A flat average, a moving distribution
The national non-resident share of US STEM research doctorates has barely moved in fifteen years. That single number is why the change is easy to miss: it describes almost no actual department. These three views take the average apart.
IPEDS Completions, research doctorates (award level 17), core-STEM CIP families. A department is one 4-digit programme at one institution — Electrical Engineering at Purdue, not "Engineering" at Purdue. Method and limits
0 · The raw numbers
Proportions hide scale, so here are the counts first: how many core-STEM research doctorates the United States awarded each year, and to whom. Both lines rise. One rises faster.
Loading…
IPEDS Completions, award level 17, core-STEM CIP families, 1995–2025. Domestic is total minus non-resident. Across 2008 and 2009 institutions were migrating between award-level codings, so both are summed for those years — using one alone understates them by about a fifth and puts a false trough in the middle of the series. IPEDS published no 1999 completions file, so 1999 is absent upstream; the line breaks there and a dashed stub bridges the gap rather than drawing through a year we do not have.
0c · The longest view
IPEDS carries no citizenship breakdown before 1995, so the series above cannot start earlier. The National Science Foundation’s Survey of Earned Doctorates can: it is a census of every research doctorate awarded in the United States, and it reaches back to 1979.
Loading…
NSF NCSES Survey of Earned Doctorates, all fields, by citizenship status at the time of the award. Not the same measure as the chart above — that one counts core-STEM doctorates at named institutions, this one counts all fields nationally, so the two are shown separately rather than joined into one line. Published five-yearly to 2004 and annually from 2009; where the data is five-yearly the line is dashed, and where it is annual it is solid. Chained across seven NCSES releases, newest winning any year they disagree on.
0b · The same picture at master’s level
The doctorate is the smaller, slower stream. The master’s is where the expansion actually happened — and where the raw counts move hardest.
Loading…
IPEDS Completions, award level 7, same eight core-STEM CIP
families, 2008–2025. The series begins at 2008 because the
CTOTALT/CNRALT residency columns do not exist in
earlier files. Note the dip in 2022 and the jump after it: those graduates
enrolled in autumn 2020, under the pandemic travel restrictions. It is a
real cohort effect, not a reporting artefact — the number of reporting
institutions moves smoothly across it (6,042 in 2022, 5,960 in 2023), and
the swing is confined to the non-resident line while the domestic line keeps
climbing.
1 · The whole distribution, then and now
Every qualifying department in the country, bucketed by its non-resident share. If the story were "nothing changed", these two shapes would sit on top of each other.
Loading…
Departments with 20+ research doctorates that year. The lowest bin includes institutions that filed a total but left the residency breakdown blank, so read it as an upper bound rather than a count of genuinely domestic programmes.
2 · Which programmes, not which universities
Pooled across every institution in the country. The range is the finding: from petroleum engineering at the top to applied behaviour analysis at the bottom, this is a property of fields rather than of particular schools.
Loading…
Programmes with 300+ research doctorates nationally, 2015–2022. Note information security near the bottom while general computer science sits near the top — the most plausible reading is that clearance-adjacent work is effectively closed to non-citizens, but IPEDS records the degree, not why anyone was admitted.
3 · Inside one university
A single share per university misleads. Pick one and the spread appears: most institutions run from near zero to near total, department by department.
Loading…
Programmes with 15+ research doctorates over 2015–2022 at that institution. Programmes below the floor are omitted, so the count here is smaller than the institution's full programme list.