Why Two Official Earnings Figures Disagree

One number measures households and takes the median; the other measures jobs and reports an arithmetic mean. Mistaking one for the other is the most common mistake in wage-reporting—and the most avoidable.

The household unit and the job unit

Each series counts a different statistical unit, and that is where the trouble starts.

The Census Bureau’s Current Population Survey Annual Social and Economic Supplement, the CPS ASEC, is the official source for annual estimates of income and poverty at the national level. Its unit is the household. That figure sweeps in everyone who shares a residence and pools their income: wages and salaries, interest, dividends, rental income, Social Security, public assistance, and the rest. A household of four with two earners and a pension counts as one line in the survey.

The Bureau of Labor Statistics runs two big wage programs, and neither looks at households. The Current Employment Statistics survey, table B-8a, covers production and nonsupervisory employees on private nonfarm payrolls—specifically production employees in mining, logging, manufacturing, and construction, plus nonsupervisory employees in service-providing industries. The Occupational Employment and Wage Statistics program, OEWS, measures occupational employment and wage rates for wage and salary workers in nonfarm establishments. Both series count jobs. A person holding two jobs shows up twice. A household with zero earners and a government transfer does not show up at all.

The consequence is mechanical: a series built from jobs will look at a different population than a series built from residences, even before anyone chooses a statistic.

Median versus mean: a statistical fork

Once the universe is set, the Census and OEWS programs pick different summarizing statistics, and that deepens the divide.

The Census Bureau defines median income as the amount that splits an income distribution into two equal halves. Mean income is the total income of a statistical universe divided by the number of units in that universe. Mean household income is obtained by dividing total household income by the total number of households. A small number of very high incomes pulls the mean above the median, so the two will never match unless the distribution is symmetric. The $87,460 is the median, not the mean.

BLS’s OEWS program computes a mean wage by dividing summed wages by total employment for an estimation cell, and it separately produces median wage estimates from the same microdata. OEWS wages are straight-time, gross pay, exclusive of premium pay—so overtime, tips, and bonuses are stripped out before the mean or median is calculated. CES average hourly earnings are also means, reported as arithmetic averages of hourly pay for the jobs in the sample.

The household median and the job mean are built from different populations, on different income concepts, using different statistical summaries. They disagree by design.

The Census Bureau’s income snapshot

The $87,460 median household number comes from the Census Bureau’s 2025 income report, and the income concept behind it matters more than people assume.

The report publishes estimates based on money income and on post-tax income. Money income includes pre-tax earnings, interest, dividends, rents, Social Security, public assistance, and similar cash receipts. Post-tax income subtracts federal and state taxes and adds the value of certain noncash benefits. The headline median is the money-income number—$87,460—and it counts the household as the unit. The mean household income for the same population would be higher because a few top earners drag the average upward.

A reader who looks at that $87,460 and tries to compare it with a pay stub is already making two mistakes: they are comparing a household pool to an individual paycheck, and they are ignoring the distinction between money income and a wage rate. Census income captures every earner in the house and every source of cash a household can tap. Payroll surveys capture straight-time wages for individual jobs.

BLS wages: the CES and OEWS systems

The wage numbers that journalists and analysts most often place next to the household median come from the BLS, and they have their own internal rules.

The CES production and nonsupervisory average hourly earnings figure is a mean derived from payroll records. It covers private-sector employees who are not managers, as defined by the B-8a footnote—production workers in goods-producing industries and nonsupervisory workers in service-providing ones. Benefits, irregular bonuses, and most tips are excluded. The figure is a per-job, not a per-person, statistic.

OEWS publishes employment and wage estimates by occupation for wage and salary jobs in nonfarm establishments. Its mean wage is the summed straight-time gross pay for an occupation divided by employment in that occupation. The program also calculates percentile estimates, so a median wage for, say, registered nurses is available. But that median is still a job-based figure. It does not count household-level resources, it does not include non-wage income, and it measures only wages from employment in establishments, not self-employment or gig work.

When the CES average hourly earnings series and the Census household median sit on the same news page, they are describing two different economies, even though both carry the word “earnings.”

Matching the question to the measurement

The rule is simple but rarely articulated clearly: household medians answer questions about living standards, job-based figures answer questions about pay.

If the question is “What does a typical household live on?” the Census median household income is the right tool. It captures the resources that people actually pool, and the median stops a few multimillion-dollar households from distorting the picture. If the question is “What does this occupation pay?” OEWS median wages or the CES average earnings series are the proper references. Using a household median to judge occupational pay is like measuring the temperature of a house with a thermometer stuck in one room and the windows open.

The mean-median distinction matters inside the job series, too. OEWS publishes both mean and median wages, and the gap between them is a rough inequality signal within an occupation. The household series publishes a median, and researchers who want a household mean are usually studying the distribution’s tail. The worst analytical error is treating a household median and a job mean as though they are interchangeable, then writing a headline about how pay has stalled or surged. They are not interchangeable.

Where the data leaves us guessing

Several moving parts remain harder to document from the agencies’ own primary-source pages than they ought to be.

The effect of multiple earners and household size on the household median is real—a rise in the number of earners per household can lift the median without any individual earning a cent more—but a single, cleanly stated Census passage explaining exactly how that mechanism operates in the CPS ASEC is hard to find. The Census Bureau publishes separate tables for full-time year-round workers, yet a side-by-side comparison with the all-households table requires historical footnotes that are not gathered in one place.

Inflation adjustment is another gap. The household income figure is released in nominal dollars, and users who want a real comparison with wage series must pick a deflator themselves and note which one they used. The BLS does not publish a page that maps its real wage indexes directly onto the Census income concept. State-level household income tables exist, but aligning them with OEWS state occupational wage medians requires reconciling different reference periods and unit definitions—work that the publications themselves do not do for you.

The definitions are public. The machinery is sound. But the stitching between the two statistical universes is left to the user. That is not a flaw so much as a warning: the two numbers disagree because they serve two different analytical jobs, and no amount of averaging will fix that. One is a household median; the others are job-based means or medians. The question that matters is not which number is right, but which unit fits the question you are actually asking.