when someone stops looking for work. That denominator sits still.
It went the other way. In May, 59.2% of the American adult population had a job. In July, 58.9% did. Over the two months in which the unemployment rate improved by two tenths of a point, the share of the population in work fell by three tenths.
The broader measure of underemployment tells a quieter version of the same thing. It counts the unemployed, plus people who want work but have stopped looking, plus people working part-time who want full-time hours. It fell from 8.1% in May to 7.9% in June - and then stopped, at 7.9% in July, in the month the headline rate fell again.
How much of the fall was the denominator
The arithmetic can be run the other way to put a size on it. Hold the participation rate at its May level of 61.8%, apply it to July's population, and the labour force would have been about 1,102,000 larger than it was. Count all of those people as unemployed and the July rate would be about 4.7%, against a published 4.1%.
That figure is derived, it is ours, and it is an upper bound rather than an estimate. It assumes every single person who left the labour force would have been counted as unemployed had they stayed, and that is not what would have happened: some would have found work, and some were never going to look. It is not a forecast of an alternative July. It is a measure of how much room the denominator had to move the answer.
| July 2026 unemployment rate | Per cent |
|---|---|
| As published | 4.1% |
| Had participation stayed at 61.8% (derived, upper bound) | 4.7% |
What this does not say
It does not say the rate is wrong. It is computed correctly from the definitions, and those definitions are the international standard rather than an American convenience. Divide 6,916,000 unemployed by a labour force of 169,094,000 and you get 4.09%, which rounds to the 4.1% that was published.
It does not say the people who left had given up. A survey that records whether someone looked for work in the last four weeks does not record why they did not. Retirement, study, caring, illness and discouragement all leave the same trace in the data, and this release cannot separate them. Anyone who tells you which one it was is telling you something the survey did not ask.
And underutilisation is not hidden unemployment. The people outside the labour force are not being concealed - they are published every month, in a series with their name on it, which is how we counted them for this piece. They are simply not in the fraction that gets read aloud.
For anyone actually looking for work, the practical reading is narrower and more useful than the headline. The rate falling does not mean there were more openings. Over the same two months the payroll count of jobs was essentially flat, ending 3,000 lower than it started. A market that is not shedding workers quickly is not the same as a market that is hiring.
We publish how we test our matching for bias, including where it still falls short. Read the bias audit.