Real Wages Fell 0.3 Percent And 0.1 Percent In The Same August

One Month, Two Divisors

On September 11 the Bureau of Labor Statistics published two tables that sit one above the other on the same page, describe the same calendar month, and disagree about how much a dollar was worth in August 2026. Table A-1 of the U.S. Bureau of Labor Statistics, Real Earnings, Table A-1 reports current-dollar average hourly earnings for all employees on private nonfarm payrolls at $37.75 and the constant-dollar version at $11.30. Table A-2 reports the same two quantities for production and nonsupervisory employees at $32.53 and $9.94.

Those four cells contain more than they appear to. Real earnings in this release are nothing but current earnings divided by a price index, so the division can be run backwards. Divide $37.75 by $11.30 and you recover 3.340707, which is a price level of 334.07 on the 1982-84 base. Divide $32.53 by $9.94 and you recover 3.272636, a price level of 327.26. Same month, same country, two price levels 6.8 index points apart. The gap is 2.08 percent.

This is not a rounding artifact and it is not an error. It is the release working as designed, and the design is stated in the footnotes. Footnote 1 to Table A-1 reads: "The Consumer Price Index for All Urban Consumers (CPI-U) is used to deflate the earnings series for all employees." Footnote 2 to Table A-2 reads: "The Consumer Price Index for Urban Wage Earners and Clerical Workers (CPI-W) is used to deflate the earnings series for production and nonsupervisory employees." Two tables, two deflators, declared on the page.

BLS does not hide the choice. The trap is that two tables produce two correct headline numbers for the same concept, and whoever quotes one is quoting a denominator picked for a population that is not the one in the numerator.

Where the Second Price Level Comes From

The CPI release published the same day carries both index numbers outright. Table 1 of the CPI news release puts the unadjusted all-items CPI-U for August 2026 at 334.980. The release text states that the CPI-W "increased 3.5 percent over the last 12 months to an index level of 328.481 (1982-84=100)." Two published index levels for August 2026, 6.499 points apart, a gap of 1.98 percent. The implied levels recovered from the Real Earnings tables run below both published levels because those tables use seasonally adjusted series, but the spread is the same spread.

The two indexes are not independent surveys. They price the same items from the same outlets in the same month. What differs is the weighting, because the target population differs. The U.S. Bureau of Labor Statistics, Handbook of Methods, Consumer Price Index concepts defines the narrower population with two conditions, and both conditions matter: the CPI-W "is based on the expenditures of urban households included in the CPI-U definition that also meet two additional requirements: more than one-half of the household's income must come from clerical or wage occupations, and at least one of the household's earners must have been employed for at least 37 weeks during the previous 12 months."

That is a household test, not a job test. It turns on the share of household income from a class of occupations and on a 37-week employment history. A household where a nurse earns 45 percent of the income and a self-employed contractor earns 55 percent fails the first condition; a household whose only earner worked 30 weeks fails the second. The result, stated in both the concepts page and the August release, is that the "CPI-W population represents approximately 30 percent of the total U.S. population and is a subset of the CPI-U population," against a CPI-U population "representing over 90 percent of the population."

The Numerator Is Counted Differently From the Denominator

Now look at what the wage series in Table A-2 actually covers. Footnote 1 to that table says the data "relate to production employees in mining and logging and manufacturing, construction employees in construction, and nonsupervisory employees in the service-providing industries. These groups account for approximately four-fifths of the total employment on private nonfarm payrolls."

Four-fifths of private payroll employment in the numerator; a household group near 30 percent of the population in the denominator. The two shares are not comparable because they are not shares of the same thing, and that is the point. The U.S. Bureau of Labor Statistics, Handbook of Methods, Current Employment Statistics concepts is explicit about its unit: "Employment is the measure of the number of jobs, as opposed to the number of employees ... Persons working at multiple worksites are counted as employed at each worksite." The wage series counts jobs. The price index weights households. A worker holding two nonsupervisory jobs enters the numerator twice and the denominator once, if at all.

The occupational list on the CES side is wider than the phrase "wage earners and clerical workers" suggests. It includes "office and clerical workers, repairers, salespersons, operators, drivers, physicians, lawyers, accountants, nurses, social workers, research aides, teachers, drafters ... and other employees at similar occupational levels." Physicians, lawyers and accountants below the working-supervisor level are production and nonsupervisory employees for CES purposes. Whether their households clear the CPI-W income-share screen is a question the CES documentation never asks.

Be precise about what the agency does and does not claim. The Real Earnings Technical Note states the pairing plainly and adds that "there are several characteristics of the series presented in this release that limit their suitability for some types of economic analyses." What no sentence in the Technical Note, in either table's footnote block, in the CPI concepts page or in the CES concepts page asserts is that the two universes are the same people. The documentation is not silent about the pairing; it is silent only about congruence, and the CPI questions-and-answers page speaks only to averaging: "BLS bases the market baskets and pricing procedures for the CPI-U and CPI-W populations on the experience of the relevant average household, not of any specific family or individual."

How Much of the Headline Gap Is the Deflator

Here the trace has to finish rather than stop at the interesting part. The two published 12-month real figures for August 2026 are minus 0.3 percent for all employees and minus 0.1 percent for production and nonsupervisory employees. Anyone who has just learned that two deflators are involved will assume the deflators caused the spread. They did not.

The nominal side comes from the U.S. Bureau of Labor Statistics, Employment Situation for August 2026, which reports that average hourly earnings for all employees "rose by 10 cents, or 0.3 percent, to $37.75. Over the year, average hourly earnings have increased by 3.1 percent." Table A-2 puts the 12-month nominal gain for production and nonsupervisory employees at 3.3 percent. On the price side, CPI-U rose 3.4 percent over 12 months and CPI-W rose 3.5 percent.

Subtract. For all employees, 3.1 minus 3.4 is minus 0.3, which is exactly the published figure. For production and nonsupervisory employees, 3.3 minus 3.5 is minus 0.2, against a published minus 0.1; the difference is the rounding that survives when BLS divides index-level series rather than subtracting rounded percentages.

Now hold the deflator fixed. Put CPI-U under both wage series and you get minus 0.3 and minus 0.1. Put CPI-W under both and you get minus 0.4 and minus 0.2. Either way the spread between the two series is 0.2 percentage point. The spread does not come from the deflator choice at all. It comes from the fact that hourly pay in the production and nonsupervisory group grew 0.2 point faster than pay across all employees, and the faster-growing series happened to be matched with the faster-growing price index, which shaved the advantage rather than creating it.

What the Deflator Choice Does Move

The deflator choice does not move this month's growth comparison. It moves the level, and it moves it by a great deal more than 0.1 point. Take the published $32.53 and divide it by the price level embedded in Table A-1 instead of the one embedded in Table A-2. That is $32.53 divided by 3.340707, or $9.74. The published figure under CPI-W is $9.94. The same paycheck, the same month, the same agency, and twenty cents of constant-dollar purchasing power separate the two answers. Run it the other way and all-employee real hourly earnings would print $11.54 under CPI-W instead of the published $11.30.

Twenty cents on $9.94 is 2.0 percent, and it is not a one-month wobble. It is the accumulated distance between two index levels that have drifted apart since the 1982-84 base, because the two baskets weight the same items differently. Switching deflators mid-series would insert a 2 percent step that has nothing to do with wages.

The Narrow Index Has a Second Job

CPI-W is not a specialist tool kept for one table of a wage release. It is the index that sets federal benefit escalation. The Social Security Administration, Latest Cost-of-Living Adjustment states that "COLAs are based on increases in the Consumer Price Index for Urban Wage Earners and Clerical Workers (CPI-W)," and specifies the window with the conditional intact: "A COLA effective for December of the current year is equal to the percentage increase (if any) in the CPI-W from the average for the third quarter of the current year to the average for the third quarter of the last year in which a COLA became effective." If the rounded increase is zero, "there is no COLA for the year."

So the 328.481 that deflates production-worker wages in Table A-2 is also one of the three monthly readings that will decide a benefit adjustment. July and August 2026 are in. September completes the third-quarter average. We have taken apart the arithmetic of that window before in our piece on why the COLA formula often misses what beneficiaries feel, and the mismatch traced there is the same species as the one here: an index weighted for one population doing duty for another.

The CPI documentation is careful here in a way commentary is not. The questions-and-answers page states that "the CPI frequently is called a cost-of-living index, but it differs in important ways from a complete cost-of-living measure," and adds that "the CPI does not necessarily measure your own experience with price change." That is the agency saying one index cannot be the right deflator for every purpose, which is why Real Earnings carries two.

A Third Price Level, From a Different Agency

A third answer to what August 2026 prices were does not come from BLS. The Bureau of Economic Analysis, Personal Consumption Expenditures Price Index page lists its current release as August 26, 2026, covering July 2026 at plus 3.7 percent from a year earlier, with the next release dated September 30, 2026. That release will put an August 2026 PCE price index reading beside the CPI-U's 3.4 percent and the CPI-W's 3.5 percent.

The PCE index is not a third slice of the same population. It is a different aggregate with a different scope, including expenditures made on households' behalf by third parties, and it is the index the central bank has committed to. The Federal Reserve, Statement on Longer-Run Goals and Monetary Policy Strategy keeps the qualifier in place: "The Committee reaffirms its judgment that inflation at the rate of 2 percent, as measured by the annual change in the price index for personal consumption expenditures, is most consistent over the longer run with the Federal Reserve's statutory maximum employment and price stability mandates."

That leaves three official price levels for one month, each correct for its purpose: CPI-U for the broad real-wage table, CPI-W for the production-worker table and for benefit escalation, PCE for the policy target. Nobody reconciles them because there is nothing to reconcile. The failure mode is quoting a real wage change without knowing which question its denominator answered. The same structure appeared when we tracked which portion of a homeowners premium enters the CPI and which enters GDP: one payment, two statistical systems, two defensible treatments.

What Would Invalidate This

Three things would break this reading. First, a month in which CPI-U and CPI-W diverge by far more than the 0.1 point seen here: the deflator, not composition, would then own the spread. Second, the January seasonal-adjustment revision described below, which can move the nominal series and its deflator together, so the 0.2-point gap is not fixed. Third, we did not test whether CPI-W households and production and nonsupervisory jobs actually overlap; no document I read measures that.

What to Watch Next Week

The BLS release calendar puts the September 2026 Employment Situation on October 2, 2026, at 8:30 a.m. That release delivers the numerators for both real-wage series: a new average hourly earnings level for all employees and a new one for production and nonsupervisory employees. It delivers neither denominator. September CPI is scheduled for October 14, 2026, which is when Real Earnings for September appears and when the two constant-dollar series can be computed at all. For twelve days the nominal wage will be public and the real wage will not exist, and any real-wage claim made in that window is using an assumed deflator.

Three checks in that window. First, whether the 0.2-point nominal gap between the two series persists; if production and nonsupervisory pay again outruns all-employee pay, the composition story is the one to follow. Second, average weekly hours for the production and nonsupervisory group, because weekly earnings in Table A-2 outran hourly earnings over the year, and that difference is hours, not pay rates. Third, on September 30, the BEA release adding the August PCE reading, the first clean three-way comparison for one month.

One calendar note catches people every January. The CPI supplemental files carry this caution: "Data in archived CPI Supplemental files may have been revised in subsequent editions. In particular, seasonally adjusted indexes and percent changes are revised with the release of January data each year for the preceding five years." Both Real Earnings tables and both deflators are seasonally adjusted, so today's monthly real-wage changes are revisable five years back. Both tables carry entries flagged preliminary.

Concrete Framework

A rule that survives contact with the release: before repeating a real-wage number, recover its denominator. Divide the current-dollar cell by the constant-dollar cell in the same row. The quotient is the price level that table used, and it names which population's basket is doing the deflating. For August 2026 it is 334.07 in Table A-1 and 327.26 in Table A-2.

Then ask three questions in order. Which wage series is in the numerator, all employees at $37.75 or production and nonsupervisory at $32.53? Which index is in the denominator, which the footnote names. And is the claim about a level or a change? That decides how much the deflator matters: for the August 2026 level the swap is worth about 2 percent, twenty cents on $9.94; for the twelve-month change it is worth about 0.1 point, and the 0.2-point spread between the two headlines is wage composition.

The lesson is narrower than "indexes differ." When an agency pairs a numerator and a denominator built on different units of count, jobs on one side and households on the other, the pairing is a judgment, and the release records it in a footnote rather than the headline. The footnote is where the definition lives. Read it before the number.