Inside the Jobs Report, Two Surveys Can Disagree

Jobs and people are not counted the same way

A payroll is a job on an employer record. One person with two jobs can therefore appear twice in payroll employment. That household measure counts employed people, including self-employed and agricultural workers who are outside the payroll survey's scope. It also supplies the labor force denominator used to calculate unemployment. Those design choices are why the series can differ without either being an error.

The samples differ as well. The August 2026 Employment Situation technical note describes the household survey as a sample of about 60,000 eligible households. Its establishment survey draws on payroll records from about 119,000 businesses and government agencies representing roughly 622,000 worksites. Neither is a census taken at midnight. Both are carefully designed estimates, which is a more interesting and more accurate description.

Three ordinary ways the headlines can separate

• A person takes a second job. Payroll employment rises, while the household count of employed people need not rise by the same amount.
• More people start looking for work. The household survey's labor force can grow, and the unemployment rate can change even as businesses are adding payroll jobs.
• Self-employment or farm work changes. Those people belong in the household employment measure but not in nonfarm payroll employment.

I would add a fourth: timing and revision. BLS says the first payroll estimates are preliminary and are revised in each of the following two months as more reports arrive. The same release also warns that both surveys are subject to sampling and nonsampling error. A dramatic one-month gap is worth noticing, but it is not a command to construct a grand story before the next month arrives.

The denominator is part of the news

Unemployment is a rate, so its denominator matters. The household measure distinguishes people who are employed, unemployed and not in the labor force. The CPS also publishes labor-force participation and the employment-population ratio, which offer different views of how many people are connected to work. If more people begin looking for a job, the labor force can grow. That can alter the unemployment rate even during a month in which employers add payroll jobs.

The reference windows are close but not identical. BLS says its household measure generally refers to the calendar week containing the 12th of the month, while the establishment survey refers to the pay period including the 12th. That is not a loophole or an excuse. It is one more reason the headline pair deserves a sentence of context before anyone declares that one number has defeated the other.

How to let both measures do their work

Start with payrolls when the question is which industries are adding or losing jobs, how many hours workers are logging and what average earnings are doing. Start with the household survey for unemployment, labor-force participation and the employment-population ratio. The CPS home page keeps those household measures together for a reason. It is a small bit of good editorial architecture in a very large federal data system.

Then look for a pattern rather than demanding synchronized monthly steps. Labor-market relationships can be lumpy. Research from the Federal Reserve Bank of San Francisco found periods in which wage growth and unemployment did not follow their usual neat relationship, partly because employers are reluctant to cut nominal wages quickly. Real economies have seams.

The pleasant payoff is that a seeming contradiction becomes an invitation to learn something more precise. I read the two headlines as a pair of windows, not a tie-breaker. Payrolls show the workplace ledger; the household survey shows how people are situated around work. Keeping both open gives the labor market more room to be real.