Methodology

Understanding where the numbers come from helps you judge how much weight to put on them. The Occupational Employment and Wage Statistics program is not a poll of workers — it is a survey of employers. Twice a year, state workforce agencies working with the BLS send questionnaires to a sample of business establishments, asking them to report how many people they employ in each occupation and what those people are paid, by wage band.

Each published estimate combines six semiannual survey panels collected over three years — roughly 1.1 million establishments covering about 57% of U.S. employment. Older panels are adjusted to current wage levels before being combined. This rolling design is what makes estimates available for ~800 occupations in small geographic areas, but it also means a published figure is best understood as a three-year centered estimate, not a snapshot of last month.

Three consequences worth knowing:

• Employer-reported wages exclude self-employment income, most equity compensation, and overtime premiums — so top-end pay in equity-heavy fields (tech, finance) is understated relative to total compensation.
• In fast-moving markets, new-hire offers can run ahead of the survey, since the data reflects all incumbents, including people hired years ago. Treat OEWS as a reliable floor and structure, and supplement with live postings for the leading edge.
• Small occupation-area cells are suppressed or noisy. When an occupation has very few workers in a state, the BLS either withholds the estimate or publishes one with a wide error band. Where data is missing on Wage Atlas, this suppression — not a site error — is usually the reason.

Our trend charts reach back to the late 1990s, which creates a problem the casual reader never sees: the federal government has changed its occupational classification system multiple times along the way. The 2010 revision of the Standard Occupational Classification split some occupations, merged others, and created new ones; the 2018 revision did the same again. "Software Developers," for example, was reorganized in 2018, and several healthcare and education occupations were redefined.

To build continuous histories, we map older SOC 2010 codes onto the current SOC 2018 taxonomy using the official BLS crosswalk tables. Where an old occupation maps cleanly to one new code, the series connects directly. Where an occupation was split or merged, a perfectly clean join is impossible — we connect the dominant successor and treat the transition years with caution. Practically, this means:

• Long-run trends for stable occupations (nurses, electricians, truck drivers) are highly reliable across the full history.
• Trends that span a reclassification boundary (notably 2009-2012 and 2018-2021) for restructured occupations can show artificial jumps that reflect the definition change, not the labor market. We flag the affected series where feasible.

For a full explanation of how the classification system works, see our SOC code guide.

We would rather you trust the data for the right reasons than for the wrong ones. These are the known limits of what Wage Atlas can show, stated plainly:

• No experience breakdown. OEWS does not record tenure or years of experience. Percentiles are a useful proxy — entry-level workers cluster near the 10th-25th percentile — but they are a proxy, not a measurement.
• Wages only, not total compensation. Health benefits, retirement matching, bonuses paid irregularly, and equity are out of scope. For some occupations these exceed 30% of cash wages.
• No employer-level detail. The data describes occupations in areas, never specific companies. A specific employer can sit anywhere in (or beyond) the published distribution.
• Cost-of-living adjustments are averages. Regional Price Parities reflect the average consumption basket; your personal exposure to housing costs may be higher or lower. See the worked examples in our cost-of-living guide.
• Projections are extrapolations. Where we show growth projections, they extend structural trends and cannot anticipate shocks — recessions, pandemics, or rapid technology shifts.

Our editorial guides — on reading percentiles, interpreting trends, and using the data in negotiations — are written specifically to help you work within these limits.