The simple version

One of these two measures holds the mix of jobs fixed on purpose. Only a change in what employers pay for the same work can move it.

The other adds up reported pay and divides by reported hours. Change who is working and that number moves on its own.

So the two can point opposite ways in the same quarter without either being wrong. They were never measuring the same thing.

The numbers

• The Employment Cost Index for the third quarter of 2026, published October 30, 2026: total compensation for civilian workers rose 3.4 percent over the 12 months ending September 2026, not seasonally adjusted. Wages and salaries rose 3.1 percent and benefit costs rose 3.9 percent over the same 12 months.
• Total compensation for civilian workers rose 0.8 percent over the three months ending September 2026, seasonally adjusted. Wages and salaries rose 0.7 percent and benefit costs rose 0.9 percent over those three months.
• The Bureau of Labor Statistics describes the Employment Cost Index as measuring the change in the cost of labor, free from the influence of employment shifts among occupations and industries.
• The agency's Monthly Labor Review says the payroll survey earnings figures reflect shifts in the number of employees between relatively high-paid and low-paid work, along with premium pay for overtime and late-shift work and changes in earnings within individual establishments.
• The same article says those earnings figures exclude items such as employee benefits, irregular bonuses and commissions, retroactive payments, and the employer's share of payroll taxes, so they do not represent total compensation costs. The Employment Cost Index covers benefit costs as well as wages and salaries.
• The agency has attributed a rise in its own earnings series to this effect, and the series it names is the one above. An Economics Daily post dated July 17, 2020 reports that real average hourly earnings for all private nonfarm employees increased 4.3 percent from June 2019 to June 2020, that this combined with a 0.3 percent increase in the average workweek to lift real average weekly earnings 4.6 percent, and that the increases in real average hourly and weekly earnings largely reflect the substantial job loss over the year among lower-paid workers.
• Those 2020 figures are inflation-adjusted and the Employment Cost Index changes above are not, so the two sets of percentages are not comparable and nothing here compares them. The post is cited for the effect it names on average hourly earnings, which is the measure this article contrasts with the index.
• Fixed employment weights in the Employment Cost Index are updated periodically to reflect the current industry and occupational employment distributions, so the mix is held still between updates rather than forever.
• Figures read at the Bureau of Labor Statistics on October 30, 2026.

Why two pay measures disagree

Pinning the mix is a deliberate design choice, not an accident. The agency sets employment weights, leaves them alone, and then asks employers what the same jobs cost this quarter.

The paycheck average has no such protection and was never meant to. It reports what was actually paid out, so every change in who is working flows straight into it.

That is why the two answer different questions. One tells you whether employers raised what they pay for a given job, and the other tells you what the typical paycheck now looks like.

The actual math on a rising average with no raises

Here is an invented example built to isolate the effect, not a real workplace. Take 100 people: 70 paid $18 an hour and 30 paid $45. The simple average is $26.10.

Now five of the people paid $18 lose their jobs and nobody else is touched. The group is 65 at $18 and 30 at $45, which averages $26.53 across 95 people.

That is a rise of 1.6 percent. Not one pay rate changed, and a measure holding the original 70 to 30 split would show zero. The arithmetic here is ours.

What this means

Match the measure to the question before you quote it. If you want to know whether employers are paying more for the same work, the mix-frozen measure is the one built for that.

If you want to know what people are actually taking home on average, the paycheck average answers it, with the caveat that a shift in who is employed can move it on its own.

The transferable habit is to ask what a published average holds constant. Two honest measures of the same subject can disagree entirely because of what each one refuses to let move.

What this is NOT

This article explains how the Employment Cost Index differs from average hourly earnings, using the Bureau of Labor Statistics figures for the third quarter of 2026, published October 30, 2026, and the agency's own technical notes and publications as the worked example. It is not a forecast of the Employment Cost Index for any later quarter. It is not advice about anyone's pay, job, or negotiating position, and we do not give personalized advice. It is not a claim that the real figures above are explained by any shift in the mix of jobs; the illustration in the math section is invented to show the mechanism in isolation, and the arithmetic in it is ours. It says nothing about whether pay is rising fast enough or too fast, nothing about what any employer should pay, and nothing about the Federal Reserve or what any figure implies for interest rates. We have not compared the two measures over any historical period, because that comparison is outside what we read here. The 2020 Economics Daily post cited above reports inflation-adjusted figures while the Employment Cost Index changes above are not inflation-adjusted, so we cite that post only for the effect it names and do not compare its percentages with the index figures. Both measures are the agency's own and we take no view on which is better. This is not a political endorsement or criticism of anyone.

Sources

• U.S. Bureau of Labor Statistics, Employment Cost Index Technical Note, 2026 Q3 results. The description of what the index measures and the section on fixed employment weights. Read October 30, 2026.
• U.S. Bureau of Labor Statistics, Employment Cost Index News Release, 2026 Q3 results, published October 30, 2026. The 12-month and 3-month changes for civilian workers. Read October 30, 2026.
• U.S. Bureau of Labor Statistics, Monthly Labor Review, One hundred years of Current Employment Statistics: busting CES myths, October 2016. What the payroll survey earnings figures reflect, what they exclude, and that they do not represent total compensation costs. Read October 30, 2026.
• U.S. Bureau of Labor Statistics, The Economics Daily, real weekly earnings up 4.6 percent for year ending June 2020, reflecting loss of lower-paying jobs, July 17, 2020. The agency attributing a rise in an earnings series to job loss among lower-paid workers. Read October 30, 2026.