Why the H-1B Wage Gap Can Change When the Question Changes

The subject here is the capped, for-profit H-1B program—and, more specifically, the relationship it creates between a petitioning employer, a named foreign worker, a scarce lottery slot, and a wage offer. The focus is why claims about the H-1B wage gap change when the comparison changes. That distinction matters because the same number can be used to make claims about a worker’s pay, an employer’s incentive, compliance with wage rules, or the wisdom of a visa fee. Those are not the same claim.

A useful conclusion does not begin by asking whether H-1B workers are simply “underpaid.” It begins by asking: compared with whom, measured when, located where, and under what institutional constraint? This series follows those questions so that a wage estimate can be read as evidence with a defined scope, rather than as a career verdict about everyone who holds an H-1B visa.

The program is an employer–worker relationship

The H-1B program admits foreign workers to “specialty occupations,” generally for a period of up to six years. For-profit employers face an annual cap of 85,000 new visas. When demand exceeds that number, a lottery decides which employers can proceed with particular hires. The scale of the competition is substantial: the Borjas paper reports that about 450,000 distinct beneficiaries participated in the lotteries on average between 2021 and 2026.

The important institutional detail is that the visa is not an abstract permission attached to a worker alone. A particular firm petitions for the temporary employment of a particular person. The worker may move to another firm, but the new employer must petition again. In the paper’s account, 63,865 H-1B workers changed jobs in fiscal year 2024 out of an estimated population of about 680,000, an annual separation rate of 9.4 percent. The paper compares that with a cited 20–25 percent range for comparable highly educated workers. The difference does not prove that every H-1B employer suppresses pay, but it does explain why the employment relationship belongs in the wage question: access to a scarce visa and the difficulty of changing sponsors can affect bargaining power as well as hiring decisions.

This also explains why the employer is central to the subject. The firm is not merely choosing between two interchangeable résumés. It is deciding whether to use a limited visa slot for a specific worker, pay the offered wage, bear the costs of sponsorship, and keep that worker in a relationship that may last several years. A wage comparison that ignores this setting risks treating the program as if it were only a demographic category.

“More” and “less” depend on the comparison

At the broadest level, H-1B workers appear to earn more than native-born workers. The Economic Innovation Group reports median H-1B pay of $120,000 in 2024, compared with $67,000 for the average native-born worker. That is a meaningful fact about the kind of labor the program attracts, but it is not an apples-to-apples comparison. H-1B workers are concentrated in high-paying occupations, metropolitan areas, and industries, and they are more likely to be young, male, and highly educated.

The Borjas paper’s own sample makes the problem visible. Its H-1B workers average about $112,600 in annual salary, while the selected native comparison group averages about $127,000, all in 2025 dollars. Those figures do not contradict the EIG comparison. They describe different populations and different summaries: a median against an average in one case, and a restricted group of private-sector, full-time, college-educated workers in the other. The H-1B sample is also unusually concentrated. About 43.9 percent work in its five largest metropolitan areas, compared with 15.1 percent of the native sample; its five largest occupations account for 63.8 percent of H-1B employment, compared with 7.8 percent for natives.

Once those differences are acknowledged, the question changes. Instead of asking whether the typical H-1B worker earns more than the typical person born in the United States, an analyst can ask how the wage differs after comparing workers with the same observable education, age, gender, occupation, industry, and metropolitan area. That is the logic of a conditional wage estimate. It is not a direct observation of two people competing for the same job. It is a statistical construction intended to reduce the influence of measurable differences between the groups.

In the revised NBER analysis, the raw comparison gives H-1B workers a wage advantage of about 4.4 percent in the initial specification. Adding geographic controls removes that advantage. Adding observable skills and occupation controls produces a disadvantage of about 13.6 percent, and adding industry controls produces an adjusted gap of roughly 16.1 percent. The result is a striking reversal, but not an arithmetic contradiction. The raw comparison answers one question; the controlled regression answers another.

That distinction is easy to lose when a single percentage is detached from its design. “H-1B workers earn 16 percent less” sounds like a statement about every worker’s paycheck. More precisely, the NBER estimate says that, within its sample and its chosen controls, the logged wages reported for H-1B hires are lower than the wages predicted for a constructed U.S.-born baseline. The coefficient is a conditional association. It does not by itself reveal whether the difference reflects employer market power, unobserved productivity, job selection, discrimination, the timing of the wage records, or some combination of those mechanisms.

flowchart LR A["Employer petitions for a named H-1B worker"] --> B["Wage offer inside a scarce visa system"] B --> C["Compare with average U.S.-born pay"] B --> D["Compare with a constructed similar-worker baseline"] B --> E["Ask whether the legal required wage is met"] D --> F["Model a possible employer fee response"] C --> G["Different questions, different conclusions"] E --> G F --> G

The reversal also does not imply that there is one H-1B experience. The supplied analyses report positive adjusted gaps at some employers and large negative gaps at others, while EIG reports a more positive pattern for younger workers and a negative pattern for older workers. The average can therefore be useful as a description of a selected population while still being a poor summary of what any particular employer or worker will experience.

A wage statistic does not settle a legal or policy question

The phrase “wage gap” leaves several essential facts unstated. Is the H-1B number a starting offer or income actually received over a year? Does the comparison use the year of the petition, the year employment began, or a survey’s trailing twelve months? Does “location” mean where the H-1B worker works or where the native worker lives? Does the native baseline include government and nonprofit workers when the H-1B sample is limited to for-profit firms? Are new H-1B hires being compared with natives who have many years of job tenure? Each choice can change the meaning or size of the estimate. The next parts of this series examine these choices rather than treating them as technical decoration.

There is also a legal question that the statistical comparison cannot answer by itself. The Department of Labor fact sheet supplied with these materials describes the H-1B required wage as the higher of the prevailing wage and the employer’s actual wage for similarly employed workers. It describes the actual wage in relation to workers with similar experience and qualifications in the specific employment, and the prevailing wage in relation to similarly employed workers in the geographic area. That is a legal floor with employer- and place-specific components. It is not identical to the wage of a constructed ACS comparison group.

Consequently, a negative regression coefficient is not proof that an employer violated the H-1B wage rules. An employer could comply with the applicable required wage while paying less than the average earnings of a carefully selected U.S.-born group. Conversely, a statistical comparison could miss a violation because it is too broad or uses the wrong time and place. Compliance requires facts about the relevant employer, job, wage determination, and payment practice that the aggregate regression does not observe.

Nor does a wage gap settle the policy question. Borjas’s model uses the estimated difference to infer how much a firm might be willing to pay for a visa over a six-year relationship. That inference adds assumptions about productivity, sponsorship costs, wage growth, job turnover, lottery participation, and the effect of a fee on which firms register. EIG’s counteranalysis argues that different timing and geographic choices reduce the gap substantially and that the age profile of H-1B workers matters. The revised NBER paper disputes that interpretation. The disagreement is not a reason to ignore the evidence; it is a reason to keep the evidence attached to the question and assumptions that produced it.

The practical judgment supported by this series is therefore narrower and more useful than a universal verdict. H-1B workers may earn more than the average native-born worker while earning less than a constructed group of comparable workers. Some employers may pay premiums, while others may pay deficits. A measured disadvantage may be relevant to bargaining power or policy design without proving unlawful underpayment. And a fee that looks attractive in a revenue model may have different effects on young, highly paid workers than on older workers in lower-paid employment.

Before accepting any H-1B wage headline, ask four questions: compared with whom, measured when, located where, and under what institutional constraint? The answers will not eliminate disagreement. They will show what the disagreement is about. The next step is to inspect the records behind the estimates—the position-level LCA, the beneficiary-level I-129 petition, and the ACS comparison sample—because a wage gap is always an answer with its question attached.

References