Read the denominator before the headline.

Picture a register of 100 adults, one tally mark each. Sixty of them have jobs. Five have no job and are looking for one. The other 35 are neither working nor looking: retired, in school, caring for someone, or simply not searching. The register is invented for this dispatch, and so are its numbers. Now take two of the five job seekers and move their marks to the third column, because they have stopped looking. Nobody found work. Nobody lost work. Yet the unemployment rate in this register just fell from 7.7 percent to 4.8 percent.

The trick is that a rate is a fraction, and the fraction changed at the bottom. Unemployment within the labor force divides the people looking for work by the people who are either working or looking: 5 out of 65 before the move, 3 out of 63 after. Employment within the population divides the people with jobs by everyone in the register: 60 out of 100, before and after. One number moved a lot; the other did not move at all. Both are correct. They answer different questions.

My argument is narrow. Read the unemployment rate together with the employment-population ratio and the participation rate before treating a change in any one of them as a plain improvement or a plain decline. The denominator is part of the economic story, not a footnote to it. This does not claim that the unemployment rate is misleading, or that people leaving the labor force is always bad news. It claims only that one rate cannot tell you which of several stories is happening.

What the July 2 release said

The real register is the Bureau of Labor Statistics' monthly Employment Situation, and the June report came out at 8:30 in the morning on Thursday, July 2. Its first sentence said that both total nonfarm payroll employment, up 57,000, and the unemployment rate, at 4.2 percent, changed little in June. Further down, the household data told a second story. The labor force participation rate fell 0.3 percentage point to 61.5 percent, and the employment-population ratio edged down 0.2 point to 59.0 percent. The unemployment rate had been 4.3 percent in May.¹

So in the same month, the share of the labor force without a job ticked down while the share of the population with a job also ticked down. That is the register's pattern in miniature. It does not mean the register's story is the true one. It means the headline rate, read alone, cannot distinguish between more people finding work and fewer people looking.

The three rates are tied together by a piece of arithmetic worth keeping in your pocket. The employment-population ratio equals the participation rate times the share of the labor force that has a job. Multiply 61.5 percent by 95.8 percent and you get 58.9 percent, which matches the published 59.0 once you allow for rounding. If participation falls while the unemployment rate holds steady, the employment ratio has to fall. If you watch only the last term in that product, you can miss a move in the first.

Two surveys, not one

The release is really two reports stapled together. Payroll employment comes from the establishment survey, about 119,000 businesses and government agencies covering roughly 622,000 worksites. The rates come from the household survey, about 60,000 households interviewed for the Bureau by the Census Bureau. BLS itself warns that the two differ in concept and method in ways that matter for their employment estimates.¹ A count of jobs and a count of people are different units. A person with two jobs appears twice in one and once in the other. It is tempting to set the payroll gain beside the household numbers and solve for what must have happened in between. The release does not support that subtraction, and this dispatch does not attempt it.

The household survey is also the noisier of the two for month-to-month changes. The technical note puts the 90 percent confidence interval on a monthly change in the unemployment rate at about 0.3 percentage point, at an unemployment rate of around 6 percent.¹ A move of 0.1 in the headline rate, or 0.2 in the employment ratio, is within the range a sample can produce on its own. That is why the Bureau wrote "changed little," and why that phrase deserves to be read as a finding rather than skipped as boilerplate. One month is not a turning point.

Leaving the count has many causes

Suppose the drop in participation is real. It still does not explain itself. A person counts as unemployed only if they had no job in the survey's reference week, were available to work, and actively looked sometime in the previous four weeks.¹ Someone who retires, starts a degree, takes on the care of a parent, falls ill, or gives up searching all leave the labor force the same way, and the headline rate records them identically.

The release offers some help in telling these apart, though not much in a single month. In June, 6.0 million people outside the labor force said they currently wanted a job, a number BLS described as little changed. Of those, 1.8 million were marginally attached, meaning they had looked within the past year but not the past four weeks, and 477,000 were classed as discouraged, believing no job was available for them.¹ Those figures are context, not a diagnosis. Nothing in the release says whether June's exits came from retirement, technology, immigration or policy, and I would not attribute them to any of those without separate evidence.

From the headline to the hiring plan

For a business planning to hire, the national numbers are weather, not the map. They tell you something about the climate in which you will recruit. They do not tell you whether the people you need, in the occupation and the place you need them, are easier or harder to find this quarter. Before changing a hiring plan on the strength of a headline, look at the evidence closer to the decision: applicant flow for the roles in question, time to fill, offer acceptance, hours and pay in the local market. If those move with the national rates, the national rates have told you something. If they do not, the headline was about someone else's register.

The same discipline applies to anyone building a workforce dashboard. A tile that shows one rate invites one reading. Put the unemployment rate, participation and the employment ratio on the same line, label each with its denominator, and the dashboard can no longer tell a story that the arithmetic does not support.

Objections, considered

Falling participation is not necessarily bad. Correct. A retirement long planned for, or a year spent in school, can be good for the person and for the economy. This dispatch attaches no single verdict to every exit. It asks only that the exit be seen as an exit, not mistaken for a hire.

The unemployment rate is still useful. It is. It is precisely defined, consistently measured and widely understood, which is why it leads the release. It measures slack among people in the market for work. It was never meant to describe everyone's labor supply or every household's well-being, and asking it to is a misuse of a good tool.

One month is noise. Often, yes, and the confidence intervals above say so. A view across several months is more informative, provided every month comes from a consistent vintage. The same July 2 release revised April and May payrolls down by a combined 74,000, a reminder that even the past is provisional.¹ The case for reading the three rates together does not depend on June being a turning point. It holds in any month.

A falling rate can describe a changing denominator as well as a changing economy.

Before the next jobs report

• Which population does this rate describe? The labor force, the whole population, or payroll jobs. Say it out loud before quoting the number.
• Do employment and participation tell the same story? If the unemployment rate falls while the employment ratio also falls, the story is about who is looking, not only who is hired.
• Is the change bigger than the survey's margin? "Changed little" is a result. Treat it as one.
• What local evidence informs the decision? Name the occupation, the place, and the measure you will check before the hiring plan moves.

Back to the register. Draw the gold ring around the 65 who are working or looking, and the blue ring around all 100. Move the two marks across the line, and watch the gold fraction shrink while the blue one stays exactly where it was. Both rings were always on the page. The headline only ever showed one of them.