Strategy and timing
The missing denominator: why published NIW outcomes mislead self-petitioners
Published NIW case summaries imply a 9% denial rate. USCIS data says 44.8% in FY2025 and over 50% in FY2026. Where the gap comes from, and what the negative cases actually say.
Updated September 10, 2026 · 9 min read
If you are considering an EB-2 National Interest Waiver petition without a lawyer, you will do what any sensible person does: look for evidence about what works. You will find a great deal of it. Law firms publish approval notices. Forums fill with success stories. Consultants post case studies with the names redacted and the outcomes intact.
What you will not find, in any quantity, is the other half of the record.
We maintain a structured database of EB-2 NIW outcomes drawn from every public source we could identify, published law firm case summaries, a compilation of anonymised approved profiles, Administrative Appeals Office decisions, and first-hand accounts on public forums. Each is coded for field, degree basis, evidence counts, whether a Request for Evidence was issued, processing time, and, where the outcome was negative, the specific reason the officer gave.
At the time of writing it holds 55 outcomes. Here is the aggregate:
| Outcome | Count |
|---|---|
| Approved | 43 |
| RFE, then approved | 7 |
| Denied | 4 |
| RFE, then denied | 1 |
Fifty of fifty-five approved. A 91% approval rate.
That number is worthless, and understanding exactly why is the most useful thing on this page.
Where the cases came from
| Source | Count |
|---|---|
| Law firm published case summaries | 35 |
| Compiled approved-profile document | 11 |
| AAO appeal decisions | 5 |
| Public forum accounts | 3 |
| Blog account | 1 |
Thirty-five of fifty-five outcomes, 64%, come from law firms publishing their own results. A law firm does not publish the petition it lost. It cannot: the client would not consent, and the firm has no reason to ask. Every law-firm-sourced record in this dataset is, by construction, an approval.
A further eleven come from a compiled document of approved profiles. The clue is in the name.
The AAO cases are the only systematically negative source, because the AAO publishes appeals, and you only appeal a denial. All five are failures. They are also, tellingly, the only records in the dataset with a coded reason attached to the outcome.
So the dataset contains two populations selected by opposite mechanisms, and neither was selected at random. The approval rate it reports is not an estimate of anything. It is an artifact of who publishes what.
The actual denominator
USCIS publishes it. The quarterly Form I-140 by Fiscal Year, Quarter and Case Status report carries a dedicated National Interest Waiver line, separate from the general advanced-degree category. Computing the rate on decided petitions, approvals divided by approvals plus denials, gives this:
| Fiscal year | Approved | Denied | Approval rate | Denial rate |
|---|---|---|---|---|
| FY2023 | 31,889 | 8,172 | 79.6% | 20.4% |
| FY2024 | 27,526 | 11,256 | 71.0% | 29.0% |
| FY2025 | 19,532 | 15,863 | 55.2% | 44.8% |
| FY2026 (Q1–Q3) | 9,238 | 9,585 | 49.1% | 50.9% |
The denial rate has more than doubled in three years, and in the first three quarters of FY2026 more NIW petitions were denied than approved.
Set that against the 9% denial rate implied by the published case summaries. The gap is not a rounding difference. Someone calibrating their expectations on law firm case studies is working from a picture roughly five times more optimistic than the agency's own numbers, and the discrepancy is widening every year, because the published sample cannot get worse while the real rate can.
Two caveats, both of which matter:
These are not cohort rates. USCIS counts decisions made in a period, not outcomes for petitions filed in it. Pending NIW petitions grew from 19,117 at the end of FY2023 to 96,297 by FY2026 Q3, so petitions decided in FY2025 were largely filed a year or more earlier. The trend direction is unambiguous. A single year's rate is not "the odds for someone filing that year."
FY2026 is incomplete. Three quarters, not four.
One more figure deserves its own line, because it is the clearest measure of what is actually happening: the pending backlog grew five-fold in three years, from 19,117 to 96,297. Receipts rose while decisions fell. That is what an adjudication system under load looks like, and it is the context every number above sits in.
Why this specifically harms people filing alone
Someone with counsel is insulated from this. Their attorney has seen their own denials. The base rate lives in that lawyer's memory whether or not it appears on the firm's website.
The self-petitioner has no such correction available. They are building an internal model of what an approvable petition looks like out of a sample that contains almost no unapprovable ones. Every case they read succeeded. The natural inference, my record resembles these, so my record is sufficient, is unsound, and there is nothing in the public evidence to check it against.
The cost of that error is not a wasted afternoon. It is the filing fee, the months, and the priority date, which is the one asset in this process that compounds.
What the negative cases actually say
Because the appeal decisions are the only records with reasons attached, they carry disproportionate information. Coding the failures produces a small taxonomy:
| Failure mode | Count |
|---|---|
| No execution plan for the proposed endeavor | 2 |
| Benefit confined to one locality or institution | 2 |
| Field is important, but the petitioner's contribution is not distinguished | 2 |
| Benefit runs to private parties, employer, clients, not the public | 2 |
| Claims not credible on the evidence supplied | 1 |
| EB-2 base eligibility never established | 1 |
| Endeavor described too broadly to evaluate | 1 |
Note what is absent. Not one is "insufficiently accomplished". None turns on citation count, publication count, or years of experience, the metrics self-petitioners obsess over, and the metrics that dominate the approval summaries.
Every one is a framing or evidentiary-structure failure. The endeavor was described at the level of a field rather than a specific undertaking. The benefit was real but ran to an employer. The importance of the domain was established and the petitioner's particular role within it was not. There was no account of how the endeavor would actually proceed.
These are fixable before filing, at no cost, by someone who knows to look for them. They are not fixable after a denial.
One failure deserves separate mention: a petition denied because EB-2 base eligibility, the advanced degree or its equivalent, was never established at all. The Dhanasar prongs were never reached. A petitioner who spends three months perfecting a national-importance argument while leaving a foreign degree unevaluated has optimised the wrong thing entirely.
Two distributions worth reading carefully
Processing time. Nineteen records report it. The median is 180 days; the range runs from 14 to 365. A median from nineteen self-selected observations should not be used to plan anything. But the spread is real information: anyone budgeting "about six months" is planning against a distribution whose tail is a full year.
Fields. No field appears more than twice. The dataset covers radiation oncology, chemical engineering, martial arts, online media, animal breeding, environmental fluid mechanics, music, and clinical toxicology. This is genuinely encouraging, and it is the one aggregate conclusion the data probably does support: the NIW is not a research-scientist category. Petitioners in applied, commercial, and artistic fields appear throughout.
Note that this, too, is shaped by who publishes. An unusual field makes a better case study than a conventional one.
Why there is no approval-likelihood score on this site
The obvious product is a percentage. It is what people ask for, it is what they would pay for, and it is unbuildable in any honest form.
Training a score on these outcomes would encode the selection bias directly into the output and hand it back wearing a number, which is worse than no score at all, because a number carries an authority that prose does not.
What the data does support is different and less flattering: a checklist of documented failure modes, each traceable to a real decision, applied to a record before it is filed. Not "you have a 78% chance" but "your endeavor is described at field level, which is the defect in two of the five appeal decisions we can point you at, and here is what those petitioners were missing."
That framing has a property a score does not. It can return the answer do not file yet. A score cannot meaningfully do that, any number above zero reads as encouragement. And the ability to say no is the only thing that makes this worth doing, because the entire cost of the problem is borne by people who filed when they should have waited.
What we are not claiming
This dataset is small, non-random, and assembled by hand. It cannot estimate an approval rate, that is its central finding, not a caveat to it, and the USCIS figures above are what an actual rate looks like. It cannot tell you whether your petition will be approved. It cannot substitute for a licensed immigration attorney, and for a complicated history, a prior denial, or a live Request for Evidence, it should not be asked to.
What it can do is the narrow thing: name the specific, documented ways petitions have failed, so you can check your own record against them before the filing fee is spent rather than after.
The public evidence base for this category is missing its denominator. Until that changes, the most useful contribution anyone can make is to stop reporting the numerator as though it were a rate.
Every record, with its source citation, is published here. Corrections and additional cited outcomes, particularly denials, are welcome.