When Different Numbers Disagree
The U.S. labor market is measured by a collection of instruments, not one national sensor. This chapter explains why unemployment, payrolls, vacancies, claims, wages, and projections can point in different directions without one automatically being wrong. The reader’s task is to label each number by what it measures, when it measures it, how it has been transformed, and whether it is an observation, an estimate, a revision, or a conditional forecast.
“The data” are several measurement systems
The Federal Reserve’s July 2026 report draws its labor-market picture from multiple sources. Civilian unemployment, participation, payroll employment, and demographic employment-to-population ratios come through Bureau of Labor Statistics and Census Current Population Survey data. Job openings and turnover use JOLTS. Online postings provide an alternative vacancy signal. Unemployment-insurance claims come from the Department of Labor. Wage measures include the employment cost index, average hourly earnings, and the Atlanta Fed’s Wage Growth Tracker. Productivity comes from another statistical series, with Federal Reserve staff calculations used in some comparisons.
These sources are not repeated measurements of exactly the same object. The unemployment rate is a share of labor-force participants without work. Payroll employment counts jobs on payrolls. EPOP relates employment to a population. JOLTS openings describe positions employers report as available, while postings count a different kind of online signal. Claims register one path into unemployment support, not every separation. A wage tracker may follow the change in pay for people observed at different times; the employment cost index measures compensation more broadly. A productivity measure connects output and hours, not job search.
The differences are useful because no single instrument covers the whole market. Payrolls can show jobs being added even when participation is changing. The unemployment rate can remain low while openings and quits flatten. Claims can stay muted while people who are not employed do not enter the labor force. A wage measure can rise in nominal terms while real purchasing power falls. Complementarity, however, is not interchangeability. Each series carries its own population, unit, timing, and blind spot.
This is also why a disagreement between indicators can be informative. Suppose payroll employment is stable, openings are falling, and layoffs remain low. That combination could mean employers are retaining current workers while reducing future hiring plans. Suppose openings are high but payroll gains are weak. Matching may be slow, or the openings may be concentrated in sectors that are difficult to fill. The source material does not supply one universal explanation for either pattern. It gives the reader a reason to inspect the instruments before choosing one.
The safest language follows the evidence. “The payroll series shows…” is narrower than “employers are confident.” “The unemployment rate is low…” is narrower than “the labor market is healthy.” “The Federal Reserve interprets these indicators as roughly balanced…” is different from asserting that demand and supply have been measured as exactly equal. Precision about the instrument protects the conclusion from carrying more certainty than the data can support.
There is a second layer beyond the name of the source: the way a source is used. Some report figures reproduce a series; others compare groups with Federal Reserve staff calculations, estimate a missing month, or smooth a history to remove a discontinuity. The resulting chart may be the right tool for the report’s question, but it is no longer a raw observation in the narrow sense.