The H-1B “gender pay gap” is a sorting story, not an unequal-pay story

A widely-quoted statistic says H-1B women earn less than men — a raw gap of roughly 4%, and much larger (16–25%) among Chinese- and other-born workers. Taken at face value it reads as unequal pay for equal work. In the FOIA'd petition data it is almost entirely occupational sorting, not unequal pay: compare men and women in the same job title and the gap all but disappears — an adjusted +0.3% at the mean, −0.9% at the median across 388 title×year strata (107,000 petitions). The story is which jobs men and women hold, not different pay for the same job.

The best-known study of this dataset, Borjas's H-1B wage-gap paper (NBER w34793), asks a different question: how H-1B pay compares to U.S.-born workers (it finds a large gap, and the finding is contested). This page instead holds nationality and visa status fixed and looks within the H-1B workforce, at the gap between men and women — a cut that paper doesn't report.

What the raw numbers look like (and why they mislead)

Trimmed to $20k–$1M (raw values run from $0.01 to $169M, so trimming is mandatory):

• Men: mean $102,766 / median $94,000 (n = 244,503)
• Women: mean $98,014 / median $90,057 (n = 119,682)
• Raw gap: +4.8% mean, +4.4% median — and narrowing over time (+6.9% FY21 → +3.9% FY24).

Taken alone, that looks like a pay-equity problem. But H-1B base pay is anchored to the Labor Condition Application position wage, so within a given role there is little room for two workers to be paid differently — the aggregate gap has to come from composition (who holds which job), not from unequal pay inside a job.

The decomposition

Comparing men and women within the same job title (title×fiscal year strata, at least 20 of each gender, weighted by stratum size):

The 4% raw gap essentially disappears within identical titles. Within an employer it widens — because at a given company men and women hold different, differently-leveled titles; hold the title fixed too and it collapses back to ~1%.

The most striking result: the 25% "Chinese-born gap" vanishes

India-born workers — two-thirds of the sample — show ~0 gap even before controls, which is what drags the overall figure down to 4%. The eye-catching 16–25% raw gaps among Chinese- and other-born beneficiaries collapse to 3% or less once you compare the same job. Occupational sorting, not unequal pay.

Education explains nothing (the raw gap persists within Bachelor's / Master's / Doctorate), and age composition works against the raw gap (men are older, and older bands show the smallest gaps).

What this comparison does and doesn't show

• Titles encode seniority. "Senior Software Engineer" is a title, so equal pay within a title does not rule out gendered differences in reaching the senior title in the first place. This method controls for the job someone holds, not the path to it — the same limitation that applies to any controlled pay-gap number.
• Base pay only. The I-129 reports prospective base compensation — no equity or bonus, which is exactly where tech-sector gender gaps tend to concentrate. This is a floor-wage comparison, not total comp.
• Employer-reported, prospective. BEN_COMP_PAID is the rate of pay stated on the petition, not verified payroll.
• Sample. New-hire cap-lottery petitions, FY21–24. 33.5% of rows have a blank job title (a redaction-era source artifact); the decomposition uses only the rows with a title, and the raw gap is nearly identical in the blank (+5.5%) and non-blank (+4.5%) halves, so dropping them doesn't skew the result. (Treating the blank-title rows as their own occupational group would instead manufacture a spurious +3.1% "within-occupation" gap, which is why they're excluded rather than pooled.)

A related figure: reported pay vs the posted wage

The same data lets you compare each worker's reported pay to the wage posted on the Labor Condition Application. The median tells the honest story: the ratio is exactly 1.000 — the typical H-1B worker's reported pay equals the posted wage, and 71.8% are within ±1% of it. The often-cited "+19–24%" is a mean effect driven entirely by the upper quartile. So "H-1B workers are paid ~20% above the posted wage" is not a defensible summary; "the typical worker is paid the posted wage, and about one in four is paid above it, sometimes well above" is.

Explore the pay data yourself

The H-1B salary explorer shows reported-vs-posted pay for each occupation — for example software engineers — and the wage search lets you filter certified H-1B and PERM wages by employer, role and state. Individual employer profiles break the same comparison down company by company.

---

Source & method

The petition-level figures come from the I-129 microdata released under the Freedom of Information Act — "sourced from USCIS, obtained by Bloomberg" (Bloomberg v. DHS), the same dataset behind Borjas's H-1B wage-gap paper (NBER w34793). Registration-pool figures (eligible registrations, selections, the FY2025 rule effects) are from USCIS's own published H-1B Electronic Registration statistics and its FY2025 selection-rule rulemaking.

Universe: fiscal years 2021–2024, cap-subject H-1B lottery petitions that were selected and filed (372,841 rows) — not the full registration pool. A frozen FOIA snapshot; per-beneficiary identifiers are redacted, so figures are aggregates only. Every number on this page was checked against the live database on 2026-07-13.

Custom data slices for research or reporting. If you're a researcher, academic or journalist and want a specific cut of this data — a breakdown by employer, occupation, country of birth, fiscal year, or any combination not shown here — get in touch and I'll put it together. Attribution to visa-bulletin.us is appreciated.

More data analysis.