Start with what each measure is built to answer. Average hourly earnings answers "what is the average hourly pay of people currently on payrolls?" It is a ratio of total payroll to total hours, so its value depends on who is employed as well as on what they are paid. The employment cost index answers "what does an hour of labor cost an employer, holding the job mix steady?" The Bureau of Labor Statistics describes it as measuring "the change in the hourly labor cost to employers, independent of employment shifts among occupations and industry categories," and its estimates cover "total compensation, including wages and salaries and benefits." The Atlanta Fed tracker answers "what happened to the pay of a typical individual worker over the past year?" by following the same respondents twelve months apart and reporting a median rather than an average.
The Federal Reserve Board leads with the employment cost index. In its Monetary Policy Report of July 2026 it reports compensation growth for private-sector workers "as measured by the employment cost index" first, then treats the other two as further measures, calling average hourly earnings "a less comprehensive measure of compensation" and noting that the tracker "reports the median 12-month wage growth of individuals responding to the Current Population Survey." The Report does not spell out why the index is more comprehensive, but the Bureau of Labor Statistics does: total compensation in the index covers "employer costs for wages and salaries and for employee benefits," and average hourly earnings covers wages alone.
The composition problem is the reason all this matters, and the clearest statement of it comes from the agency that publishes the affected series. In its Current Employment Statistics highlights for April 2020, the Bureau of Labor Statistics wrote: "Average hourly earnings for all leisure and hospitality workers rose $1.14 in April to $18.00 as lower-earning workers were removed from payrolls." That is the statistical agency saying, in its own publication, that an average rose because low-paid workers had lost their jobs. Nobody in that industry received a raise on account of it. Any use of average hourly earnings as a summary of "what workers' pay did" inherits this problem in both directions: the average rises when the low-paid are laid off and falls when they are hired back.
Who indexes to what is worth knowing, because the phrase "wage growth" turns up in Social Security calculations and does not refer to a Bureau of Labor Statistics series there. Social Security indexes a worker's past earnings to its own administrative series. The governing regulation defines the "average of the total wages" for years after 1990 as "all remuneration reported as wages on Form W-2 to the Internal Revenue Service for all employees for income tax purposes," plus certain deferred compensation contributions, "divided by the number of wage earners," and directs that figures for recent years be published in the Federal Register. That series, commonly called the national average wage index, is computed from tax reporting rather than from a household or establishment survey, and it is what drives the indexing of career earnings and several annually adjusted Social Security amounts. The mechanics of that indexing belong to average indexed monthly earnings and to the cost-of-living adjustment.
One further split is worth knowing about rather than re-deriving. The Atlanta Fed tracker publishes separate series for job switchers and job stayers, and the definition of a switcher is broader than it sounds. That distinction, and what it does and does not measure, is covered under job hopping.
