How the EF EPI 2025 Is Built: 2,200,000 Tests, 400 Test Takers Minimum
The EF EPI 2025 rests on 2,200,000 free online tests, needs 400 test takers for a country to appear, and averages three years of scores into one number.
នៅលើទំព័រនេះ
Country rankings invite comparison, so the rules behind them decide what the comparison means. This report reads the methodology appendix of the EF English Proficiency Index 2025 and sets out what the numbers actually rest on: 2,200,000 free, self-selected online tests taken in 2024, a minimum of 400 test takers for a country and 100 for a city, and a three-year rolling average that blends the current edition with the two before it. The index covers 123 countries and regions and correlates with other tests at r = 0.74 for TOEFL iBT and r = 0.61 for IELTS Academic. The short version: the index is a large, consistent and deliberately self-selected snapshot of people who chose to test their English online, not a sample of each country's population.
Key findings
- The 2025 index is based on more than 2,200,000 EF SET tests taken in 2024 and covers 123 countries and regions.
- A country needs at least 400 test takers to enter the index and a city needs 100, which sets the smallest group the report will publish.
- Each national score is a three-year rolling average, so the current year is 1 of the 3 years behind the published figure.
- The sample is 46% female and 40% male, with 14% not stating a gender, and its average age is 26; 85% of respondents who gave a birth year were under 35.
- The index correlates at r = 0.74 with TOEFL iBT scores from 2023 and r = 0.61 with IELTS Academic scores from 2024, which EF presents as evidence that the tests reveal similar national trends despite different designs.
Background
The EF English Proficiency Index is one of the few public datasets that puts a single English score on more than a hundred countries and regions, which is why its country comparisons travel so widely in education policy and in employer marketing [1]. It is also built from an open online test rather than a national exam, so its sampling rules are the part a careful reader needs first. EF itself states that the test-taking population is self-selected, does not claim that it is representative, and says the resulting bias could skew scores in either direction [1]. This report collects the method facts that the index publishes, without re-ranking anything.
Data and method
Everything here comes from the 2025 report's appendices: the methodology section "About the Index" (Appendix A), the score calculation subsection, and the section on sampling biases [1]. No country score is recalculated and no ranking is reproduced, so the figures below are the report's own statements about its inputs and rules rather than new analysis.
The relevant definitions are EF's. A national score is the average EF SET score of test takers residing in that country during the previous calendar year, averaged with the published scores for the two earlier editions to produce a three-year rolling average. Region and world scores are population-weighted by country, while scores for subgroups inside a country, such as an age band or a gender, are not population-weighted and are calibrated against the national score instead. Skill scores, published for the first time in 2025, are the average of that skill's tests inside the country and do not necessarily average up to the national score.
The sample is 2,200,000 test takers, all of whom chose to take a free online test. EF publishes the gender and age composition of that group and the share who tested on a mobile device, but not a per-country count, so a country that barely clears the floor and one with a far larger testing population carry the same weight in any comparison of their scores. Every figure in this report is aggregated: no individual test record is used or published, no person is identifiable in any number quoted here, and no group smaller than the index's own thresholds appears.
Results
The base is large and open. More than 2,200,000 EF SET tests went into the 2025 edition, spread over 123 countries and regions, and the test is free and online, which is what makes a sample of that size possible at all.
Figure 1. Composition of the EF EPI 2025 sample as published in Appendix A, shares of respondents, n = 2,200,000 test takers. Source: EF EPI 2025 [1].
The sample is young and slightly female. Forty-six per cent of respondents were female, 40% male and 14% did not provide gender information; the average age was 26, and among respondents who gave a birth year, 85% were under 35 and 99.5% under 60, while 10% left the birth year blank. Forty-six per cent of 2024 test takers completed the exam on a mobile device, which matters because a phone-based session is a different setting from a proctored test centre.
The publication rules are the second constraint. Only cities and regions with at least 100 test takers enter the index, and only countries with at least 400, so the smallest country-level figure rests on a few hundred people. National scores are three-year rolling averages: the current year's test results are averaged with the two previously published scores, which smooths year-to-year noise and also means a country's published score moves slowly by design.
Figure 2. Correlation coefficients between EF EPI 2025 scores and other national test results, published by EF, n = 123 countries and regions. Source: EF EPI 2025 [1].
The cross-test correlations are the report's own validity claim. EF reports r = 0.74 against TOEFL iBT scores from 2023 and r = 0.61 against IELTS Academic scores from 2024, and describes the pattern as evidence that tests with different designs and different test-taker profiles reveal similar national trends. A correlation of that size supports comparability at the country level; it does not make the three tests interchangeable for an individual candidate, and EF does not claim that they are.
| Method element | What the 2025 report states |
|---|---|
| Base | More than 2,200,000 EF SET tests taken in 2024 |
| Coverage | 123 countries and regions |
| Minimum group | 100 test takers for a city, 400 for a country |
| National score | Three-year rolling average |
| Region score | Population-weighted by country |
| Subgroup score | Not population-weighted |
| Sample selection | Self-selected, free, online |
Read the table as the rulebook behind every number in the index: it tells you the smallest group that can appear, how much of a score comes from the current year, and which aggregates are weighted by population.
Limitations
Four limits follow directly from the method. First, the sample is self-selected: people who seek out a free online English test are not a random draw from their country, and EF states that the resulting bias could push scores either up or down. Second, the format excludes anyone without reliable internet access, which EF expects to pull scores upward by leaving out poorer and less educated populations. Third, the published composition is global rather than per country, so a country's own mix of ages and genders is not visible, and the minimum of 400 test takers is a floor rather than evidence that a country's sample represents its population. Fourth, the three-year rolling average and the population weighting of regional scores mean that a country's published number is a blended figure, and adding speaking and writing to a test that previously covered only reading and listening makes comparisons across editions harder still.
What this means for learners
Read a country score as a description of people who chose to take a free online test in that country, not as a statement about how well everyone there speaks English. If you are using the index to pick a place to study or work, treat the band as context and look for the requirement that actually applies to you, which is usually a specific test score from a specific provider. For your own progress, a test you can repeat cheaply and often is more useful than a national ranking, because the number that matters is your own and how it moves. Hibay offers spoken practice with feedback between tests, though its feedback is not an EF SET score and does not predict one.
Frequently Asked Questions
How many people take the test behind the EF EPI? More than 2,200,000 EF SET tests went into the 2025 edition, all taken in 2024 by people who chose to sit a free online test (see Key findings).
How many test takers does a country need to appear in the EF EPI? At least 400 for a country and 100 for a city or region, per the methodology appendix of the 2025 report (see Results).
Is the EF EPI a representative sample of each country? No. EF describes the sample as self-selected and does not claim that it represents each country's population, and it notes that the free online format excludes people without internet access, which tends to pull scores upward (see Limitations).
How does the EF EPI compare with TOEFL and IELTS? EF reports a correlation of r = 0.74 with TOEFL iBT 2023 scores and r = 0.61 with IELTS Academic 2024 scores at the country level, which supports broad comparability of national trends rather than equivalence for an individual candidate (see Results).
How to cite
Hibay Product Team. (2026, November 2). How the EF EPI 2025 Is Built: 2,200,000 Tests, 400 Test Takers Minimum. Hibay. https://www.hibay.ai/en/news/ef-epi-2025-how-the-index-is-built
Figures are from the public sources listed below and may be cited with attribution.
References
- EF Education First. EF English Proficiency Index 2025: A Ranking of 123 Countries and Regions by English Skills. https://www.ef.com/assetscdn/WIBIwq6RdJvcD9bc8RMd/cefcom-epi-site/reports/2025/ef-epi-2025-english.pdf. Accessed 2026-11-02.