GCSE评分基准 NRT参考测试 2017基线 显著性
每年夏天 GCSE 放榜时，公众看到的是分数线与等级比例，看不到的是这些数字背后的证据链。国家参考测试（National Reference Test，NRT）就是这条链上最关键的一环：它用一场不记成绩的抽样测试，判断全国 16 岁学生在英语语言与数学上的整体表现有没有发生变化。2024 年的结果同时出现了英语 4 分档的显著下降与数学 7 分档的显著上升，Ofqual 最终决定两科都不调整评分标准。这篇文章把 2024 年的数据、判断逻辑与后续影响一步步拆开讲。
When GCSE results come out each summer, the public sees grade boundaries and grade proportions, not the evidence chain behind them. The National Reference Test (NRT) is the key link in that chain: a no-stakes sample test that shows whether national attainment of 16-year-olds in English language and maths has shifted. The 2024 results produced a statistically significant fall at grade 4 in English and a significant rise at grade 7 in maths, yet Ofqual decided not to adjust standards in either subject. This article walks through the 2024 data, the reasoning behind that decision, and what followed in later years.
NRT 是什么：330 所学校的抽样测试
每年二月到三月，英格兰一部分抽样学校的 Year 11 学生参加一场一小时的短测试，题目风格与 GCSE 英语语言和数学考题接近。这场测试由 Ofqual 委托国家教育研究基金会（NFER）开发、实施并分析，2017 年第一次正式施测，此后每年一次。每科有 8 套题册，每道题同时出现在 2 套题册里，这样所有题目可以合在一起分析，得到每科单一的表现尺度。英语测试分阅读与写作两部分，各 25 分；数学测试有 13 或 14 题，共 50 分。目标样本是每科至少 330 所学校，每所学校最多 30 名学生参加英语或数学。它是一场无利害测试，不产生学生个人成绩，也不公布学校层面的结果，更不用于学校问责。很多学生以为这是一场全国模考，其实它在制度上的位置完全不同。
Every February and March, a sample of Year 11 students in England sits a one-hour test built to look like GCSE English language and maths papers. Ofqual commissions NFER to design, run and analyse it; each subject uses eight booklets, every question appears in two of them, and the results are combined into a single scale per subject; the target is at least 330 schools per subject with up to 30 students each. The test carries no personal or school-level results and plays no part in accountability, which is precisely why its evidence is treated as neutral.
2017 基线：改革后 GCSE 的第一次评分
NRT 的所有比较都回到 2017 年，因为那是改革后的 GCSE 英语语言与数学第一次评分。当年的 GCSE 结果给出了三个分数界限上学生的百分比，这些百分比被映射到 2017 年 NRT 的成绩上，得到一条对应的水平线。此后每一年，只要估计有多少学生达到这条水平线，就能看出全国表现是上升、下降还是持平。Ofqual 官网在解释这些数字时特别提醒一点：新资格刚推出的头几年，考生表现通常会先下滑再回升，这种锯齿效应必须纳入判断。2019 年与 2020 年关于数学的决定就建立在这个考虑之上。疫情让 2021 年与 2022 年的测试严重受限，但比较基准依旧是 2017 年。所以读到任何一年的 NRT 结果，第一件事都是看它相对 2017 年变了多少。
Everything in the NRT is measured against 2017, the year reformed GCSEs in English language and maths were awarded for the first time. GCSE outcomes from that year set the percentages of students at three grade thresholds, and those percentages were mapped onto the 2017 test to create the reference levels used ever since. Ofqual is explicit that new qualifications tend to dip in the first year and recover afterwards, so the sawtooth effect is weighed up whenever a change against 2017 is examined.
三个分数界限：7 分、5 分与 4 分
NRT 只在三个分数界限上报结果：7 分、5 分与 4 分，报告形式是预计达到该分数及以上的学生百分比。5 分界限由考试局按算术方式设定，通常不会成为调整对象，因此每年的年度声明主要讨论 7 分与 4 分。抽样测试必然带有误差，官方目标是让每个界限上的估计在 95% 置信区间内的误差不超过正负 1.5 个百分点。所有公布的百分比都带置信区间，两条区间的重叠程度直接决定结论能不能站住。分数界限本身并不改变试卷难度，它的作用是抵消难度差异：卷子难，界限就低；卷子容易，界限就高。常见的错误是把界限当成一条在年年变动的及格标准，其实它首先是同一年内不同试卷之间的一道平衡装置。
Results are reported at three thresholds only: grades 7, 5 and 4, expressed as the expected percentage of students reaching each grade or above. Grade 5 is set arithmetically by the exam boards and is not normally adjusted, which is why the annual statement concentrates on grades 7 and 4; the aim is a 95% confidence interval no wider than plus or minus 1.5 percentage points. Boundaries exist to cancel out paper difficulty rather than to move the standard, so a harder paper comes with lower boundaries and an easier paper with higher ones.
2024 年测试窗口与样本构成
2024 年的测试在 2 月 16 日到 3 月 4 日之间完成。超过 13,000 名 Year 11 学生参加，其中英语的完成样本来自 339 所学校，数学来自 341 所，都高于每科 330 所的目标。抽样按学校过去在 GCSE 英语语言与数学上的成绩以及学校规模分层，并对学校类型做了核对。实际完成率是英语 81%、数学 82%，低于疫情前平均 85% 的水平。缺席的主要原因是生病或其他获准理由，占英语缺席的 58%、数学的 62%；这反映的是 2024 年 3 月中学出勤率整体低于疫情前，而不是学生专门躲开了这场测试。获得特殊安排的考生比例从 2023 年的 6.8% 降到 5.8%，其中使用文字处理器的有 584 人。
Testing in 2024 ran between 16 February and 4 March, with more than 13,000 Year 11 students taking part. The achieved samples covered 339 schools in English and 341 in maths, above the target of 330 per subject, and they were stratified by historical GCSE performance and school size; completion rates were 81% in English and 82% in maths against a pre-pandemic average of 85%. Absence was mostly illness or another authorised reason, matching the higher absence rates seen in state secondary schools that March rather than any reluctance to sit the test itself.
英语语言 4 分档：从 69.9% 到 65.1%
与 2017 年相比，英语语言在 4 分档出现了统计学上显著的下降，显著性达到 0.01 水平。2017 年的预计值是 69.9%，置信区间 68.4 到 71.4；2024 年是 65.1%，置信区间 63.4 到 66.8。7 分档从 16.8% 变为 16.4%，置信区间 15.3 到 17.6，与 2017 年没有显著差异。Ofqual 的额外分析显示，样本的先前成绩构成无法解释这个下降，2024 年的样本在英语这一科甚至可能略强。学生调查确实记录了动机与准备度的下降，但建模结果显示这不足以解释 4 分档的变化。读这类结论时值得记住一点，显著指的是统计上不太可能是偶然，不是指幅度大。换句话说，这个下降更可能反映这一届学生整体的真实表现，而不是抽样带来的噪声。
English language shows a statistically significant fall at grade 4 against 2017, significant at the 0.01 level. The expected figure moved from 69.9% in 2017, with a confidence interval of 68.4 to 71.4, to 65.1% in 2024 with 63.4 to 66.8; at grade 7 the movement from 16.8% to 16.4% is not significant. Ofqual's extra analysis found that the prior attainment profile of the sample cannot account for the fall, and modelling suggested that lower test motivation was not a major factor either.
数学 7 分档：从 19.9% 到 22.4%
数学在 7 分档出现了显著上升，但只达到 0.05 的显著性水平。2017 年的预计值是 19.9%，置信区间 18.6 到 21.2；2024 年是 22.4%，置信区间 21.0 到 23.7。数学 4 分档从 70.7% 变成 70.9%，置信区间 69.4 到 72.4，没有显著变化。这个上升与 2019 年的数学结果相似，而在 2019 年，Ofqual 判断上升更可能来自新资格头几年的锯齿效应，因此没有调整评分标准。当年那份声明的逻辑在 2024 年被重新拿出来对照。很多学生以为 2.5 个百分点的上升意味着分数线一定会下调，实际上显著性水平与决策规则把它挡了回来。
In maths the rise at grade 7 is significant, but only at the 0.05 level. The expected percentage went from 19.9% in 2017, with a confidence interval of 18.6 to 21.2, to 22.4% in 2024; at grade 4 the change from 70.7% to 70.9% is not significant at all. That pattern closely resembles 2019, when Ofqual read the rise as a likely sawtooth effect and chose not to adjust, and that earlier decision shaped what happened in 2024.
显著性水平：0.05 与 0.01 的差别
0.05 的显著性水平意味着这种差异有二十分之一的概率是偶然出现的，0.01 水平把这个概率降到百分之一。NFER 两个水平都会报告，而 Ofqual 在解释结果时沿用往年的做法，以 0.01 为准，理由是 NRT 与 GCSE 结果都属于高风险判断。三个界限与多个年份一起比较，会让碰巧显著的机会迅速累积，统计上因此使用了 Bonferroni 校正来控制整体误差。0.01 门槛更高，结论更难达成，这是谨慎原则的直接体现。常见的错误是看到某一年的数字变了就认定标准一定跟着动，其实门槛、方向与历年一致性三件事要一起满足。把这两个水平混在一起读，是每年结果公布后最容易出现的误读。
A 0.05 significance level means a one in twenty chance that the difference is coincidental; at 0.01 the chance drops to one in a hundred. NFER reports both, while Ofqual interprets the evidence at the 0.01 level because of the high stakes attached to the test and to GCSE results. Because three thresholds are compared across many years, a Bonferroni adjustment is applied so that the overall level of significance holds, which makes a finding harder to reach than a single comparison would suggest.
Ofqual 对英语的判断：证据不足不下调
4 分档的显著下降本可以被解读为，这个夏天在英语语言的 4 分档小幅下调评分标准是合适的。Ofqual 却给出了更保守的结论：要做出调整，必须先确信下降反映了这一届考生真实的能力变化。疫情之后连续几年的背景让这种确信难以成立，因此 2024 年夏天没有下调。官方在声明里列出三条原则：年份之间与科目之间保持一致，方向不影响判断；把学生调查等背景证据一并纳入；谨慎行事并公开说明理由。按这个框架，下降本身不构成调整的充分条件。很多学生会以为数据变差就该放宽给分，制度上的顺序恰好相反，先要排除样本与动机等因素。
The fall at grade 4 could have been read as a case for a small downward adjustment to the English language standard this summer. Ofqual took the cautious route instead: an adjustment would need confidence that the fall reflected a genuine change in what this cohort knows, understands and can do, and the years after the pandemic made that confidence impossible to justify. Read together, consistency across years and subjects, contextual evidence such as the student survey, and a stated preference for caution explain why no adjustment was made.
数学为什么不调整：与 2019 年保持一致
数学 7 分档的上升只在 0.05 水平显著，这一点在决策里很关键。2019 年 Ofqual 判断数学相对 2017 年的上升更可能来自锯齿效应，当年决定不调整。如果 2024 年作出上调，等于奖励一个此前已经明确不奖励的表现水平，这与决定要在年份之间保持一致的原则直接冲突。声明里还写明，变化的方向不会改变判断标准，无论上升还是下降都按同一套规则处理。所以数学 7 分档在 2024 年夏天同样没有调整，只留下趋势观察。2025 年的结果公布后，Ofqual 再次决定英语与数学都不调整，一致性被继续执行。
The maths rise at grade 7 was significant only at the 0.05 level, which mattered to the decision. Ofqual had already decided in 2019 that a similar rise was more likely to reflect the sawtooth effect than a real improvement, so acting in 2024 would have rewarded a level of performance it had previously declined to reward. Consistency is the operative principle here; the direction of a change does not alter the test applied to it, and 2025 produced another no-adjustment decision for both subjects.
学生调查与先前成绩：为什么不能只看分数
参加 NRT 的学生在测试后要填一份短问卷，内容涉及测试动机、GCSE 备考情况以及对科目的态度，用来解释结果变化的背景。与 2017 年相比，2024 年两科学生都更不看重 NRT、对自己的测试成绩更无所谓、准备也更少，数学考生还报告努力程度下降。抽样完成者在先前成绩上带有向上的偏差，不过这种偏差在八次测试里都相对稳定，建模显示它不足以完全解释当年的变化。在英语这一科，样本先前成绩的差异方向与成绩下降方向相反，说明样本并不弱。问卷还记录了学生对科目本身的感受，这部分内容帮助解释成绩之外的背景。任何调整决定都要把这些证据放在一起看，单看百分比会得出错误结论。
Students who sit the NRT also complete a short survey on motivation, GCSE preparation and attitudes to the subject, which gives context to any movement in results. Compared with 2017, students in 2024 placed less importance on the NRT, felt more indifferent about their own performance and prepared less; in maths they also reported putting in less effort. The achieved samples carry an upward prior attainment bias that has been stable across eight administrations, and in English that bias runs against the fall rather than explaining it.
分数界限怎么被影响：从证据到指令
NRT 在制度上只是额外证据：出现统计学显著差异时，Ofqual 可以要求考试局在设定 GCSE 英语语言与数学的分数界限时调整评分标准。考试局设界限仍然靠数据与定性证据结合，阅卷员的专业判断在其中占很重的分量，专家会拿当年考生的作答与往年比较。2023 年评分回到疫情前标准，2024 年延续正常评分。2024 年 8 月 22 日，英格兰发出超过 560 万份 GCSE 成绩，报考量比 2023 年多出 25 万以上。整体看，7 分及以上占 21.7%，2023 年是 21.6%；4 分及以上占 67.4%，2023 年是 67.8%。这些数字与两科都不因 NRT 调整的结论是彼此吻合的。
The NRT is an additional source of evidence: where a difference is statistically significant, Ofqual can require exam boards to adjust grade standards when setting boundaries in GCSE English language and maths. Boards still combine data with qualitative evidence, and examiner judgement on the quality of students' work carries real weight. Grading returned to pre-pandemic standards in 2023 and continued normally in 2024, when more than 5.6 million GCSE results were issued in England, entries rose by more than 250,000, grade 7 and above reached 21.7% and grade 4 and above 67.4%.
2024 年真正被调整的科目：法语、德语与计算机科学
2024 年夏天被要求更宽松给分的是法语、德语与计算机科学，这三项调整与 NRT 没有关系。法语在 7 分与 4 分做了小幅正向调整，德语在 9 分、7 分与 4 分做了调整，这是跨学科可比性工作的延续，目标是让法语与德语的评分标准向西班牙语靠拢。计算机科学的调整来自对多年评分标准的研究，结论是 2014 到 2019 年间标准可能略偏严格，所以要求考试局在 9 分、7 分与 4 分放宽。相比之下，英语语言与数学在 2024 年没有因 NRT 做任何调整。很多学生以为 NRT 一定会动自己那一科的分数线，看清是哪个机制在起作用很重要。
The subjects required to be marked more generously in summer 2024 were French, German and computer science, and none of that came from the NRT. French was adjusted at grades 7 and 4, German at grades 9, 7 and 4, as part of long-running work on inter-subject comparability aimed at aligning French and German with Spanish. Computer science was adjusted at grades 9, 7 and 4 after research suggested that standards had become slightly more stringent between 2014 and 2019, while English language and maths carried no NRT-driven change at all.
2025 年与 2026 年：NRT 第一次真正动用
2025 年的测试规模继续扩大，近 14,000 名学生、超过 350 所学校参加，Ofqual 决定英语语言与数学都不调整。2026 年 8 月 20 日公布的结果显示，数学成绩出现了轻微但持续的长期改善，英语语言没有出现同样的趋势。为了保证标准稳定，Ofqual 要求考试局在设定数学部分分数界限时把 NRT 结果考虑进去，这些界限因此低于原本应有的水平。带来的影响是数学高分段达到高分的学生略有增加，4 分档保持稳定。英语语言 4 分档出现了下降迹象，趋势还不够一致，2026 年不做调整，2027 年结果出来后会再评估一次。把 NRT 当成预测自己分数的模考是最常见的误解，它测的是全国水平，不产生个人成绩。
Almost 14,000 students from more than 350 schools took the 2025 test, and Ofqual again decided that neither subject needed an adjustment. The 2026 results, published on 20 August, showed a slight but sustained long-term improvement in maths and no equivalent trend in English language, so exam boards were told to take the NRT into account when setting some maths boundaries. Those boundaries ended up lower than they would otherwise have been, adding a small number of higher grades at the top end while grade 4 stayed stable; English language remains under review for 2027.
需要一对一辅导？Need One-on-One Tutoring?
A-Level / GCSE / IB / KS2-KS3 全科辅导
16621398022 同微信
Visit aleveler.com
或截图后用「扫一扫 → 相册」选择本图