Does your city change your pay band? Barely.

Starting point

The job sets the band, and the range is enormous

Across 369 occupations the band ratio runs from 1.24× to 3.01× — a 2.4-fold difference in how spread out the middle of the market is. Half of all occupations fall between 1.44× and 1.72×. The narrow end is not low-paid work: pharmacists earn a median of $140,910 and have one of the tightest bands in the country. It is work where the rate is set by something other than a conversation — a licence, a union schedule, a federal pay table.

That half of the question has its own write-up, with the full ranking, the licensed professions that pay well and flex little, and what it means for an actual counter: which jobs have room to negotiate. It reports the same gap as a share of the median rather than as a ratio, so its numbers run on a different scale — 0.22–1.26 where these run 1.24×–3.01×. Either is fine; mixing them in one sentence is not.

What follows is the part that write-up does not cover: whether any of this depends on where you happen to work.

It is worth knowing what the published band is not the same as, either. A growing number of states now require employers to put an expected salary range in the job posting — Colorado from January 2021, with New York, California and Washington following by January 2023. Arnold, Quach and Taska found that the Colorado law raised the share of postings carrying salary information by 30 percentage points, and that among employers posting before and after, posted salaries rose about 3.6%. A posted range is one employer's stated range for one role. The band on this page is the realised distribution of what an entire occupation is actually paid across every employer in an area. They will rarely match, and the second is the harder number to argue with.

Finding 1

Where you live barely moves it

Comparing metro areas honestly takes one correction. The survey publishes far more occupations for New York than for a rural area, and the extra ones are the professional jobs with wide bands — so ranking places on "all occupations published there" measures the occupation mix, not the pay setting. We instead fixed a basket of 23 occupations published in at least 90% of areas — nurses, accountants, electricians, truck drivers, office supervisors — and ranked 472 areas on that basket alone.

The correction matters, and the cleanest illustration is the top of the uncorrected ranking: Midland, TX comes first in the country at 1.60× — an area for which the survey publishes just 94 occupations, against 350 for New York-Newark-Jersey City. It is not that pay is unusually spread out there; it is that the occupations thin enough to go unpublished are the narrow-banded ones. On the fixed basket the entire country collapses into a very narrow range.

Narrowest area

1.34×

Median area

1.52×

Widest area

1.68×

Widest ÷ narrowest

1.25×

Every one of the 472 comparable areas in the United States sits between 1.34× and 1.68×, and the middle half sits between 1.49× and 1.55×. The largest quarter of areas by employment average 1.53×; the smallest quarter average 1.51×. That is a difference of 0.026 — next to a spread of 1.77 across occupations, it is nothing.

This lines up with what Hazell, Patterson, Sarsons and Taska found on job-posting microdata: 40–50% of a firm's posted wages for a given job are identical across its locations, and the firm rather than the location explains most of the wage variation within a job. Our aggregates cannot test their mechanism — but a country where place barely changes band width is what their result predicts.

One thing this is not: a claim that your own metro is irrelevant. It says the typical area is unremarkable once occupation mix is held constant. Fix the occupation and areas do separate — our companion analysis of which jobs have room to negotiate finds software developers' bands running roughly twice as wide in Austin as in San Jose, and reads that as base pay being the standardised part of the package wherever equity dominates. Finding 2 below is the same effect measured across every occupation.

Finding 2

Individual job-and-place cells still differ a lot

Everything above is about averages, and averages hide the cells. Take one occupation and look at it area by area and the band ratio moves sharply — Dentists, General runs from 1.08× in the narrowest area to 4.59× in the widest. The typical area is unremarkable; specific job-and-place combinations are not.

Occupations published in at least 100 areas. The ranking uses each occupation's 90th- versus 10th-percentile area, so a single freak cell cannot buy a place on this list; the two columns then show the actual extremes. The most stable occupations — pharmacy technicians, bookkeeping clerks, medical assistants — barely move at all.

Take the whole table

Every figure on this page comes from one file, and the file is yours. 64,113 occupation-by-area rows plus 398 national rows, with the published percentiles, all four derived band measures, and the flag columns used here — so you can reproduce these numbers or disagree with them on the same data.

Methodology

Source. The Occupational Employment and Wage Statistics survey run by the U.S. Bureau of Labor Statistics, May 2025 reference period, released 2026-05-15. That is the survey's reference period, not its publication date — the May 2025 estimates are the most recent that exist. We use the published annual 10th, 25th, 50th, 75th and 90th percentiles together with employment counts. Using this survey to describe the shape of the wage distribution rather than its level is not new — BLS did it itself in the Monthly Labor Review in 2014, comparing OES-derived inequality measures against the Current Population Survey. What is new here is doing it per occupation and per area, and publishing the resulting table.

Measures. For each occupation-and-area cell we compute the band ratio (p75 ÷ p25), the decile ratio (p90 ÷ p10), the dollar width (p75 − p25) and the width as a share of the median. The band ratio is the headline because it describes the ordinary middle rather than the tails, and because a ratio is comparable between a $40,000 job and a $400,000 one.

Exclusions. Two, applied at different stages, and only one of them is reversible from the file. First, before this file is written: 3,840 of the 67,953 cells that clear the employment and median thresholds report two adjacent percentiles as an identical number, which is the signature of interval-scale collection (see Limitations). Those are dropped in the dataset build, because the counter-offer calculator built on the same data must not anchor on one either. They are therefore in no row of the CSV, and the 64,113 cells it does carry all have strictly increasing percentiles. Second, inside this study: 3,297 cells belonging to SOC residual categories — the "All Other" buckets — because those pool dissimilar jobs by construction, so a wide band there is a fact about federal bookkeeping rather than about any job. Without this exclusion the two widest occupations in the country are both residual buckets; these are flagged rather than deleted, so anyone who disagrees can put them back. That leaves 60,816 cells and 369 national occupations in the analysis.

Comparing areas. Areas are ranked only on a fixed basket of 23 occupations published in at least 90% of areas, and only areas publishing at least 90% of that basket are ranked at all (472 of 528). Ranking on all locally published occupations instead measures occupation mix — we report both so the difference is visible. The comparison ranking labelled "uncorrected" above is uncorrected for occupation mix only; the two exclusions apply to it as well.

Reproducibility. Two scripts and one manual step. BLS blocks automated retrieval of its bulk files by policy, so the oe.* time-series files are downloaded once through a browser. Then scripts/build-comp-dataset.mjs converts them into the wage dataset and applies the thresholds and the artefact filter, and scripts/build-payband-study.mjs emits both the figures on this page and the CSV from it. Every number above is read from that output; none is typed by hand.

Limitations

These are the reasons not to over-read the numbers above. They are listed here rather than in a footnote because a measure of pay dispersion is easy to misuse.

1. This is not a measure of negotiation room. A p25–p75 gap contains at least four things: differences between firms, differences between workers' seniority and specialisation, differences in industry mix inside one occupation code, and whatever bargaining occurred. Published aggregates cannot separate them, and the work cited below suggests the first is the largest. Treat a wide band as a wide band.
2. Base pay only. OEWS measures straight-time gross pay. Equity, RSUs and nonproduction bonuses are excluded from the survey entirely. For the occupations flagged with an asterisk above — executives, technology, finance, sales — this omits precisely the part of compensation that moves most.
3. Wages are collected on an interval scale, not as exact rates. When many of an occupation's workers fall into one interval, adjacent published percentiles collapse onto the same value and the ratio stops meaning anything. That is not hypothetical: 3,840 of 67,953 cells (5.7%) show at least one collapsed pair. One metro reports an identical 10th and 25th percentile that would otherwise produce a fictitious 13× band; another publishes a median identical to its 75th percentile at $618,090 against a 25th percentile of $211,400. Unlike the residual buckets these are removed rather than flagged, so they cannot be put back from the CSV — the filter that drops them is in scripts/build-comp-dataset.mjs and runs on the public BLS files, so a reader who disagrees can reproduce them from the source.
4. No published reliability measure for percentiles. BLS publishes sampling-variability estimates for employment and mean wages, not for the percentile estimates this study is built on. So a difference between two similar band ratios cannot be called statistically significant, and we never do.
5. Each estimate pools three years. An OEWS cell combines six semiannual panels collected over three years, with wages from earlier panels aged forward using the Employment Cost Index. Measured dispersion therefore contains some imperfectly-removed wage growth — which will mechanically widen the band for fast-growing occupations more than for flat ones.
6. One release, no trend. BLS discourages using OEWS as a time series, because occupation coding, area definitions and survey methods all change between releases. Nothing here is a change over time, and we make no claim about one.