Salary-range laws have not narrowed the pay band

Mandate vs no mandate

+0.62%

Band width, same 216 occupations

Could be chance

p = 0.50

4,000 random relabellings

Placebo group

−1.40%

Law passed, not yet in force

State-to-state spread

10.7pp

17× the mandate gap

The band ratio is the 75th percentile of annual base pay divided by the 25th, inside one occupation in one state. Across the occupations compared here it sits around 1.44×: where the ordinary middle starts at $100,000 it ends near $144,072, a spread of $44,072. The mandate difference is worth about $893 of that.

The question

"California requires salary transparency, Texas does not"

That was a reader's explanation for why pay is spread out differently in different places, and it is a good hypothesis: if every employer must publish a range, the ranges should converge, and the gap between the person at the bottom of the ordinary middle and the person at the top should close. It is also testable, because the survey publishes the same percentiles for both states.

On the 215 occupations both states publish, California's median band ratio is 1.54× and Texas's is 1.50×. California — the mandate state — is 2.4% wider. The two states in the comment sit next to each other at the wide end of the country, on opposite sides of the law.

One pair of states proves nothing either way, which is why the rest of this page is the same comparison run across every state and every occupation the survey publishes.

Finding 1

Occupation by occupation, the mandate is invisible

Comparing states honestly takes one correction, and it is the correction that decides the answer. Mandate states are the big coastal economies; they employ a different mix of jobs from Mississippi, and professional jobs have wider bands. Ranking states on "all occupations published there" would therefore measure the occupation mix and call the result "the law". So every comparison here is made within an occupation: a fixed basket of 216 occupations published in at least 90% of the 51 jurisdictions — nurses, accountants, electricians, truck drivers, cooks, office supervisors — with each occupation's mandate-state median set against its own no-mandate median, and the per-occupation differences pooled at the end.

The median occupation's band is +0.62% in mandate states (1.48× against 1.43×). Bands are narrower in 91 of 216 occupations — 42%, which is a coin toss. And the direction is the wrong one for the hypothesis: the sign says wider.

Each row is the median across occupations of that group's median band ratio divided by the same occupation's no-mandate median. "Longest" means the law covered at least 4 of the six semiannual panels the survey pools.

The right question about a +0.62% gap is not which direction it points but whether any 9 states picked out of the 42 would have produced something that big. Reshuffling which states carry the label 4,000 times, a gap at least this large comes up 50% of the time (p = 0.50). There is nothing here to explain.

Finding 2

The states without a law yet look more "compliant" than the states with one

A cross-section like this can always be dismissed with one sentence — mandate states are different states — so it needs a group that answers it. There is a natural one. 6 jurisdictions (DE, MA, ME, NJ, VA, VT) had passed a posting law that had not yet taken effect during the survey window. They resemble mandate states in the ways a sceptic worries about — politics, wage level, urbanisation, the kind of legislature that passes this kind of law — and differ in the one way that is supposed to matter.

Their bands are −1.40% against no-mandate states — narrower in 139 of 216 occupations. That is a larger deviation than the mandate group's, in the direction the hypothesis predicts for the law, from jurisdictions where the law was not operating. Whatever produces these small differences, it is not the requirement to publish a range.

Two other checks say the same thing. Measuring the wider p90/p10 spread instead of the middle band moves the mandate figure to +0.68% and the placebo figure to −3.61% — same ordering, same conclusion. And restricting the comparison to the richer half of no-mandate states, on the grounds that mandate states are the high-wage ones, moves the headline to +1.64%. A number that swings this much when the comparison group is redrawn is a description of which states these are, not a measurement of what the law did.

Finding 3

States do differ — the differences just are not law-shaped

Holding occupation constant, band width still runs from 6.5% below the national figure in South Dakota to 4.2% above it in California — a spread of 10.7 points, roughly 17 times the gap the law is supposed to explain. Below, every jurisdiction on that one axis.

Read the two lists side by side and the pattern is not legal. The widest bands in the country belong to California, New York, Texas, Mississippi and Oklahoma — 2 of them mandate states, 3 of them not. The narrowest belong to South Dakota, Vermont, Maine, Wisconsin and North Dakota, not one of which had a posting mandate in force. What those groups share is size and industry structure: big, dense, high-variance labour markets at one end and small, employer-concentrated ones at the other. That is the same conclusion Hazell, Patterson, Sarsons and Taska reach from job-posting microdata, where the firm rather than the location explains most within-job wage variation.

The honest version

What would actually answer this — and what it found

A cross-section of paid wages is the wrong instrument for a law about adverts. The right one is posting microdata with a before-and-after design around the effective dates, and it exists: Arnold, Quach and Taska study Colorado's January 2021 law with a difference-in-differences on job postings. They find employers increased the share of postings carrying salary information by 30 percentage points and that wages rose 1.3–3.6% across three datasets — and, in the same abstract, "no impacts on pay dispersion". The measure this page cannot identify is the measure the paper that can identify it reports as null.

There is a second reason a state-by-state comparison of paid wages is doomed here, and it is worth stating plainly because it is what most commentary misses: the treatment leaks. A national employer that must publish a range for a Denver role usually publishes it for the same role everywhere, and a remote posting is typically covered by the strictest state it can be performed in. The "control group" reads mandated ranges too.

None of this makes transparency laws ineffective — it makes this outcome the wrong scoreboard. The literature's findings sit elsewhere: Cullen and Pakzad-Hurson model how transparency shifts bargaining power in equilibrium, and Cullen's review reports that transparency between coworkers inside a firm narrowed coworker pay gaps while leading employers to bargain harder and average wages to fall. That is a different mechanism from a posted range in a job advert, and this page does not test it.

A cleaner design on this data would exploit the staggered effective dates — Colorado 2021 through Vermont 2025 — as a panel. We are not running it. Each OEWS estimate pools six semiannual panels over three years, so a "2025 law" state has one panel in six of exposure and the treatment date is smeared across the outcome; a difference-in-differences on that would look rigorous and mean very little. If somebody outside this project wants to review the design, the data below is the whole input.

Take the whole table

35,446 occupation-by-state rows: the published percentiles, four derived band measures, and each jurisdiction's posting requirement with its effective date and how many of the survey's six panels it covers. Every exclusion this study makes is a flag rather than a deletion, so you can undo any of them and disagree with us on the same file.

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 — these are the most recent estimates that exist. We use the published annual 10th, 25th, 50th, 75th and 90th percentiles at state level, together with employment counts.

Legal coding. A jurisdiction counts as a mandate only if the range must appear in the posting itself; owing a range on request, before an offer or after an interview is coded separately. The table was verified against law-firm trackers on 2026-08-06 rather than recalled, and the verification changed the answer twice: Maine is routinely listed as a mandate state but its law took effect 2026-07-29, fourteen months after this survey's reference period, and Washington, D.C. has been one since 2024-06-30 and is routinely left out. Both corrections are baked into the groups below.

"Panels of 6" is how many of the six semiannual collection panels behind a May 2025 estimate post-date the law. Zero means the estimate contains no post-law data at all, which is what puts DE, MA, ME, NJ, VA, VT in the placebo group. Connecticut, Nevada and Rhode Island owe a range on request, after an interview or at hire and are reported separately throughout.

Comparison. Within occupation, always. A fixed basket of 216 occupations published in at least 90% of jurisdictions; for each, the median band ratio among one group against the median among no-mandate states; the reported figure is the median of those per-occupation ratios. An occupation enters a comparison only when both sides have enough jurisdictions for a median to mean anything (at least 3 and 15 respectively).

Exclusions. All flagged rather than deleted in the published file. 2,697 cells in SOC residual categories (the "All Other" buckets), which pool dissimilar jobs by construction; 1,990 cells where two adjacent published percentiles are identical, the signature of interval-scale collection; 6,316 cells below 200 employees, where a percentile describes almost nobody; and 928 rows for Puerto Rico, Guam and the Virgin Islands, which none of these laws reach. That leaves 23,515 cells in the analysis.

Chance. The permutation test reshuffles which jurisdictions carry the mandate label across the 42 states in the comparison and recomputes the whole within-occupation statistic 4,000 times. The reported p-value is the share of reshuffles producing a gap at least as large in absolute value. The shuffle is seeded, so the number does not drift between builds.

Reproducibility. Two scripts, no manual steps beyond the annual download BLS requires a human to perform: scripts/build-oews-state-slice.mjs extracts the state slice and scripts/build-disclosure-study.mjs emits every figure on this page and the CSV. Nothing here is typed by hand.

If you reproduce this from the CSV you will land within about a hundredth of a point of the figures above rather than exactly on them: the file stores band ratios rounded to three decimals, while the analysis runs on the unrounded values. Recomputing the headline from the published file gives +0.63% against the +0.62% reported here. If you get something further away than that, we would like to know.

Limitations

A null result can be misread as confidently as a positive one. These are the reasons not to over-read it.

1. Absence of an association is not proof of no effect. This design would miss a real effect that is small, slow, or concentrated in the occupations and firms that actually hire through public postings. It rules out a large, fast, economy-wide compression of paid wages. It does not rule out a modest one.
2. Wages paid, not ranges posted. The survey never sees a job advert. If posting mandates compress the advertised range while leaving the realised distribution alone — which is exactly what the posting-data study reports — this page would look precisely as it does.
3. The treatment leaks into the control group. Multi-state employers standardise postings, and remote roles are typically covered by the strictest applicable state. There is no clean untreated group anywhere in this data.
4. Each estimate pools three years. Six semiannual panels collected over three years, aged forward with the Employment Cost Index. A law effective in January 2025 touches one panel in six of a May 2025 estimate, so recent mandates are heavily diluted by construction — the reason the effective-date column exists and the reason no diff-in-diff appears here.
5. Base pay only. OEWS measures straight-time gross pay. Equity, RSUs and nonproduction bonuses are excluded from the survey entirely — precisely the components that move most in the high-wage states that make up most of the mandate group.
6. No published reliability measure for percentiles. BLS publishes sampling-variability estimates for employment and mean wages, not for the percentiles this study is built on. The permutation test asks whether the labelling could have produced the gap; it cannot price the survey's own sampling error on top.
7. A wide band is not negotiating room. The p25–p75 gap contains differences between firms, industries and workers' seniority as well as whatever bargaining occurred, and published aggregates cannot separate them. The companion study of pay band width by metro area sets out that argument in full.