Who gets paid to be here
A doctorate is not something you buy. In science and engineering it is a job: an assistantship, tuition waived, a stipend, and the work funded largely out of federal research grants. The question is who gets that deal.
| Primary source of support | US citizen or permanent resident | Temporary visa |
|---|---|---|
| Research assistantship | 35.4% | 53.6% |
| Teaching assistantship | 17.7% | 22.4% |
| Funded position, either kind | 53.1% | 76.0% |
| Fellowship or scholarship | 27.6% | 17.4% |
| Paying their own way | 16.0% | 3.5% |
An American in a STEM doctoral programme is more than four times as likely to be funding it out of their own pocket. And this is not one odd discipline: the gap runs through every one of the nineteen fields the survey covers, from social sciences to engineering. Not one runs the other way.
NSF Survey of Earned Doctorates, Table 4-1 (2024) — a census of research doctorate recipients, administered by institutions and completed before graduation, not a sample. See the distributions · Method and limits
What happens when citizenship is required
Computer science doctorates go to non-residents about 63% of the time. Information-security doctorates — the field closest to classified work, where a clearance is normally a condition of the job — go to non-residents 2.5% of the time.
Same discipline. Largely the same universities. The difference is that one of them requires citizenship and the other does not — and where it is required, Americans fill the programmes. That is worth sitting with: the composition of these fields is not a fact of nature or a shortage of American talent. It responds to what is asked for.
The security programmes are small — 511 doctorates against 8,624. What IPEDS records is which degrees were awarded, not why anyone was admitted. How far this can be pushed.
The same university, two programmes
The clearance contrast holds across two fields. This one holds inside a single institution — same campus, same degree level, same admissions office — which removes the easiest objection to all of it.
Lamar confers three times more counseling psychology degrees than computer science degrees, and not one of them went to a student on a temporary visa. Nobody can answer that this is a university that happens to recruit internationally. It recruits internationally into one programme.
It is not one university either. 165 institutions run one master’s programme at least 50 points above another of their own, and 99 do it at doctoral level. Illinois Tech awards electrical engineering doctorates at 93.2% non-resident and psychology doctorates at 1.8%.
The pattern is consistent: the high side is computing, analytics, statistics and engineering; the low side is psychology, counselling and education. See all of them, at both levels · What this does and does not show
The departments
One field, one institution, 2015–2022. Conservative by design — 80+ doctorates awarded, and 70%+ non-resident in at least six separate years. One unusual cohort cannot qualify.
These are federal records, not estimates.
Every figure in this table comes from IPEDS Completions — the Integrated Postsecondary Education Data System, collected by the National Center for Education Statistics at the US Department of Education. Reporting is mandatory for every institution that participates in federal student aid programmes, and each institution reports its own degrees conferred. It is a census, not a sample or a survey of opinion.
“Non-resident alien” is a federal reporting category for students on temporary visas. It is recorded by the institution at the time the degree is awarded. Nothing on this table is inferred from names, ethnicity, or country of origin — those cannot be derived from this data and are not attempted here.
The files are public. Every row below shows the exact steps to rebuild it: which file, which columns, which filters. Download IPEDS data yourself · Method and limits
| Institution | Field | Doctorates | Non-resident | Years ≥70% |
|---|---|---|---|---|
| Loading the 49 departments… | ||||
Who the expansion was built for
Between 2012 and 2025, American universities added about 159,000 core-STEM master’s places. Nearly 96,000 of them — three in every five — went to students on temporary visas. The system did not stand still and quietly change hands. It doubled, and the majority of what it built was not for Americans.
159,213 core-STEM master’s places added, 2012–2025 · IPEDS Completions
This does not depend on where you start counting. 2012 is one baseline; the result survives every other one. Measured to 2025 from any starting year between 2008 and 2018 — eleven different windows — non-residents took the majority of the new core-STEM master’s places every single time, between 50.2% and 60.8% of them. There is no baseline in that range that turns the majority around.
A funded place is a real thing with a real limit. Teaching assistantships are set by how many sections a department runs; research assistantships by what its grants will carry. When a department grows its intake and fills most of the new places with international students, that is a choice about who gets trained on the public’s money.
The scope of this finding. Domestic core-STEM master’s places rose 68% across the same window. What is measured here is who the growth was built for, and who pays for it. Method and limits
What the public puts in
$874,630. That is the median federal research obligation behind a single STEM research doctorate, across 99 institutions. Not tuition, not a grant a student applied for — public money flowing into the departments that train them, at close to nine hundred thousand dollars for every doctorate they produce.
STEM doctoral students are paid for: federal research assistantships, teaching assistantships, fellowships, tuition waived, stipend paid. The table below shows what that intensity looks like institution by institution, and what share of those doctorates went to students on temporary visas.
| Institution | Federal obligations | STEM doctorates | Per doctorate | Non-resident |
|---|---|---|---|---|
| Loading… | ||||
Both columns are federal records: obligations from the NSF National Center for Science and Engineering Statistics (FY2023), doctorates from IPEDS Completions (2015–2022). Method and limits.
The dissertation corpus
National trends
Data, methodology and machine access
Get in touch
Corrections, questions, and data requests are all welcome. If you think something here is wrong, say so — the sources are public and the working is published.
- Email andybarr@protonmail.com
- Signal andybarr.25
- Tip line Send something anonymously
- Support Buy me a coffee
Method and limits
Everything here comes from federal files anyone can download — IPEDS Completions for degrees, the NSF NCSES Federal Support Survey for funding. The methodology page carries the full chain, and the bulk exports let you recompute any of it yourself.
- Doctorates, not master’s. Cost figures cover research doctorates, where federal support concentrates. Master’s students mostly pay tuition; mixing the two would overstate the federal subsidy.
- An intensity ratio, not a price tag. Federal obligations buy research — papers, instruments, facilities, staff — not degrees. A campus with a national laboratory ranks high for reasons unrelated to graduate training.
- Why programmes differ is inferred, not recorded. The high side of every within-institution contrast is STEM-OPT eligible and feeds the H-1B route; the low side leads to state-licensed practice in the United States. That is the obvious reading and it is only a reading — IPEDS records degrees awarded, not who applied, who was admitted, or why anyone chose a programme. The gaps are measured. The explanation is not.
- The clearance contrast is a reading. Information security runs at 2.5% non-resident against computer science at 63.1%, and the obvious explanation is that a clearance is normally a condition of the job. IPEDS cannot confirm that: it records degrees awarded, not why anyone was admitted, and 511 doctorates is a small base. The contrast is measured; the explanation for it is inferred.
- Awards and funding, not admissions. IPEDS records which degrees were awarded and to whom; it does not record who applied or who was admitted. Domestic STEM doctorates rose about a third between 2012 and 2025. What this measures is who the growth was built for — and who pays for it.
- Non-resident is a visa status at graduation, not ethnicity or country of origin. These files cannot be split by nationality, and they do not record who applied, was admitted, or was funded.
- Reported, not inferred. Every figure on this page comes from a federal collection an institution was required to file: IPEDS Completions for degrees, the NSF federal support survey for funding. Elsewhere on this site, department-level and name-origin analysis is estimated from dissertation records, and that work is labelled as such. The two are never mixed: a name cannot establish a visa status, and this site does not claim otherwise.
- Self-reported. Institutions file their own completions, and NCES publishes what they file. That makes the data authoritative and auditable rather than independently audited — which is why every case here also shows whether that institution has reported consistently over time.