Key Findings

• We estimate that 61 percent of H-1B registrations would achieve a higher wage level classification by reclassifying into one of the five most closely related occupations, with an average gain of 1.3 additional lottery tickets among those who reclassify to the most closely related occupation offering a higher wage level. When only the single most closely related occupation is considered as an alternative, 41 percent would reclassify to a higher wage level.
• Under a strategy in which the employer reclassifies into the most closely related occupation among the top five neighbors (our benchmark scenario), Architecture and Engineering occupations gain the most selection share (+1.78 pp), while Computer and Mathematical occupations lose an additional 1.40 pp beyond what the DHS rule alone removes.
• Reclassification would partially reverse the rule’s intended reallocation: the (pre-reclassification) Level I selection share would rise from 13 percent under the DHS rule back to 21 percent (compared to 28 percent under the random lottery), while the (pre-reclassification) Level IV share would fall from 31 percent to 23 percent (compared to 18 percent under the random lottery).
• We estimate that 42 percent of the expected compensation increase under the DHS rule is offset by reclassification, reducing the gain in average compensation among selected registrants from 10 percent to 5.8 percent relative to the random lottery.

---

Introduction

Our previous analysis projected that the new DHS wage-level-weighted H-1B lottery rule would shift visa allocation toward higher-paid workers, increasing the average annual compensation of selected applicants by about 10 percent. These estimates assume that employers will continue classifying workers in the same occupations under the new rule as they did under the prior random lottery.¹

This assumption may not hold. The weighted lottery assigns selection weight based on a registration’s Occupational Employment and Wage Statistics (OEWS) wage level — the highest level that the offered salary equals or exceeds for a given occupation and metropolitan statistical area (MSA). Because OEWS wage thresholds vary substantially across occupations, employers have an incentive to reclassify positions into related occupations with lower thresholds. Policy analysts have flagged this concern: closely related occupations often have materially different OEWS wage thresholds in the same labor market, so employers may be able to increase lottery weight by adjusting how they describe a position without changing the underlying job duties, worker qualifications, or (importantly for the goals of the policy) offered compensation.² In the final rule, DHS acknowledged the concern but stated that “there are sufficient provisions to detect and deter occupational misclassification during registration.”³

The scope for such strategic reclassification depends on two factors: the availability of closely related alternative occupations for any given position, and the extent to which prevailing wage structures differ across those alternatives in local labor markets. To systematically assess this possibility, we conduct a counterfactual analysis using occupational relatedness data from the O*NET program. For each H-1B registration in the FY 2024 lottery data, we identify the set of most strongly related occupations as determined by O*NET’s expert-reviewed similarity scores. These scores reflect overlap in tasks performed, knowledge required, and occupational nomenclature. We then determine whether reclassifying the position into any of these related occupations would yield a higher equivalent wage level at the same offered compensation.

This exercise provides an estimate of strategic reclassification potential: it captures what employers could do given occupational proximity and prevailing wage differentials, not what employers actually will do. For our main analysis, we consider only the most plausible reclassification candidates (the five nearest occupational neighbors) and hold offered compensation fixed. In practice, employers might consider a broader set of occupations or adjust compensation offers to jointly maximize lottery advantage and minimize compensation costs. Our analysis also does not account for adjudicator scrutiny or legal risk, which may deter some reclassification attempts. However, because the rule conditions selection weight on occupational classification, and because job boundaries across related occupations are fluid, even modest reclassification can alter lottery outcomes.

Reclassification Strategies

A registration reclassifies if at least one of the five most closely related occupations yields a higher wage level classification, holding the worksite MSA and offered compensation fixed. To illustrate, consider a Software Developer in the Washington-Arlington-Alexandria MSA with an offered salary of $134,581. Under the original classification, the salary falls in Level II ($126,090–$150,758), yielding 2 lottery entries. The second-most related O*NET occupation, Computer Systems Analyst, has a Level III threshold of $129,480 in the same MSA — below the offered salary. By reclassifying, the employer moves the registration to Level III and ends up with 3 lottery entries without changing the compensation.⁴

Job title data from H-1B approvals in FY2021–FY2024 support the plausibility of this switch. Here, “job title” refers to the employer-specific position name on the filing (e.g., “Application Developer”), as distinct from the standardized SOC occupational category. Although only around five percent of unique Software Developer job titles also appear under Computer Systems Analyst, these overlapping titles are disproportionately common: they account for nearly 70 percent of Software Developer approvals with observed job titles. In other words, most Software Developer H-1B positions already carry a job title that has been approved under both SOC codes, suggesting that many employers could reclassify without altering the position title itself.

To systematically assess reclassification potential, for each H-1B registration in FY 2024 we:

1. Identify the five most closely related occupations from O*NET.
2. Retrieve OEWS wage levels for each related occupation in the registration’s worksite MSA.
3. Compute the equivalent wage level that the offered salary would achieve under each related occupation in the same MSA.
4. Flag whether any related occupation yields a higher wage level than the original classification.

Our counterfactual exercise considers two distinct strategies employers might use when selecting among the five most closely related occupations:

• First-alternative strategy: The employer reclassifies into the most closely related occupation, among the five considered, that yields a higher wage level at the offered compensation, regardless of whether a more distantly related occupation would yield a larger wage level gain. This strategy stays as close as possible to the original occupation, minimizing the occupational distance of the switch while still capturing any available upward reclassification.

• Maximum-gain strategy: The employer selects whichever of the five most closely related occupations yields the largest increase in equivalent wage level at the offered compensation. If multiple alternatives produce the same maximum gain, we take the one ranked most closely related. This strategy reflects a more aggressive approach in which the employer is willing to reclassify further afield within the set of plausible substitutes to maximize lottery odds.

In the results below, we report outcomes under both strategies. The first-alternative strategy provides a conservative lower bound on the wage level gains achievable through reclassification, while the maximum-gain strategy provides an upper bound given the five-occupation constraint. The gap between the two strategies reflects the extent to which the closest occupational neighbor is not always the one with the most favorable prevailing wage structure (in terms of lottery odds).

Results

Strategic Reclassification Potential

Using the FY 2024 registration data, we find that 60.7 percent of H-1B registrations achieve a higher wage level classification by reclassifying into one of the five most closely related O*NET occupations, holding the offered compensation fixed. Among those who reclassify, the average gain is 1.3 wage levels under the first-alternative strategy.

Figure 1 below traces how reclassification potential grows as we expand the set of alternative occupations considered. The number of occupations considered can be interpreted as how aggressively employers search for reclassification opportunities, from the single most obviously similar occupation on the left to increasingly distant occupational neighbors on the right. We use five, corresponding to O*NET’s “Primary-Short” tier of closest neighbors, as a relatively conservative assumption. With just the single most closely related occupation, 41.1 percent of registrations reclassify to a higher wage level; with five, 60.7 percent do; and the curve flattens around 13 occupations, plateauing at approximately 81 percent when all 20 related O*NET occupations are considered—close to the theoretical ceiling of approximately 83 percent (the total share of registrations below Level IV that could potentially reclassify upward).

Heterogeneity by Baseline Wage Level

The table below decomposes the 60.7 percent that reclassify by source and destination wage level, expressed as a share of all registrations. Level I registrations account for 28.6 percent of the pool; of these, 80 percent reclassify to a higher wage level through at least one related occupation, with an average gain of 1.5 levels under the first-alternative strategy. Most move to Level II (15.1 percent of all registrations), with smaller shares reaching Level III (4.2 percent) and Level IV (3.6 percent). Level II, the single largest wage level group at 39.2 percent of the pool, reclassifies at a 69.8 percent rate with an average gain of 1.3 levels; the dominant flow is to Level III (19 percent of all registrations), while 8.4 percent jump to Level IV. Level III (14.8 percent of the pool) reclassifies at a 70.3 percent rate but can gain only one level by definition. Level IV registrations, the remaining 17.4 percent, already carry the maximum weight and cannot benefit from reclassification.

The largest single flow in absolute terms is the Level II to Level III transition, which alone accounts for 19.0 percent of all registrations.

Compensation Effects

We now assess how strategic reclassification affects the composition of lottery-selected registrations. Under the DHS rule without switching, the wage-level weighting raises average compensation of selected registrants by $11,172 (10 percent) relative to the random lottery baseline. Under the first-alternative strategy with five related occupations, this gain shrinks to $6,461 (+5.8 percent relative to the random lottery baseline), a reduction of 42.2 percent. Reclassification alone offsets $4,711 of the average compensation increase the rule was expected to produce. We summarize these results in Table 2.

Figure 2 below traces the compensation effects as a function of the number of related occupations considered. The DHS rule line is flat by construction, while the reclassification scenarios decline as more alternative occupations become available for switching. The gap widens monotonically but begins to plateau around 13 occupations, consistent with the flattening of reclassification potential documented above. Even when employers can reclassify into only the single most closely related occupation, average compensation among selected registrants already falls below the DHS rule level: the gain relative to the random lottery is 2.5 percentage points smaller than under the rule without reclassification (Figure 2, leftmost point).

The baseline (pre-reclassification) wage level composition of selected registrations tells the same story. Under the random lottery, Level I registrations account for 28.1 percent of selected slots and Level IV for 18 percent. The DHS rule shifts this distribution sharply: Level I falls to 13.3 percent while Level IV rises to 30.9 percent. Under the first-alternative strategy, reclassification partially reverses that reallocation, pushing the baseline Level I share back to 20.9 percent and the baseline Level IV share down to 23.1 percent. Figure A2 in the Appendix shows the corresponding distribution by submitted (post-reclassification) wage level.

Who Gains Most from Reclassification?

Not all occupation groups gain equally from strategic reclassification. Under the first-alternative strategy, Architecture and Engineering occupations gain 1.78 pp of selection share relative to the DHS rule without reclassification, most of it redirected from Computer and Mathematical occupations, which lose an additional 1.40 pp beyond what the DHS rule alone already removes (Table A3). These shifts reflect three margins: what share of a group’s registrations sit at wage levels below the maximum (exposure), what fraction of those workers can actually reclassify upward given the prevailing wages of related occupations (reclassification rate), and how many wage levels they jump when they do (average jump). We formalize this decomposition in the Appendix and plot the result for each major occupation group below.

Groups to the right of the dashed line have above-average expected extra entries from reclassification and generally gain selection share. Architecture and Engineering scores high on all three margins: its registrations are concentrated at Levels I and II, related occupations frequently offer a higher classification at the same salary, and the resulting jumps are large. Computer and Mathematical occupations have a reclassification rate near the pool average (61 percent) but switchers tend to land only one level higher, because prevailing wages do not vary as much across the software- and data-centric occupations that populate its O*NET neighborhood. Business and Financial Operations, with the lowest reclassification rate among the major groups (42 percent), sees much of the share it gained under the DHS rule eroded as groups with stronger reclassification potential crowd it out.

Discussion

The new DHS rule aims to preferentially allocate H-1B visas to workers whose offered wages are high relative to prevailing wages in their occupation and location. Our analysis suggests that this mechanism is sensitive to strategic occupational reclassification. Because the boundaries between related occupations are often ambiguous and prevailing wages differ substantially across similar occupations, employers can increase their lottery odds without actually offering higher compensation.

Along the extensive margin (the set of occupations employers might consider), our estimates represent a lower bound on the scope for strategic reclassification. We consider only the five nearest occupational neighbors and hold compensation fixed. Employers with greater sophistication or tolerance for audit risk may identify additional reclassification opportunities, and, as we show, expanding the set of related occupations from five to twenty raises the share with reclassification potential from 61 to 81 percent. However, if many employers reclassify into the same destination occupations, crowding at higher wage levels would partially offset the individual advantage, and the realized impact could be smaller than our partial-equilibrium exercise implies. Actual reclassification rates will depend on enforcement intensity, adjudicator expertise, and employers’ perceptions of the legal risks. Classification patterns in upcoming H-1B filing cycles will provide the first direct test of whether and how employers respond.

---

Appendix

Data and Simulation

H-1B Registration Data

We use H-1B lottery registration data for fiscal year 2024, obtained by Bloomberg through Freedom of Information Act requests. The FOIA extract links registration records to the corresponding I-129 petitions and Labor Condition Applications (LCAs) filed by selected registrants. We draw offered compensation and worksite location from the petition records and the SOC occupation code from the associated LCA. Because these fields are available only for selected registrations that proceeded to the petition stage, our analysis sample consists of this subset of registrations.

O*NET Related Occupations

The O*NET program provides a crosswalk of related occupations based on an approach that incorporates three dimensions of occupational similarity: what workers do (tasks), what they know (knowledge domains), and what the occupations are called (title similarity).⁵ For each O*NET-SOC occupation, the database identifies the 10 primary and 10 supplemental most similar occupations, ranked by relatedness. In our main analysis, we focus on the “Primary-Short” tier, which contains the five most strongly related occupations after expert review. We take these as the most plausible reclassification candidates. O*NET similarity reflects task and knowledge overlap, which may not map directly to the criteria immigration adjudicators apply when evaluating whether a position fits a given SOC code under DOL guidance. Our reclassification estimates rest on the assumption that these O*NET neighbors represent legally viable H-1B alternatives; in practice, some neighbors may differ in specialty-occupation requirements, narrowing the set of feasible switches.

Prevailing Wage Data

OEWS wage levels are set according to prevailing wages at different points of the wage distribution within each occupation and geographic area. We obtain these data from the Office of Foreign Labor Certification (OFLC), which publishes wage percentiles by SOC code and MSA. Wage Level I corresponds approximately to the 17th percentile, Level II to the 34th percentile, Level III to the 50th percentile (median), and Level IV to the 67th percentile.

Simulation Procedure

Because the wage, occupation, and worksite fields we use are available only for selected registrations that proceeded to the petition stage, we generate 1,000 synthetic registration pools based on these observed records. In each iteration, selected registrations are treated as representative of the broader set of registrations with similar characteristics, and the synthetic pool is scaled to match the total number of FY 2024 registrations. We then draw a lottery sample from each synthetic pool and simulate the random lottery, the DHS rule, and the DHS rule under each reclassification strategy (first-alternative and maximum-gain). Final estimates are computed by averaging outcomes across the 1,000 simulated lotteries.

More on the O*NET Related Occupations Methodology

Our occupational relatedness data come from the O*NET program’s related occupations framework, most recently updated using the O*NET 30.0 Database. The method scores every possible pairing of the 923 data-level O*NET-SOC occupations on a composite similarity index, then subject-matter experts at the National Center for O*NET Development review and refine the rankings.

The composite equally weights three components: (1) work-based similarity, derived from sentence embeddings of each occupation’s task statements and detailed work activities; (2) knowledge-profile similarity, the cosine between occupations’ importance ratings across O*NET’s 33 knowledge domains; and (3) title similarity, based on word embeddings of each occupation’s job titles, weighted to downplay generic terms common within a job family.

Each occupation receives a ranked list of related occupations. Expert reviewers compare these against structural relationships in the O*NET-SOC taxonomy and may override where the empirical ordering seems off. The final related occupations matrix assigns each occupation five “Primary-Short” related occupations (its closest neighbors) along with five more “Primary-Long” and ten “Supplemental” tiers for broader occupational similarity searches.

Decomposition of Reclassification Gains by Occupation Group

Under the DHS rule, a registration at wage level $w$ receives $w$ lottery entries (1 for Level I, up to 4 for Level IV). Reclassifying from level $w$ to $w' > w$ adds $w' - w$ entries. For occupation group $g$, the expected extra entries per worker are

$$E[\Delta\omega]_g = \sum_{w \in \{I,\, II,\, III\}} \pi_g(w) \times r_g(w) \times \ell_g(w)$$

where $\pi_g(w)$ is the share of group $g$‘s registrations at source level $w$ (exposure), $r_g(w)$ is the fraction that can reclassify upward through one of the five most related O*NET occupations (reclassification rate), and $\ell_g(w)$ is the mean number of levels gained by reclassifiers starting at $w$ (average jump). Level IV is excluded from the sum because those registrations already carry the maximum weight.

The exact condition for a group to gain selection share is that its proportional increase in average lottery weight — $E[\Delta\omega]_g$ divided by the group’s baseline average weight — exceeds the pool-wide average proportional increase.⁶ Because reclassification-driven gains are small relative to baseline weights of 1–4, $E[\Delta\omega]_g$ alone closely tracks the proportional ranking, and groups with above-average $E[\Delta\omega]_g$ generally gain selection share. The stacked bars in the main-text figure show each group’s $E[\Delta\omega]_g$ decomposed by source wage level; hovering over a segment reveals the underlying $\pi$, $r$, and $\ell$ values for that cell.

The figure below plots each occupation group’s proportional gain in average lottery weight (x-axis) against its simulated share change from reclassification (y-axis), with circle area proportional to the group’s registration share. The strong positive relationship confirms that the decomposition captures the dominant source of variation in simulated outcomes: groups whose workers collect disproportionately more extra lottery tickets systematically gain selection share, while those below the pool-wide average (dashed line) lose share.

Submitted Wage Level Distribution

Figure 3 in the main text reports the baseline (pre-reclassification) wage level composition of selected registrations. The figure below shows the corresponding distribution by submitted (post-reclassification) wage level — i.e., the wage level that determines each registrant’s lottery weight after any strategic reclassification.

Additional results (considering five related occupations)

---

This analysis was produced by PWBM staff under the direction of Alex Arnon. This brief was written with the encouragement of Heidi Williams who also provided valuable feedback.

Footnotes

1. These figures correspond to FY 2024 lottery data. See Table 1 in our previous brief. The dollar figures in this brief differ slightly from the previous analysis due to simulation variation across runs. ↩

2. For illustrative examples and discussion of this concern, see Economic Innovation Group, “EIG Letter: DHS Should Revise Proposed H-1B Weighted Lottery to Prioritize Top Talent,” October 2025, https://eig.org/eig-letter-dhs-should-revise-proposed-h-1b-weighted-lottery-to-prioritize-top-talent/; Jeremy Neufeld, “The ‘Wage Level’ Mirage: How DHS’s H-1B Proposal Could Help Outsourcers and Hurt U.S.-Trained Talent,” Institute for Progress, September 2025, https://ifp.org/the-wage-level-mirage/. ↩

3. “Weighted Selection Process for Registrants and Petitioners Seeking To File Cap-Subject H-1B Petitions,” 90 Fed. Reg. 60864, 60916 (Dec. 29, 2025), https://www.federalregister.gov/d/2025-23853. ↩

4. This example is based on an actual FY 2024 registration for a Software Developer in this MSA, updated with a plausible offered salary (assuming growth equal to that of the median wage for this occupation) and with OEWS wage-level thresholds for the FY 2027 H-1B lottery. ↩

5. See O*NET Resource Center, “Developing Related Occupations for the O*NET Program,” https://www.onetcenter.org/reports/Related_2022.html and “Updates to Related Occupations for the O*NET Program Using the O*NET 30.0 Database,” https://www.onetcenter.org/reports/Related_2025.html. ↩

6. Figure A1 in the Appendix plots the proportional measure directly against simulated share changes. ↩