The headline finding: no statistical agency in any of the five markets publishes the number of postgraduates writing their thesis in a second language. Not Eurostat, not IPEDS, not HESA, not Destatis. The figure quoted in business cases is always mobility — students enrolled outside their country of origin — and mobility is wrong in both directions at once.
It undercounts, because a Dutch master’s candidate on an English-taught programme in Utrecht is writing in an acquired language and is not mobile. It overcounts, because an Austrian doctoral candidate in Germany is mobile and is writing in her first language.
Below is what is actually published, read first-hand on 23 August 2026, with the errors named so the figures can be used honestly in a paper rather than quietly misused in one.
The best published proxy: mobile students from abroad, ISCED 7 and 8
Eurostat’s dataset on mobile students from abroad counts students enrolled in a reporting country whose country of origin is elsewhere. Paired with total enrolment at the same level, it gives a share. Reference year 2023; the mobility table and the enrolment table were both last updated on 11 August 2026.
Master’s level (ISCED 7), 2023
| Country | Mobile from abroad | Total enrolled | Share |
|---|---|---|---|
| EU-27 | 786,433 | 5,517,114 | 14.3% |
| Netherlands | 53,139 | 195,387 | 27.2% |
| Germany | 221,295 | 1,129,883 | 19.6% |
| France | 141,361 | 1,004,770 | 14.1% |
| Sweden | 19,777 | 152,849 | 12.9% |
| Spain | 45,699 | 418,653 | 10.9% |
| Italy | 54,897 | 824,686 | 6.7% |
| Poland | 27,442 | 447,020 | 6.1% |
Doctoral level (ISCED 8), 2023
| Country | Mobile from abroad | Total enrolled | Share |
|---|---|---|---|
| EU-27 | 158,749 | 721,447 | 22.0% |
| France | 25,429 | 69,639 | 36.5% |
| Sweden | 6,948 | 19,283 | 36.0% |
| Germany | 48,061 | 205,302 | 23.4% |
| Spain | 19,652 | 97,227 | 20.2% |
| Italy | 5,874 | 42,688 | 13.8% |
| Poland | 1,002 | 23,454 | 4.3% |
| Netherlands | not reported | 40,612 | — |
The Dutch row is not an omission on our part. The query returns an enrolment figure and no mobility value at ISCED 8. If you have been quoted a Dutch doctoral international share, ask where it came from, because it did not come from this table.
The doctoral figures are roughly half again as high as the master’s figures in most countries, and in France and Sweden they are close to triple. That is the single most useful pattern here for anyone sizing writing support: the exposure concentrates at doctoral level, which is exactly where provision is thinnest, as set out in why your international cohort needs the most support and is offered the least.
The United States: a different question, answered by a membership survey
The US figure most often cited comes from the Institute of International Education’s Open Doors census. Reading the academic-level workbook directly on 23 August 2026, for the 2024/25 reporting year:
| Level | 2023/24 | 2024/25 | Change |
|---|---|---|---|
| Total graduate | 502,291 | 488,481 | −2.7% |
| Master’s | 319,402 | 306,123 | −4.2% |
| Doctoral | 149,618 | 153,418 | +2.5% |
The denominator trap, stated plainly. Open Doors also prints a percentage beside each of those counts — 41.5% for total graduate in 2024/25. That is the share of all international students who are studying at graduate level. It is not the international share of US graduate enrolment. Quoting it as though it were the second thing is the most common error we see in institutional business cases using this source, and it is the same class of denominator failure catalogued in the analysis of writing centre utilisation data.
Note also the divergence in the last row: master’s numbers fell while doctoral numbers rose. Any planning assumption that treats international postgraduate demand as one trend is already wrong for 2024/25.
The United Kingdom: a documented decline
We attempted to read the HESA student domicile pages on 23 August 2026 and received HTTP 403. We are therefore not quoting a UK figure. It exists and it is good; it simply was not readable by us on the day, and repeating a remembered number would be exactly the practice this desk exists to argue against.
The three reasons none of these numbers answers your question
1. English-medium instruction makes domestic students second-language writers
This is the largest single source of undercounting, and it is concentrated in exactly the countries with the highest mobility shares. A Dutch, Swedish or German candidate on an English-taught doctoral programme writes the thesis in an acquired language and is invisible to every mobility statistic. In a department where the working language is English and the domestic intake is substantial, mobility may capture less than half of the affected population.
2. Mobility counts first-language writers as mobile
The reverse error, and it is not small. Austrians in Germany, Belgians in France, Irish students in the UK, Latin American students in Spain and Portuguese-speaking students in Portugal are all mobile and all writing in a first language. In Spain, whose intake skews strongly towards Latin America, this correction runs in one direction and it is material.
3. Origin is not language
Country of origin is a passport question. A candidate educated in an English-medium international school in a non-English-speaking country may write English more comfortably than a domestic candidate. Any figure built on nationality is a proxy for a proxy, and it should be labelled as one every time it appears in a paper.
The proxy you can build, and the three inputs it needs
Because the published figure does not exist, the defensible move is to construct one internally and to state its construction. Three inputs, all of which your institution already holds:
- Language of prior degree-level education, not nationality. Most admissions systems capture the institution and country of the previous qualification. This is a far better proxy than domicile and it costs nothing to extract.
- Language of instruction of the programme, held in the programme catalogue. Cross this with input one and the picture changes sharply in English-medium programmes.
- Whether an English language qualification was required at admission, which is already recorded for visa and admissions purposes.
Combine them into a simple three-band flag — likely first language, likely second language, unclear — and publish the method alongside the number. A named proxy with a stated construction survives a governance committee. A borrowed national percentage does not.
Then disaggregate the three operational measures that matter against that flag: writing support waiting times, corrections rates at examination, and detector flag rates. If none of the three has ever been broken down this way, that absence is itself the finding worth putting in the paper.
How current is any of this?
The Eurostat figures above describe the 2023 reference year and were published in 2026. That is a normal lag for official education statistics and it is not a defect, but it does mean that a strategy paper written today using them is describing a cohort that has largely graduated. The lag by source type is set out in how current your higher education data actually is, and the same caution applies to completion and attrition evidence, mapped in where completion and attrition data comes from.
For the supply side that pairs with these numbers — how many academic staff exist to supervise the candidates counted here — see how many postgraduates and how many staff to supervise them. That article counts the same population from the capacity side; this one counts the language exposure inside it.
How to cite these figures
Name the dataset, the reference year, the extraction date and the exact filter. For example: Eurostat, mobile students from abroad enrolled by education level, sex and country of origin, reference year 2023, ISCED 8, world total excluding the reporting country, extracted 23 August 2026. A committee will accept that. It should not accept a percentage with no denominator attached, which is how most of these figures currently travel.
What follows operationally
If your doctoral cohort sits near the EU-27 mobility figure of 22.0%, and your programmes are taught in English, the realistic second-language share of your thesis writers is materially higher than 22% and you do not currently know by how much. That is a measurement task with a one-week cost and a provision decision behind it.
Tesify for Institutions is provision rather than measurement, but it removes the ability-to-pay variable from the outcome for the cohort this article is about, and it produces usage data broken down the way you need it. If you are sizing writing support against a cohort you have not yet measured, request an institutional evaluation and we will scope a free departmental pilot with the cohort breakdown agreed before it starts.
Frequently asked questions
How many postgraduates write their thesis in a second language?
No statistical agency publishes this. The nearest published proxy is student mobility, which showed 22.0% of EU-27 doctoral enrolment and 14.3% of master’s enrolment coming from abroad in 2023, but mobility both undercounts and overcounts second-language writers.
Which country has the highest share of international doctoral candidates?
Of the countries examined, France at 36.5% and Sweden at 36.0% in 2023, followed by Germany at 23.4% and Spain at 20.2%. The EU-27 figure was 22.0%. Source: Eurostat, reference year 2023, extracted 23 August 2026.
Why is there no Dutch doctoral mobility figure?
The Eurostat query returns a total enrolment figure of 40,612 for the Netherlands at doctoral level in 2023 and no mobility value. If you have been given a Dutch doctoral international share, ask for its source, because it did not come from this table.
Why does mobility undercount second-language writers?
Because English-medium instruction makes domestic students second-language writers too. A Dutch, Swedish or German candidate on an English-taught programme writes the thesis in an acquired language and appears in no mobility statistic at all.
Why does mobility also overcount them?
Because many mobile students write in their first language. Austrians in Germany, Belgians in France, Irish students in the UK and Latin American students in Spain are all counted as mobile and all writing in a first language.
What does the 41.5% figure in Open Doors actually mean?
It is the share of all international students in the United States who are studying at graduate level. It is not the international share of US graduate enrolment, and quoting it as the second thing is the most common error we see in business cases using this source.
Did US international graduate numbers rise or fall?
Both, depending on level. In 2024/25 the total graduate figure fell to 488,481 from 502,291, master’s fell 4.2% to 306,123, and doctoral rose 2.5% to 153,418. Treating international postgraduate demand as a single trend is already wrong for that year.
Why is there no UK figure in this article?
Because the HESA pages returned HTTP 403 to our fetch on 23 August 2026. The data exists and is good; it was simply not readable by us on the day, and quoting a remembered number would be worse than leaving the row blank.
What is a better internal proxy than nationality?
Language of prior degree-level education, crossed with the language of instruction of the programme and with whether an English language qualification was required at admission. All three are already held in admissions and catalogue systems.
How should we present the proxy to a committee?
As a three-band flag with its construction stated: likely first language, likely second language, unclear. A named proxy with a published method survives scrutiny; a borrowed national percentage does not.
How current are the Eurostat figures?
They describe the 2023 reference year and were last updated on 11 August 2026. That is a normal lag for official education statistics, but it means a paper written today with them describes a cohort that has largely graduated.
What should we measure once we have the flag?
Writing support waiting times, corrections rates at examination and detector flag rates, each disaggregated against the flag. If none has ever been broken down this way, that absence is itself a finding worth reporting.