# ESAT score statistics: the numbers, and their limits

> UAT-UK's 2025/26 technical report prints every ESAT module's mean, SD and percentiles for 14,193 candidates. Here they are, and what they cannot tell you.

[Formatted version of this page](https://www.frontiervue.com/exams/esat/guides/score-statistics)

Most ESAT pages give you the range, 1.0 to 9.0, and leave it there. UAT-UK's annual technical report goes considerably further, printing a mean, a standard deviation and four percentiles for every module in every sitting, and almost nobody turns that into something a candidate can look up. This page does. It lays out the 2025/26 cycle in full: 14,193 candidates, October 2025 and January 2026 side by side. The limits get the same treatment as the numbers. Nothing here is a university's requirement, nothing here adds up into a total, and nothing here can be measured against last year.

## Key takeaways
- The 2026 cycle had 14,193 ESAT candidates in total, everyone sitting Mathematics 1: 10,116 (71%) in October 2025 and 4,077 (29%) in January 2026, up from 11,919 in 2024/25.
- The scale is constructed, not observed: UAT-UK fixes the October median candidate at 4.5 and the 90th percentile at 7.0, which is exactly why every module's October median lands on 4.5.
- Module means across both events, in the report's own order: Mathematics 1 at 4.58, Biology 4.44, Chemistry 4.61, Physics 4.25, Mathematics 2 4.40. They are not a league table. The report states modules are not comparable and there is no total.
- Every January mean sits below its October counterpart, 3.55 against 4.58 in Physics. That is not a difficulty gap: one set of scaling constants covered both events, and the report states scores are comparable across them.
- Scores carry an error bar: the scaled-score standard error of measurement was 0.54 to 0.73 for Mathematics 1 in October and 1.03 to 1.05 for Biology. A gap smaller than one SEM is not a real gap.

### Quick answer: the official reference points for an ESAT score

| Item | Value |
| --- | --- |
| Cohort size | 14,193 candidates: 10,116 (71%) in October 2025 and 4,077 (29%) in January 2026 |
| Scale anchors | The October median is fixed at 4.5 and the 90th percentile at 7.0, capped at 1.0 and 9.0, reported to one decimal place |
| Module means, both events | Mathematics 1 4.58, Biology 4.44, Chemistry 4.61, Physics 4.25, Mathematics 2 4.40 |
| What the data cannot do | Module scores cannot be aggregated or compared with one another, and this cycle cannot be compared with 2024/25 |

> A percentile is not a threshold. UAT-UK publishes no pass mark and no score requirement, so every figure here describes where candidates landed, never what a university asks for.

## Contents
1. The official distribution, module by module
2. October and January: one ruler, two separate cohorts
3. Reading your own score: percentiles, and the error bar around them
4. Group differences: what the tables show, and how far they stretch
5. Speededness and reliability: the two numbers behind the score

## 1. The official distribution, module by module

Start with what you came for. What follows is the all-events block of Table 5 in the official technical report, covering every candidate in the 2025/26 cycle. One rule governs it: read down a column, never across. Inside a module the numbers compare; between modules the report forbids it.

|  | Mathematics 1 | Biology | Chemistry | Physics | Mathematics 2 |
| --- | --- | --- | --- | --- | --- |
| Candidates | 14,193 | 1,547 | 3,146 | 11,290 | 12,113 |
| Mean | 4.58 | 4.44 | 4.61 | 4.25 | 4.40 |
| Standard deviation | 1.47 | 1.87 | 1.74 | 1.71 | 1.63 |
| 25th percentile | 3.6 | 3.2 | 3.5 | 3.0 | 3.3 |
| Median | 4.4 | 4.5 | 4.5 | 4.1 | 4.2 |
| 75th percentile | 5.3 | 5.7 | 5.7 | 5.3 | 5.4 |
| 90th percentile | 6.7 | 7.0 | 7.0 | 6.5 | 6.5 |

> Every module bottoms out at 1.0 and tops out at 9.0. Source: Table 5 of the UAT-UK 2025/26 ESAT technical report, the all-events rows covering both the October and January sittings.

- **14,193** ESAT candidates this cycle, every one of them sitting Mathematics 1
- **4.5** Every module's October median, fixed there by the scaling procedure
- **7.0** Every module's October 90th percentile, the other fixed anchor
- **27 items / 40 min** Per module, identical across all five

**How 14,193 candidates combined their modules**: Mathematics 1 is compulsory; most candidates add two or three more (Table 4)

- Maths 1 + Maths 2 + Physics: 10,530 candidates, 74%
- Maths 1 + Biology + Chemistry: 1,334 candidates, 9%
- Maths 1 + Maths 2 + Chemistry: 1,155 candidates, 8%
- The remaining combinations: Each printed at 5% or below
  - Maths 1 + Chemistry + Physics: 642, 5%
  - Maths 1 + Maths 2: 303, 2%
  - Maths 1 + Biology + Maths 2: 110, 1%
  - Maths 1 + Biology + Physics: 103, 1%

> Two sentences from the report worth memorising. 'The scaled scores are not comparable across modules and there is no aggregate or total score.' And: 'This means that the scaled scores from this cycle cannot be directly compared to those from the 2024/25 cycle.' The first kills any talk of a combined total; the second kills year-on-year comparison.

## 2. October and January: one ruler, two separate cohorts

Which sitting is easier comes up second-most often here. At first glance January looks harder, since every module's January mean falls below its October counterpart. The scaling section rules that reading out. One set of constants was applied to both events, deliberately, so that a score carries the same meaning whichever event produced it. The rows below therefore hold two entirely separate groups of candidates, 10,116 and 4,077 of them, measured against a single unchanging ruler.

|  | Candidates | Mean | Median | 90th percentile |
| --- | --- | --- | --- | --- |
| Mathematics 1, October 2025 | 10,116 | 4.73 | 4.5 | 7.0 |
| Mathematics 1, January 2026 | 4,077 | 4.19 | 4.1 | 5.4 |
| Chemistry, October 2025 | 2,814 | 4.72 | 4.5 | 7.0 |
| Chemistry, January 2026 | 332 | 3.63 | 3.5 | 5.7 |
| Physics, October 2025 | 7,717 | 4.58 | 4.5 | 7.0 |
| Physics, January 2026 | 3,573 | 3.55 | 3.4 | 5.5 |
| Mathematics 2, October 2025 | 8,082 | 4.73 | 4.5 | 7.0 |
| Mathematics 2, January 2026 | 4,031 | 3.75 | 3.7 | 5.4 |

> Biology is left out here: only 16 candidates sat it in January and the report prints NA in all nine statistics columns. Its October figures were 1,531 candidates and a mean of 4.46. Chemistry in January is also the one row in Table 5 whose maximum is not 9.0 but 8.9.

> Two lines, quoted as printed. On method: 'The same scaling constants were used for both the October 2025 and January 2026 events to ensure the scaling was consistent and scaled scores were comparable across events.' On what that means for one candidate: 'Therefore, a candidate who scored 6.5 in Chemistry, for example, in either January or October has a higher ability than a candidate who scored 4.2 in Chemistry in either event.' A 6.5 in Chemistry is a 6.5 in Chemistry, whichever event issued it.

### Why the October median lands exactly on 4.5

Because it is defined that way. Candidate ability is calibrated with a Rasch model, the October median ability is then pinned to 4.5 and the 90th-percentile ability to 7.0, a regression line between those two points yields the scaling constants, and the result is capped at 1.0 and 9.0. A median of 4.5 says nothing about how strong this year's cohort was. It says the scaling ran.

### Are January scores discounted?

No. Both events share one set of scaling constants and the report says outright that scores are comparable across them. Sample size is the thing that actually deserves caution: 332 candidates sat Chemistry in January, and 16 sat Biology, whose nine statistics columns are all printed as NA. The smaller the row, the shakier its percentiles as a reference point.

### Why the January percentiles bunch together

Look at the spread. Mathematics 1 had a standard deviation of 1.59 in October and just 1.03 in January. The January scores cluster far more tightly, so the run from the 25th percentile at 3.6 up to the 90th at 5.4 is much shorter than October's 3.6 to 7.0. In January the same handful of tenths moves you further through the rank order.

## 3. Reading your own score: percentiles, and the error bar around them

The reflex on opening a result is to ask what it is worth. There is an answer, but only in the currency of position, never in the currency of sufficiency. These four steps are the order in which the two tables above actually work, and the fourth is the one everybody skips. If you would rather build a score before benchmarking one, the [question bank](/papers/esat) is the place to start.

1. **Stay inside one module** (Step 1)
Find your module's column. A 5.0 in Chemistry and a 5.0 in Physics are not the same achievement, and the report provides no total.

2. **Match your sitting** (Step 2)
The two events run different percentile ladders. A 5.4 in Mathematics 1 falls between the October median of 4.5 and the October 75th percentile of 5.6, while in January that same 5.4 is exactly the 90th percentile.

3. **Read a band, not a point** (Step 3)
Only the 25th, 50th, 75th and 90th are printed, with nothing to interpolate in between. 'Between the 50th and 75th percentile' is an honest sentence. A precise-sounding rank is not.

4. **Add the error bar back** (Step 4)
A scaled score is a measurement and carries error. Apply the standard errors below and a lot of apparent gaps stop existing.

- **Scaled-score SEM:** 0.54 to 0.73 (Mathematics 1, across the October forms)
- **Scaled-score SEM:** 1.03 to 1.05 (Biology, October forms, the largest of any module)
- **About 68% probability:** True score within ±1 SEM (The classical test theory framing the report cites)
- **About 95% probability:** True score within ±2 SEM (Biology carries the largest SEM, so its band is the widest)

> The red line, stated plainly: a percentile is not a threshold. UAT-UK publishes no pass mark, and no fixed score requirement is published for you to clear, so any chain of reasoning that turns '90th percentile' into 'safe' is something a reader added. Position is what this page offers. A promise is not. Check each university's own pages for what actually counts.

## 4. Group differences: what the tables show, and how far they stretch

A large part of the technical report breaks scores down by gender, first language, area of residence and school type. Worth reading, with one rule set first: these are summary statistics. They record what a group scored, not what a group is capable of. The means below come from those tables and are still read downwards, comparing groups inside a single module.

|  | Mathematics 1 | Chemistry | Physics | Mathematics 2 |
| --- | --- | --- | --- | --- |
| Candidates identifying as men | 4.66 | 4.88 | 4.36 | 4.47 |
| Candidates identifying as women | 4.37 | 4.31 | 3.93 | 4.22 |
| English as a first language | 4.14 | 4.35 | 4.10 | 4.06 |
| Another first language | 5.09 | 4.93 | 4.44 | 4.78 |
| Resident in the UK | 4.07 | 4.28 | 4.00 | 3.98 |
| Resident in the EU | 3.75 | 3.57 | 3.56 | 3.68 |
| Resident outside the UK and EU | 5.21 | 5.08 | 4.62 | 4.92 |

> Biology is omitted to keep the table readable; the report draws a separate conclusion about its residence split, quoted in the fold-out below. All figures are the mean columns of Tables 7, 8 and 11 in the technical report.

### Gender: the choice gap is wider than the score gap

In the full cohort, meaning Mathematics 1, 69% identified as men and 30% as women. Who picks what varies far more than how they score: Biology has the highest share of women at 60%, Physics the lowest at 24%, and the report notes that this matches last year. On the scores it gives a range: 'The difference in mean scaled score ranges from 0.57 for Chemistry to 0.25 for Maths 2.'

### First language: the counter-intuitive one

Between 53% and 56% of each module's candidates have English as a first language, and the report flags that the Mathematics 2 share stood at 65% last year. The direction of the gap tends to surprise people: candidates whose first language is not English score higher. In the report's words: 'The gap is smallest for Biology, where the mean score for non-English first language candidates was 0.02 higher than those with English as a first language, and largest for Maths 1 (0.96).'

### Residence: Biology is the exception

Between 48% (Mathematics 2) and 51% (Chemistry) of candidates live in the UK, a thin slice in the EU, and most of the rest outside both. The report's wording: 'For Biology, UK (mean of 4.43) and Other (mean of 4.54) candidates are broadly comparable, but for the other modules, Other candidates outperform UK candidates.' Anyone testing from outside the UK and EU falls into that Other group, whose mean is the highest of the three in every module of Table 8. That is a fact about the distribution, not a requirement from anyone.

### School type and socio-economic background

Among UK candidates the commonest setting is a further education or sixth form college, from 37% (Biology) to 41% (Physics and Mathematics 2). Grammar school candidates make up 17% to 21%, which the report sets against the roughly 5% of state-funded secondary pupils who attend one nationally. Between 7% and 10% receive free school meals, against almost 25% of pupils nationally. These sit in the report precisely because the candidate pool is not a cross-section of school leavers.

> A gap between group means is not the same thing as items being unfair to a group, and the report keeps the two apart. Its differential item functioning analysis flags very few items at category C this cycle: one in Mathematics 1 (first language), one in Biology (gender), two in Chemistry, two in Physics (one favouring men, one favouring women) and none at all in Mathematics 2. The analysis only runs where there are at least 50 responses per group and 200 in total.

## 5. Speededness and reliability: the two numbers behind the score

This is the section with training value in it. The report counts how many candidates never reached the final items, and how reliable each module proved to be. Running out of time in mocks is a documented property of the paper rather than a personal defect. For pacing, see the [test day guide](/exams/esat/guides/test-day).

- **40%** of Mathematics 2 candidates left an item unreached or spent under 5 seconds on one, still the most speeded module, down from over 50% last cycle
- **13%** of Mathematics 1 candidates had at least one item they never reached, down from 15%
- **7%** the same figure for Physics, down from 10%, which the report calls borderline speeded
- **Over 90%** of candidates used between 35 and 40 minutes, except in Biology at 86%

### Reliability: why 0.70 clears the bar on a test this short

The usual rule of thumb asks for a Cronbach's alpha of at least 0.80, but the report writes that 'A reliability of over 0.70 would still be considered satisfactory for the lengths of the ESAT modules', because each module runs to only 27 items. Raw score reliabilities this cycle: Mathematics 1 from 0.68 to 0.84, Chemistry's October forms 0.76 to 0.80, Physics 0.70 to 0.77, Mathematics 2 0.66 to 0.81 against 0.53 to 0.77 last cycle, and Biology 0.65 to 0.69 on a much smaller sample.

### Item quality: almost everything cleared the bar

The normal target is that 80% of new items clear the criteria. This cycle Mathematics 1, Biology and Chemistry each hit 100%, Physics 98% and Mathematics 2 99%. Mathematics 2 improved noticeably as well: items with a p value below 0.20 fell from 16% last cycle to 8%, which is to say fewer near-impossible questions.

### Timing: nearly everyone uses the full window

Each module is capped at 00:40:00, every module's mean time lands between 39 and 40 minutes, and the median is a flat 40 minutes for Mathematics 1 and Mathematics 2, fractionally under for the rest. Around 4% of candidates per module were excluded from the timing analysis for having extra time, and 614 candidates, about 5% of the cohort, had access arrangements on Mathematics 1.

> Turned into training decisions: Mathematics 2 and Mathematics 1 are where pacing practice pays. Biology and Chemistry are far less pressured, with only 2% of candidates leaving an item unreached, so accuracy is the better target there. Time your practice in 40-minute blocks, one per module, to match the real rhythm.

> Once you have your own scores, results-day details and where to check them are on [ESAT results day](/results/esat).

## Questions
**What counts as a good ESAT score?**

There is no published pass mark and no published score requirement, so good can only mean good relative to the distribution. Mathematics 1 in October ran 3.6 at the 25th percentile, 4.5 at the median, 5.6 at the 75th and 7.0 at the 90th. The same four numbers for January were 3.6, 4.1, 4.7 and 5.4. A score means something inside one module and one sitting, and nowhere else.
**Is the October or the January sitting easier?**

The January means are genuinely lower, 3.55 against 4.58 in Physics. That still says nothing about paper difficulty: one set of scaling constants covered both events and the report states scores are comparable across them, so the same score means the same ability either way. What does deserve caution is the January sample size, 332 candidates in Chemistry and 16 in Biology, whose statistics are printed as NA. On which event your universities accept, see the [registration and dates guide](/exams/esat/guides/registration-and-dates).
**Can I add my module scores into a total?**

No. The report is explicit: 'The scaled scores are not comparable across modules and there is no aggregate or total score.' The reason sits in the method. Each module is scaled on its own with its own constants, a multiplier of 1.0570 for Mathematics 1 against 2.3107 for Biology. A 5.0 therefore rests on a different performance in each module, and adding or averaging them means nothing.
**Is this year's 4.5 the same as last year's?**

The report does not vouch for it. Scaling is redone each admissions cycle and pinned to that cycle's October ability distribution, and the report states this cycle's scaled scores cannot be directly compared with 2024/25. The only comparison that holds is within this cycle, October against January. For context, the cohort grew from 11,919 Mathematics 1 candidates last cycle to 14,193 this one.
**Does a small difference in score matter?**

A scaled score is a measurement and carries error with it. The scaled-score standard error of measurement ran from 0.54 to 0.73 for Mathematics 1 in October, and as high as 1.03 to 1.05 for Biology. Under the classical test theory the report cites, a true score falls within one SEM of the observed score about 68% of the time and within two SEM about 95% of the time. Practically: a gap narrower than one SEM is not a real gap.

## Official sources
- [UAT-UK ESAT Technical Report 2025-26, Table 5: scaled score summary statistics by module and event](https://uat-wp.s3.eu-west-2.amazonaws.com/wp-content/uploads/2026/08/07161800/UAT-UK-ESAT-Technical-Report-2025-26.pdf)
- [Same report, Tables 2 and 3: the scaling anchors and each module's scaling constants](https://uat-wp.s3.eu-west-2.amazonaws.com/wp-content/uploads/2026/08/07161800/UAT-UK-ESAT-Technical-Report-2025-26.pdf)
- [Same report, Table 4: the distribution of module combinations](https://uat-wp.s3.eu-west-2.amazonaws.com/wp-content/uploads/2026/08/07161800/UAT-UK-ESAT-Technical-Report-2025-26.pdf)
- [Same report, Tables 7, 8, 11 and 12: statistics by gender, residence, first language and school type](https://uat-wp.s3.eu-west-2.amazonaws.com/wp-content/uploads/2026/08/07161800/UAT-UK-ESAT-Technical-Report-2025-26.pdf)
- [Same report, Tables 17 to 22: reliability, standard error of measurement, timing and unreached items](https://uat-wp.s3.eu-west-2.amazonaws.com/wp-content/uploads/2026/08/07161800/UAT-UK-ESAT-Technical-Report-2025-26.pdf)
- [UAT-UK Annual Reports index (public entry point for all four 2025-26 reports)](https://esat-tmua.ac.uk/about-uat-uk/annual-reports/)
