Documentation

Who controls the most internationally prominent universities of each country, and what does a domestic student pay to attend them? This dataset codes the top universities of 63 countries and territories — 376 institutions, selected by three rankings (below) — for control (public or private), the standard annual charge to a domestic first-degree student relative to GDP per capita, and the existence of means-tested support, by who funds it and how it is paid. Every institution is backed by documented sources (3,948 in total), each with its URL, access date and a verbatim quote.

1. Sample
2. What is measured
3. Country summaries
4. How the data were collected and checked
5. Main decisions
6. Admissions: selectivity and formal equity
7. Limitations and known issues
8. Files
9. Variable dictionary
10. Versions and updates

1. Sample

Three samples, each fixed before any price of its new institutions was collected. The two QS 2027 samples cover 63 countries and territories; the QS 2024 sample covers the 55 of the earlier releases. In every sample, all institutions tied at the fifth position are included and countries with fewer than five ranked institutions keep all they have.

• QS 2027, overall rank (the default in the visualisation): the five highest-ranked institutions per country in the QS World University Rankings 2027 (published June 2026).
• QS 2027, academic reputation: the five institutions per country with the highest QS Academic Reputation score, from QS's survey of academics. This is arguably closer to what "elite" means, because the overall rank also rewards indicators such as the faculty/student ratio on which small institutions score high.
• QS 2024, overall rank: the sample of the first release (QS 2024, published 27 June 2023), kept as a robustness check.

The two 2027 samples share 268 institutions; the overall-rank sample shares 248 with QS 2024 and the reputation sample 225. The dataset holds the union of the three. Aid and enrolment refer to 2023/24; prices are given both for 2023/24 and for the latest year (section 2). The 2027 samples use a recent ranking to choose the institutions, not to date the prices. Which institutions are "top 5" can change a lot across editions: in Uruguay, for example, the Universidad de Montevideo was first in QS 2024 and the Universidad de la República first in QS 2027.

The sample is internationally ranked research universities, not the universe of nationally selective programmes. The countries were chosen by the investigators to cover every world region and a range of higher-education models; they are not a random sample. Hong Kong SAR is a separate unit; Taiwan is included; Macau is not. Countries: Argentina, Australia, Austria, Azerbaijan, Belgium, Brazil, Canada, Chile, China, Colombia, Croatia, Czechia, Denmark, Ecuador, Egypt, Estonia, Finland, France, Germany, Greece, Hong Kong, Hungary, India, Indonesia, Ireland, Israel, Italy, Japan, Kazakhstan, Kyrgyzstan, Lithuania, Malaysia, Mexico, Netherlands, New Zealand, Nigeria, Norway, Oman, Pakistan, Paraguay, Peru, Philippines, Poland, Portugal, Romania, Saudi Arabia, Singapore, Slovakia, Slovenia, South Africa, South Korea, Spain, Sweden, Switzerland, Taiwan, Thailand, Turkey, Ukraine, United Arab Emirates, United Kingdom, United States, Uruguay, Uzbekistan.

2. What is measured

Reference student and year. A citizen, locally resident, full-time student in a first undergraduate degree, at a standard load, in academic year 2023/24 (or calendar 2023). Where a source for that year could not be found and the same regime clearly applied, the nearest year is used and flagged (36 institutions carry a year label other than 2023/24 or calendar 2023; see year_observed).

Control (strict). An institution is public if the state owns it (a public-law body created by the state, with no private founder) or appoints a majority of its governing body; otherwise it is private (non-profit or for-profit). Public funding and academic autonomy do not enter this definition. Separately, a government-dependence flag marks private institutions that receive substantial public funds explicitly for tuition or teaching (for example per-student teaching grants or the state paying students' fees). Research funding and student loans do not count.

Headline charge. Standard annual tuition plus compulsory recurring fees paid by every reference student, after reductions that apply to all students in every year of study, and before any targeted aid. Housing, health insurance, application and entrance-exam fees are excluded. Where prices vary by programme, the headline is the midpoint of the published range (minimum and maximum are kept); where they depend on family income, it is the full-pay rate. Reductions limited to the first year are not part of the headline.

Price bands. The headline charge divided by GDP per capita 2023 in current local currency (World Bank; for Taiwan, the official national accounts), grouped in bands fixed before coding: free < 1%, low 1–5%, low–moderate 5–10%, moderate 10–25%, high ≥ 25%. An alternative set of cut-offs (2.5 / 5 / 15 / 30%) is reported as a sensitivity check.

Two price years. Every institution has a 2023/24 price (columns without prefix) and a latest price (latest_* columns), collected in September–October 2026 with the same rules for academic year 2025/26 or calendar 2025. Each price is divided by GDP per capita of its own year (World Bank, July 2026 release; Taiwan from its national accounts). The latest view has a band for 367 institutions: 343 with a 2025/26 price and 24 that use an earlier year’s price with that year’s GDP per capita (no 2025/26 price could be documented, or no GDP or exchange rate exists for its year; marked * and explained in latest_year_flag). 13 institutions are in a different band in the two views. Control, aid and enrolment are not recoded in the latest view. Its 1,265 sources are in price_sources.csv; the independent checks of the rows that changed band or rules, corrected a 2023/24 figure, or belong to the countries added in 2026 (80 rows) are in price_verification.csv.

Means-tested support. Non-repayable support whose eligibility depends on household or parental income, coded in four cells: funded by the government or run by the institution, times paid as a fee reduction (waiver, income-scaled fee, or grant paid against the charge) or as cash to the student (for example living-cost grants). Each cell is yes, no (searched, not found), unknown, or not_applicable (a fee reduction where the charge is zero). Loans, in-kind support and merit-only awards are excluded. Aid is never subtracted from the headline charge. Where documented, the share of students receiving each instrument is recorded.

QS classification (context only). QS publishes, for each ranked institution, a Status (public / private not for profit / private for profit), a size class, a subject focus and a research intensity; they are released as context variables (qs_*) and are not used in any coding. QS Status is not our control variable: QS "Public" also covers private universities that depend on public funding. Of the institutions classified as private under the strict test, 18 are "Public" in QS (for example Oxford, Cambridge, UCL, Imperial, the Israeli universities and Hungary's foundation universities); 5 institutions public under the strict test are "Private" in QS.

Enrolment. One figure per institution for 2023/24 or 2023: first-degree headcount where published (242 institutions), otherwise undergraduate headcount, total headcount or full-time equivalents (flagged; 114 institutions). Found for 356 of 376 institutions. Enrolment is used only for weighting within countries.

3. Country summaries

3×3 grid. Institutions are placed by control (public; private and government-dependent; private and not government-dependent) and price (free; low to low–moderate; moderate or high). A country is placed in the cell that holds the largest number of its priced institutions (unweighted) or the largest share of their enrolment (enrolment-weighted); a country tied between two cells counts half in each. Institutions without an established price are left out.

Country type (rule fixed in advance). Control type from the share of public institutions: public if at least 0.7 (4 or 5 of 5), mixed if 0.3 to 0.7, private below 0.3. Price type is the median band; it is marked incomplete when a missing price could move the median. Each country type is recomputed under several alternative definitions (top 1 and top 3 only; alternative band cut-offs; government-dependent private counted as public; enrolment weights; the low and high ends of programme ranges; pricing the free track where free places are rationed by exam score; setting the price to zero where a documented majority of entrants pay nothing). These are the sens_* columns of countries.csv.

Weights. Summary shares are shown pooled (each institution counts once), per country (country shares averaged with equal weight) and per country, enrolment-weighted. The association with income inequality uses the World Bank Poverty and Inequality Platform Gini (latest national survey 2013–2023) and is descriptive only.

4. How the data were collected and checked

Collection. The data were collected with the help of AI research agents (Anthropic's Claude) working under a written protocol set by the investigators. Collectors were given the definitions and the distinctions to keep separate, and were never told what result to expect for a country. Every fact had to be supported by an opened source: the exact URL, a verbatim quote in the original language, a locator and the access date. Where a live page had changed since 2023/24, contemporaneous archived captures (Internet Archive) were used. Unknown is recorded as unknown: a missing price or a missing grant is never read as zero.

Independent check. Each institution was checked once by a second agent working in a fresh context, which re-opened the cited sources (and others where needed) and asked whether they give the same control and the same price band. Exact figures need not match; categories must. Control agreed for 373 of 376 institutions and the band for 361; the exceptions are listed in section 6. Disagreements were resolved with evidence or left unresolved, and every correction was applied by script, so all tables can be rebuilt from the recorded evidence. The checker is the same kind of model as the collector: a fresh context reduces anchoring, but errors need not be independent.

Human review. The investigators set the protocol and made every methodological decision (section 5), reviewed contested cases and commissioned a skeptical review of the country-level conclusions. Not every individual value has been re-verified by hand. That is one reason every source is published: anyone can follow a coding back to the page it rests on.

5. Main decisions

All 76 decisions, with the options considered and the rationale, are in decision_log.csv (early entries concern the organisation of the work). The ones that shape the results:

6. Admissions: selectivity and formal equity

For the 308 institutions of the QS 2027 academic-reputation sample, the release adds how first-year, first-degree entry worked for 2023/24 entry (or the nearest year, flagged): applicants, admitted, entrants and places, the admission regime, and whether the institution applied a formal equity admission rule.

Formal equity admission (equity_formal) means an explicit rule that reserves places, adds points, lowers the admission threshold or opens a separate entry track for applicants from groups defined by disadvantage strongly correlated with income: household income or socioeconomic status, race or ethnicity (including Indigenous peoples), caste or tribe, public or under-represented schools, rural or remote origin and disadvantaged regions, first generation, care leavers and orphans. It is not: holistic “considered” factors; individual hardship clauses; channels that apply the same criteria (QuestBridge); quotas defined by territory, religion or language (e.g. Greenland, Azores and Madeira, the Thrace Muslim minority, the Sámi language quota, occupied or earthquake regions) or by gender; merit, athlete, legacy, veteran, international or service-obligation routes; and financial aid. Rules for applicants with disabilities are recorded separately (equity_disability_rule) and do not make equity_formal yes (decisions D066–D074). equity_scope says whether the rule covers first-degree entry across the institution, only named programmes, or only an access or foundation programme. Reserved and filled counts are given with their own denominators; a share is reported only as stated by the source or computed from stated counts.

Rates. Admission rate (admitted / applicants) and entry rate (entrants / applicants) are released only where the independent check found the figures on the same base (population, year and round): 63 institutions. Applicant counts are often applications (one person counted once per programme) rather than persons (sel_unit), and under central allocation (e.g. Greece, Turkey, China) institutions have no applicant count at all.

Collection and check. One collector run per country, then an independent checker per group of countries that coded equity blind before seeing the draft and re-opened the main source of every figure: equity agreed for 289 of 308 institutions before reconciliation; 188 corrections were accepted by the coordinator (admissions_verification.csv, sources in admissions_sources.csv).

Which institutions have formal equity rules? 136 of the 303 institutions with a known code (45%); 5 remain unknown. By group (shares over known codes; full tables by price band, QS Status, admission regime, rank position, country type and region in equity_by_type.csv):

Basis among institutions with a rule (several possible): income or socioeconomic 82, region or rural 60, other 52, race ethnicity 51, public school 17, caste tribe 5, first generation 2. Instruments: reserved places 101, separate track 64, lower cutoff 51, bonus points 19, contextual offer 3. Countries where all sampled institutions have a rule: Australia, Brazil, Canada, Chile, China, France, Hungary, India, Indonesia, Ireland, Kazakhstan, Malaysia, New Zealand, Nigeria, Paraguay, Portugal, Romania, South Korea, Taiwan, Ukraine, United Kingdom, Uzbekistan. Countries where none has: Argentina, Austria, Azerbaijan, Belgium, Czechia, Denmark, Egypt, Estonia, Finland, Germany, Greece, Hong Kong, Italy, Japan, Lithuania, Mexico, Netherlands, Norway, Poland, Singapore, Slovakia, Slovenia, Sweden, Switzerland, Turkey, United Arab Emirates, United States, Uruguay.

Country characteristics. At the country level (share of each country's sampled institutions with a rule), formal equity rules are more common in poorer countries: Spearman rank correlation with GDP per capita (PPP, 2023) -0.40, with the Gini index 0.16. By thirds of countries:

GDP per capita: World Bank WDI (GDP per capita in local currency / PPP conversion factor), Taiwan not available. Gini: World Bank PIP, as in section 3; six countries have no Gini, and six of the least unequal third report consumption-based Ginis, which run lower than income-based ones. Tables in equity_by_gdp_tercile.csv, equity_by_gini_tercile.csv and equity_country_correlates.csv. With about 60 countries and strong regional clustering these are associations, not effects.

These are descriptive counts for a ranking-defined sample, not estimates for each country's system; within a country, institutions usually share one national rule, so countries, not institutions, are the effective units.

7. Limitations and known issues

• Which universities. The sample is defined by an international research ranking. It says nothing directly about the rest of each system (non-ranked public universities, vocational and short-cycle institutions), where most students often study. Five institutions per country is a small window.
• Which countries. 63 countries and territories chosen by the investigators, not a random sample. Countries with fewer than five ranked institutions, or more because of ties, are listed per sample in section 1.
• Which ranking. The top 5 depends on the ranking edition and criterion (section 1). The three samples are reported side by side so that results can be compared across definitions; none of them is the universe of selective institutions.
• Sticker price, not net price. The headline charge is what a student pays before targeted aid. For low-income students at institutions with generous means-tested aid (for example elite US private universities) the net price can be far lower. Aid is coded as present or absent, not by generosity; coverage shares are recorded only where documented. 569 of the 376×4 aid cells are unknown, so shares of institutions with aid are lower bounds.
• One reference student. Charges for international students, part-time students, second degrees and separately priced tracks are not covered.
• Programme variation. Where prices vary by programme, the midpoint can sit in a different band from the cheapest or most expensive programmes (band_unstable, and the low-end / high-end sensitivities).
• Denominator. GDP per capita in current local currency. For Ireland it is inflated by multinational accounting, which lowers Irish ratios; purchasing-power comparisons across countries are not the aim.
• Close to a cut-off. 50 institutions sit within 10% (relative) of a band threshold, so a small change in price or GDP would move them.
  Show the 50 institutions

  • University of Cambridge (United Kingdom, 23.0% vs 25%)
  • University of Oxford (United Kingdom, 23.0% vs 25%)
  • Imperial College London (United Kingdom, 23.0% vs 25%)
  • University College London (United Kingdom, 23.0% vs 25%)
  • The University of Tokyo (Japan, 10.8% vs 10%)
  • Kyoto University (Japan, 10.8% vs 10%)
  • Osaka University (Japan, 10.8% vs 10%)
  • Tohoku University (Japan, 10.8% vs 10%)
  • The University of Melbourne (Australia, 10.3% vs 10%)
  • The University of New South Wales (UNSW Sydney) (Australia, 10.3% vs 10%)
  • The University of Sydney (Australia, 10.3% vs 10%)
  • Australian National University (ANU) (Australia, 10.3% vs 10%)
  • Monash University (Australia, 10.3% vs 10%)
  • University of Otago (New Zealand, 10.5% vs 10%)
  • Victoria University of Wellington (New Zealand, 10.5% vs 10%)
  • University of Waikato (New Zealand, 10.1% vs 10%)
  • Singapore Management University (Singapore, 10.5% vs 10%)
  • Politecnico di Milano (Italy, 10.7% vs 10%)
  • University of Lagos (Nigeria, 9.8% vs 10%)
  • Cairo University (Egypt, 5.4% vs 5%)
  • Yonsei University (South Korea, 22.6% vs 25%)
  • Vilnius University (Lithuania, 22.9% vs 25%)
  • National Tsing Hua University (Taiwan, 5.4% vs 5%)
  • National Taiwan University of Science and Technology (Taiwan Tech) (Taiwan, 5.2% vs 5%)
  • University of Bucharest (Romania, 5.2% vs 5%)
  • Stefan cel Mare University of Suceava (Romania, 4.9% vs 5%)
  • Budapest University of Technology and Economics (Hungary, 10.2% vs 10%)
  • The University of Hong Kong (Hong Kong, 10.7% vs 10%)
  • The Chinese University of Hong Kong (CUHK) (Hong Kong, 10.6% vs 10%)
  • The Hong Kong University of Science and Technology (Hong Kong, 10.6% vs 10%)
  • The Hong Kong Polytechnic University (Hong Kong, 10.6% vs 10%)
  • City University of Hong Kong (Hong Kong, 10.6% vs 10%)
  • Universiti Sains Malaysia (USM) (Malaysia, 5.0% vs 5%)
  • Universiti Teknologi Malaysia (Malaysia, 5.0% vs 5%)
  • Mahidol University (Thailand, 26.7% vs 25%)
  • Thammasat University (Thailand, 25.9% vs 25%)
  • Université de Montréal (Canada, 5.1% vs 5%)
  • University of Geneva (Switzerland, 1.1% vs 1%)
  • University of Milan (Università degli Studi di Milano) (Italy, 10.0% vs 10%)
  • University of Canterbury (Te Whare Wānanga o Waitaha) (New Zealand, 10.2% vs 10%)
  • National Taiwan Normal University (NTNU) (Taiwan, 5.2% vs 5%)
  • Baku State University (Azerbaijan, 26.8% vs 25%)
  • Auezov South Kazakhstan University (SKU) (Kazakhstan, 9.4% vs 10%)
  • Kyrgyz Russian Slavic University (Kyrgyzstan, 25.1% vs 25%)
  • Dhofar University (Oman, 26.0% vs 25%)
  • National University of Uzbekistan named after Mirzo Ulugbek (Uzbekistan, 22.9% vs 25%)
  • Samarkand State University (Uzbekistan, 22.9% vs 25%)
  • Tashkent Institute of Irrigation and Agricultural Mechanization Engineers - National Research University (TIIAME-NRU) (Uzbekistan, 22.9% vs 25%)
  • Tashkent State Transport University (Uzbekistan, 22.9% vs 25%)
  • Tashkent State University of Economics (Uzbekistan, 23.9% vs 25%)
• No established price. 10 institutions (excluded from every price statistic):
  • Austral University (Argentina)
  • Ain Shams University (Egypt)
  • Assiut University (Egypt)
  • Mansoura University (Egypt)
  • Universidad de Montevideo (Uruguay)
  • Universidad ORT Uruguay (Uruguay)
  • Universidad Católica del Uruguay (Uruguay)
  • Universidad Espíritu Santo (UEES), Ecuador (Ecuador)
  • University of Nigeria Nsukka (Nigeria)
  • The University of Lahore (Pakistan)
• Checks that did not agree. The independent check did not confirm the band for:
  • Austral University (Argentina): band unresolved
  • University of Palermo (Argentina): band unresolved
  • Ain Shams University (Egypt): band unresolved
  • Assiut University (Egypt): band unresolved
  • Mansoura University (Egypt): band unresolved
  • University POLITEHNICA of Bucharest (Romania): band disagree
  • Universidad de Montevideo (Uruguay): band unresolved
  • Universidad ORT Uruguay (Uruguay): band unresolved
  • Universidad Católica del Uruguay (Uruguay): band unresolved
  • Universidad Espíritu Santo (UEES), Ecuador (Ecuador): band unresolved
  • University of Nigeria Nsukka (Nigeria): band unresolved
  • The University of Lahore (Pakistan): band unresolved
  • Khalifa University (United Arab Emirates): band unresolved
  • Baku State University (Azerbaijan): band unresolved
  • Auezov South Kazakhstan University (SKU) (Kazakhstan): band disagree
• Contested control. Some classifications turn on legal details and are reported with sensitivities: English universities are private non-profit and not flagged as government-dependent (borderline); Edinburgh and McGill are private non-profit and flagged; the Israeli universities and Hungary's foundation universities are private and flagged; Antwerp's control decides whether Belgium is mixed or private. Board composition was not documented systematically, so a stricter test requiring a state-appointed board majority is not reported.
• Latest prices. Some latest prices rest on weaker evidence than the 2023/24 rows (press reports, a university's statement that fees were unchanged, or estimates), flagged with low confidence in latest_confidence. Prices in high-inflation economies (Argentina, Turkey, Egypt) depend strongly on the month observed; the price and GDP years are matched, but not the months.
• Gaps from the 2027 update. The institutions added from QS 2027 were collected and checked in September 2026. Some have no established price (a fee schedule was not published or not reachable; see the list above), and a few prices rest on weaker evidence (a nearest year, a news report or field averages), flagged with low confidence in confidence and explained in notes. Part of the checking of these additions was done without web search: only the cited and directly reachable sources were re-opened.
• Open items in this release. Whether Saudi Arabia's Prince Mohammad bin Fahd University should be flagged as government-dependent (the share of its entrants paid by the Ministry is unknown; currently not flagged). POSTECH is coded free under the rule for tuition waived unless a student falls behind; this application is under review. Khalifa University (UAE) is coded free for nationals, but its scholarship tiers may leave some Emiratis paying (AED 82,500 a year); Baku State University is high on its main campus and moderate if its Qazax branch is counted; who appoints the University of Nizwa's board is not documented (counted in the “unresolved as public” sensitivity). Four of the six Uzbek prices rest on the national base contract schedule, which the universities were required to keep in 2023/24, not on their own schedules.
• Enrolment. Measures differ across institutions (first-degree, undergraduate, total headcount or FTE), and some counts are course enrolments rather than students. Enrolment is missing for 376−356 institutions; a country with any such institution in a sample is left out of that sample's weighted results rather than imputed.
• Inequality. The Gini comparison is an association across a small number of countries with strong regional clustering, not an effect. Consumption-based Ginis (China, Egypt, India, Indonesia, Kazakhstan, Kyrgyzstan, Nigeria, Pakistan, Philippines, South Africa, Thailand, Ukraine) are lower than income-based ones; Azerbaijan, Hong Kong, New Zealand, Oman, Saudi Arabia, Singapore have no survey on the platform; Argentina has only an urban survey.
• Admissions variables. Many institutions publish no applicant or offer counts; figures come from different sources (institutional reports, national statistics, admission systems) and scopes, documented in sel_scope and adm_notes. Equity rules were read for 2023/24 entry where an archived or dated text was available; some rest on the nearest year or on secondary sources (flagged in notes). Five institutions remain unknown because their sources were blocked or not found. Finance (revenue by source) was piloted but is not released.
• Quotes. Source quotes longer than 400 characters are shortened in this release (1067 quotes, marked in quote_truncated); the URLs lead to the full text.

8. Files

All files are UTF-8 CSV. Download everything (.zip), which also includes this documentation and the visualisation.

Institutions link to sources, verification and the visualisation through institution_id; countries through country_iso3. Decision IDs (D0nn) and run tags in notes refer to decision_log.csv and verification.csv.

9. Variable dictionary

The main variables of institutions.csv and countries.csv. The full dictionary, including sources, verification and the decision log, is in dictionary.csv.

Show the dictionary

Variable	Definition	Values
institution_id	Stable identifier: ISO3 country code + short institution code.	text
country_iso3	Country or territory (ISO 3166-1 alpha-3; HKG = Hong Kong SAR, TWN = Taiwan).	text
country_name	Country or territory name.	text
institution	Institution name in English, as listed by QS.	text
in_qs2027_overall	In the sample 'QS 2027, top 5 per country by overall rank' (ties at the fifth position included).	yes | no
in_qs2027_ar	In the sample 'QS 2027, top 5 per country by Academic Reputation score' (ties included).	yes | no
in_qs2024	In the sample 'QS 2024, top 5 per country by overall rank' (the v1.0 sample).	yes | no
qs2027_rank	Overall rank as printed in the QS World University Rankings 2027 (published June 2026); '=' marks a tie, ranges are rank bands; blank = not ranked in 2027.	text
qs2027_ar_score	QS 2027 Academic Reputation score (0-100), the basis of the reputation sample.	number
qs2027_pos_overall	Position within the country by QS 2027 overall rank (rank bands in QS order).	integer
qs2027_pos_ar	Position within the country by QS 2027 Academic Reputation score.	integer
qs2024_rank	Rank as printed in the QS World University Rankings 2024 (published 27 June 2023), for institutions of the QS 2024 sample.	text
qs_status	QS 2027 'Status' classification, as published by QS (not our control variable; see documentation).	Public | Private not for Profit | Private for Profit | blank
qs_size	QS 2027 size class (student body): S, M, L, XL.	S | M | L | XL
qs_focus	QS 2027 subject focus: FC fully comprehensive, CO comprehensive, FO focused, SP specialist.	FC | CO | FO | SP
qs_research	QS 2027 research intensity: VH very high, HI high, MD medium, LO low.	VH | HI | MD | LO
control	Strict control: public if the state owns the institution (a public-law body it created, no private founder) or appoints a majority of its governing body; otherwise private. Public funding and academic autonomy do not enter.	public | private_nonprofit | private_forprofit
control_basis	What establishes the control code: state ownership, state appointment of a board majority, both, or (private institutions) the legal form.	ownership | appointment | ownership_and_appointment | legal_form
govt_dependent_flag	Private institutions only: yes if the government transfers substantial resources explicitly to cover tuition or teaching (per-student teaching grants, the state paying students' fees). Research funding and student loans do not count. Always 'no' for public institutions.	yes | no
control_group	Row of the 3x3 grid: public; private and government-dependent; private and not government-dependent.	public | private_govt_dependent | private_not_govt_dependent
tuition_std	Standard annual tuition for the reference student, local currency.	number (local currency)
compulsory_fees_std	Other annual charges every reference student must pay (registration, student contribution, student-service fees, compulsory levies). Excludes housing, insurance, application and entrance-exam fees, optional items.	number (local currency)
headline_charge	tuition_std + compulsory_fees_std: the standard annual charge after reductions that apply to all students in every year, before any targeted aid. Blank = not established.	number (local currency)
currency	Currency of the charges (ISO 4217).	text
averaging_rule	How the headline was set when prices vary: one rate; midpoint of the published range across programmes; full-pay rate of an income-based scale.	uniform | midpoint_of_range | full_pay_rate_of_scale
charge_min	Lowest charge in the range (programmes or scale points), when not uniform.	number (local currency)
charge_max	Highest charge in the range, when not uniform.	number (local currency)
gdp_pc_2023_lcu	GDP per capita 2023, current local currency (World Bank WDI NY.GDP.PCAP.CN; Taiwan from DGBAS national accounts).	number (local currency)
charge_ratio	headline_charge / gdp_pc_2023_lcu.	share (0.05 = 5%)
charge_band	Band of charge_ratio, fixed before coding: free < 1%; low 1-5%; low_moderate 5-10%; moderate 10-25%; high >= 25%.	free | low | low_moderate | moderate | high | blank
price_group	Column of the 3x3 grid: free; low to low-moderate (1-10%); moderate or high (>= 10%).	free | low_to_low_moderate | moderate_or_high | blank
band_min	Band of charge_min.	band
band_max	Band of charge_max.	band
band_unstable	yes if charge_min and charge_max fall in different bands.	yes | no
charge_band_alt	Band under the alternative cut-offs 2.5 / 5 / 15 / 30% (sensitivity).	band
govt_fee_aid	Government-funded means-tested (household income) waiver, income-scaled fee or grant paid against the charge. no = searched, not found; not_applicable = the headline charge is zero.	yes | no | unknown | not_applicable
govt_cash_aid	Government-funded means-tested non-repayable cash paid to the student (e.g. living-cost grants).	yes | no | unknown
inst_fee_aid	Institution-run means-tested fee reduction (own waivers, income scales, need-based scholarships against the charge; donor-funded schemes run by the institution count).	yes | no | unknown | not_applicable
inst_cash_aid	Institution-run means-tested cash paid to the student.	yes | no | unknown
govt_fee_share_pct	Documented share of students receiving govt_fee_aid (national or institutional, see notes). Blank = not found.	percent
govt_cash_share_pct	Documented share of students receiving govt_cash_aid.	percent
inst_fee_share_pct	Documented share of students receiving inst_fee_aid.	percent
inst_cash_share_pct	Documented share of students receiving inst_cash_aid.	percent
zero_pay_entrant_share_pct	Documented share of first-year entrants who pay nothing (state-funded places, gratuidad, full income-based exemption). Used only in the majority-of-entrants sensitivity.	percent
year_observed	Academic year the charge applies to. Target 2023/24 (or calendar 2023); a nearest year is flagged in the text.	text
confidence	high = official source for the year and a uniform or clear rule; medium; low.	high | medium | low
control_check	Result of the independent check of control.	agree | disagree | unresolved
band_check	Result of the independent check of the charge band.	agree | disagree | unresolved
checker_run	Identifier of the check run (see verification.csv).	text
enrolment	Enrolment used for weights, 2023/24 or calendar 2023.	number of students
enrolment_measure	first_degree_headcount where published; otherwise the flagged fallback.	first_degree_headcount | undergraduate_headcount | total_headcount | fte | blank
enrolment_year	Year of the enrolment figure, as stated by the source.	text
enrolment_confidence	Confidence in the enrolment figure.	high | medium | low
enrolment_url	Source of the enrolment figure.	URL
enrolment_note	Scope of the enrolment figure (what is counted, derivations).	text
n_sources	Number of documented sources for the institution in sources.csv.	integer
latest_academic_year	Academic year of the latest documented price (target 2025/26 or calendar 2025).	text
latest_headline_charge	Latest headline charge (same definition as headline_charge), local currency. Rows with latest_year_flag use an earlier year.	number (local currency)
latest_charge_min	Lowest charge in the latest range.	number (local currency)
latest_charge_max	Highest charge in the latest range.	number (local currency)
latest_price_year	Year of the price and of the GDP per capita used for the latest ratio (2025, or earlier where flagged; D058, D060, D062).	2023 | 2024 | 2025
latest_gdp_pc_lcu	GDP per capita of latest_price_year, current local currency (World Bank WDI, July 2026 vintage; Taiwan DGBAS).	number (local currency)
latest_charge_ratio	latest_headline_charge / latest_gdp_pc_lcu.	share
latest_charge_band	Band of latest_charge_ratio (same cut-offs).	band
latest_band_unstable	yes if the latest minimum and maximum fall in different bands.	yes | no
latest_year_flag	Set when the latest view uses an earlier price year (shown with * in the visualisation and figures).	text
change_vs_2023_24	How the latest price relates to the 2023/24 row.	no_change | price_update | regime_change | correction_of_2023_24 | not_established
latest_confidence	Confidence in the latest price.	high | medium | low
notes	Justification of the coding, including checker and recode notes (tags such as [R103] identify the run that added the text).	text
sample	Sample the row is computed on (one row per sample and country).	qs2027_overall | qs2027_ar | qs2024
country_iso3	Country or territory code.	text
country_name	Country or territory name.	text
n_institutions	Institutions of the country in this sample (top 5 with ties; fewer where QS lists fewer).	integer
n_priced	Selected institutions with an established headline charge.	integer
public_share	Share of selected institutions with control = public (equal weights).	share
free_share	Share of priced institutions in the free band.	share
free_or_low_share	Share of priced institutions in the free or low band.	share
public_and_free_or_low_share	Share of selected institutions that are public and free or low.	share
control_type	public if public_share >= 0.7; mixed if 0.3-0.7; private if < 0.3.	public | mixed | private
price_type	Median charge band; 'incomplete(a..b)' when a missing headline could move the median between a and b.	band | incomplete(..)
country_type	control_type-price_type (rule fixed before coding).	text
latest_price_type	Price type computed with the latest prices (2025/26; earlier years where flagged).	band | incomplete(...)
latest_country_type	Country type computed with the latest prices.	text
grid_cell_unweighted	3x3 cell holding the plurality of the country's priced institutions (control_group|price_group); ties list both cells separated by ' / '.	text
grid_cell_enrolment_weighted	3x3 cell holding the plurality of the enrolment of the country's priced institutions. Blank = not computable (an institution lacks enrolment).	text
n_govt_fee_aid_yes	Institutions with govt_fee_aid = yes (similarly n_govt_cash_aid_yes, n_inst_fee_aid_yes, n_inst_cash_aid_yes).	integer
n_govt_fee_aid_unknown	Institutions with govt_fee_aid = unknown (similarly for the other three aid variables).	integer
n_band_unstable	Institutions whose range spans two bands.	integer
n_low_confidence	Institutions coded with low confidence.	integer
sens_top1	Country type using only the top-ranked institution.	text
sens_top3	Country type using the top 3.	text
sens_alt_bands	Country type with band cut-offs 2.5 / 5 / 15 / 30%.	text
sens_govdep_as_public	Country type counting government-dependent private institutions as public.	text
sens_enrolment_weighted	Country type with enrolment weights (public share weighted; price = weighted median band). Blank = not computable.	text
sens_range_low_end	Country type using charge_min where prices vary.	text
sens_range_high_end	Country type using charge_max where prices vary.	text
sens_free_track_headline	Country type pricing exam-rationed free places (instead of fee-paying places) where both exist.	text
sens_majority_entrants_price	Country type setting the headline to 0 where a documented majority of entrants pay nothing.	text
gdp_pc_2023_lcu	GDP per capita 2023, current local currency.	number
gini	Gini index, World Bank Poverty and Inequality Platform: latest national survey 2013-2023.	0-100
gini_year	Survey reporting year.	year
gini_welfare_type	income or consumption (consumption-based Ginis are lower).	income | consumption
gini_coverage	Survey coverage.	national | urban
gini_flag	Reason the Gini is excluded from fits (e.g. non-national coverage).	text
selectivity_regime	Admissions (QS 2027 academic-reputation sample only, 2023/24 entry): how first-year first-degree places are allocated.	own_admission | central_allocation | mixed | open_admission
adm_ref_year	Entry year or cycle the admissions figures refer to (nearest year flagged).	text
applicants	Applicants for first-year first-degree entry (unit in sel_unit; scope in sel_scope).	number
admitted	Offers or admissions for first-year first-degree entry.	number
entrants	Students who enrolled (first-year first-degree entry).	number
places	Places offered or approved intake capacity.	number
sel_unit	Whether applicants count persons or applications (a person applying to several programmes counts several times).	unique_applicants | applications | not_applicable | blank
sel_scope	Scope of the admissions figures (campuses, levels, routes, nationalities, sources).	text
sel_confidence	Collector's confidence in the admissions figures.	high | medium | low
rate_comparability	Independent check: are applicants, admitted and entrants on the same base? Rates are released only if ok.	ok | mixed_bases | na
admission_rate	admitted / applicants, computed only where rate_comparability = ok.	0-1
entry_rate	entrants / applicants, computed only where rate_comparability = ok.	0-1
equity_formal	Formal equity admission rule for 2023/24 entry for groups defined by income-correlated disadvantage (documentation section 6; disability-only rules excluded).	yes | no | unknown | blank (not in sample)
equity_scope	Coverage of the equity rule.	all_programmes | some_programmes | access_programme_only
equity_basis	Groups targeted (several separated by ;).	income_or_socioeconomic; race_ethnicity; caste_tribe; public_school; region_or_rural; first_generation; other
equity_instrument	How the rule works (several separated by ;).	reserved_places; bonus_points; lower_cutoff; separate_track; contextual_offer
equity_mandate	Who sets the rule.	national_law; government_regulation; institutional
equity_reserved_share_pct	Share of places the rule reserves, as stated by the source (several categories summed, noted).	percent
equity_reserved_places	Number of places reserved, where published.	number
equity_reserved_places_base	Total places the reserved count is a share of.	number
equity_filled_entrants	Entrants admitted under the equity rule, where published.	number
equity_filled_entrants_base	Total entrants the filled count is a share of.	number
equity_filled_share_pct	Filled share as stated by the source.	percent
share_includes_disability	yes if the stated reserved share is one indivisible figure that also covers disability.	yes | no | blank
equity_disability_rule	A group rule (reserved places, lower threshold, separate track) for applicants with disabilities exists; not counted as formal equity.	yes | no | unknown
equity_confidence	Confidence in the equity coding.	high | medium | low
adm_notes	Admissions notes: sources, conflicts, coordinator and checker corrections.	text

10. Versions and updates

The dataset will be extended with more institutions per country and more countries. Each release gets a new version number, is generated by code from the recorded evidence, and is never edited by hand; the changes are listed here.

• v2.2 (October 2026): admissions variables for the 308 institutions of the QS 2027 academic-reputation sample: applicants, admitted, entrants, places, admission regime, formal equity admission rules (with scope, basis, instrument, mandate, reserved and filled counts) and disability rules, each collected and independently checked (admissions_sources.csv, admissions_verification.csv); a table of which types of institutions have formal equity rules (equity_by_type.csv, equity_by_country.csv, section 6). Prices and classifications unchanged from v2.1.
• v2.1 (October 2026): latest prices (academic year 2025/26 or calendar 2025) for every institution next to the 2023/24 prices, each divided by GDP per capita of its own year (latest_* columns, price_sources.csv, price_verification.csv, a price-year switch in the visualisation); eight countries added (Slovakia, United Arab Emirates, Oman, Philippines, Kazakhstan, Uzbekistan, Azerbaijan, Kyrgyzstan); 376 institutions in 63 countries and territories. Corrections of 2023/24 figures: Universidad de Lima (an enrolment fee of S/ 250 per semester had been missed), Universidad de Granada (a file-opening fee of €59.10, not €26.30), NTU “KhPI” (the range maximum was an English-taught programme), the University of Haifa (a range across programmes), PSL (range minimum); no band changed.
• v2.0 (September 2026): samples from the QS World University Rankings 2027 (overall rank and academic reputation) added as headline samples, QS 2024 kept as a robustness check; 328 institutions in 55 countries and territories (54 new); QS classification variables added; prices still 2023/24. Correction: a build-order error had let an earlier recode overwrite a later ruling, so the Free University of Berlin and RWTH Aachen showed institution-funded fee aid; both are now no (the fund is run by the student body and tested on the student's own income).
• v1.0 (September 2026): first public release. 274 institutions in 55 countries and territories (QS 2024 top 5); protocol v1.4.

Generated by build_release.py v0.5.