Evaluating the Impact of Tutoring Subsidies for 16-19 Year Olds | A-Level阶段辅导成效评估

2026-10-15 13:07 · 本文部分内容由AI辅助创作

📚 Evaluating the Impact of Tutoring Subsidies for 16-19 Year Olds | A-Level阶段辅导成效评估

Every year, schools, colleges, charities and families spend substantial sums on tutoring for students aged 16 to 19, yet the evidence base for that spending is far thinner than the equivalent evidence for younger pupils. This article sets out how to evaluate whether tutoring funding actually improves A-Level outcomes, which measures to use, which statistical tools to apply, and which mistakes routinely inflate or conceal impact.

每年，学校、学院、慈善机构和家庭都会为16至19岁学生的补习投入大量资金，但支持这些投入的证据基础却远比低龄阶段的同类证据薄弱。本文将系统说明如何评估补习经费是否真正提升了A-Level成绩、该采用哪些成效指标、该使用哪些统计工具，以及哪些常见错误会系统性地夸大或掩盖真实成效。

1. Why the 16-19 Phase Needs Its Own Evidence Base | 为何16-19岁阶段需要独立的证据基础

The 16-19 phase is structurally different from Key Stage 3 or Key Stage 4. Students study three or four linear two-year courses, cohorts are small, and the outcome that matters most is not a scaled score but a set of letter grades that determine university offers. Evidence generated for GCSE cannot simply be transferred, because the outcome measures, the funding rules and the stakes are all different.

16-19岁阶段在结构上不同于KS3或KS4。学生通常修读三到四门两年制线性课程，年级规模较小，而最关键的产出并非标准化分数，而是一组决定大学录取的字母等级。为GCSE生成的证据无法直接迁移，因为成效指标、资助规则和风险水平都不同。

There is a second reason. Much of the strongest tutoring evidence comes from primary and lower-secondary settings, where effect sizes of roughly 0.2 to 0.5 standard deviations are commonly reported for high-dosage small-group or one-to-one tuition. Whether those effects survive at A-Level, where content is harder, students are more self-directed, and grading is more compressed at the top, is an empirical question that has not been settled.

还有第二个原因。最有力的补习证据大多来自小学和初中阶段，高剂量的小组或一对一辅导通常报告约0.2至0.5个标准差的效应量。这些效应能否在A-Level阶段延续——该阶段内容更难、学生自主性更强、高分段评分更为压缩——仍是一个尚无定论的实证问题。

Evaluators therefore need to treat 16-19 tutoring as a distinct field of study, with its own baseline measures, its own outcome logic and its own cost benchmarks.

因此，评估者应将16-19岁补习视为一个独立的研究领域，拥有自己的基线测量、成效逻辑和成本基准。

2. The Funding Landscape: Who Actually Pays | 资金来源：究竟是谁在付费

Before evaluating impact, it is essential to identify the funding stream, because different streams carry different accountability requirements and different student populations. Confusing a means-tested bursary with a universal tutoring subsidy will destroy the internal validity of any comparison.

在评估成效之前，必须先厘清资金流，因为不同资金流对应不同的问责要求和不同的学生群体。把经过经济状况审查的助学金与普惠性补习补贴混为一谈，会直接破坏任何比较的内部效度。

Common funding routes in the UK and comparable systems include the following.

在英国及类似体系中，常见的资金渠道包括以下几类。

16-19 Bursary Fund and college discretionary funds, targeted at disadvantaged students but not ring-fenced for tutoring.
16-19助学金及学院自由裁量基金，面向弱势学生，但并未限定用于补习。
Government tutoring programmes, such as England's National Tutoring Programme, which operated from 2020 to 2023 and focused mainly on pupils aged 5 to 16.
政府补习计划，如英格兰2020至2023年实施的"国家补习计划"，主要面向5至16岁学生。
Charitable and widening-participation programmes funded by universities or the Office for Students.
由大学或学生事务办公室资助的公益及扩大参与类项目。
Direct parental spending, typically the largest single source at 16-19 in cash terms.
家长直接支出，按现金规模计算通常是16-19岁阶段最大的单一来源。

Each route implies a different counterfactual. A means-tested programme must be compared against similarly disadvantaged non-participants; a parental-purchase programme must be compared against families who could afford tutoring but chose not to buy it, which is a very different group.

每条渠道都对应不同的反事实。经过经济状况审查的项目，必须与同等弱势但未参与的学生比较；而家长自费项目，则需与"有能力购买但选择不购买"的家庭比较，这是完全不同的一组人群。

3. Defining the Counterfactual | 界定反事实：不补习会发生什么

The central question in any evaluation is not "did grades improve?" but "did grades improve more than they would have done without the funding?" Progress in A-Level is the norm, not the exception, so a simple before-and-after comparison will almost always show an improvement and almost always overstate the tutoring effect.

任何评估的核心问题都不是"成绩是否提高了"，而是"成绩是否比没有这笔资助时提高得更多"。在A-Level阶段，进步是常态而非例外，因此单纯的前后对比几乎总会显示进步，也几乎总会高估补习效应。

Four designs are practical at this scale.

在这一规模下，有四种设计是可行的。

Randomised controlled trial, where eligible students are randomly allocated to funded tutoring or a waiting-list control.
随机对照试验：将符合条件的学生随机分配到获资助补习组或候补对照组。
Regression discontinuity, exploiting a sharp eligibility threshold such as a household income cut-off.
断点回归：利用明确的资格门槛，如家庭收入分界线。
Difference-in-differences, comparing change in the funded group with change in a comparable unfunded group.
双重差分：比较受资助组与可比未受资助组的变化量。
Propensity score matching, pairing funded students with statistically similar unfunded students on prior attainment, subject, school and background.
倾向得分匹配：依据先前成绩、科目、学校与背景，将受资助学生与统计特征相近的未受资助学生配对。

The difference-in-differences estimator is written as follows.

双重差分估计量的表达式如下。

ΔΔ = (Y₁,post − Y₁,pre) − (Y₂,post − Y₂,pre)

Where Y₁ denotes the funded group and Y₂ the comparison group. The estimate is unbiased only if the two groups would have followed parallel trends in the absence of the programme, an assumption that should always be tested using pre-programme data.

其中Y₁代表受资助组，Y₂代表对照组。只有当两组在无项目情况下会呈现平行趋势时，该估计才是无偏的；这一假设应始终用项目前数据进行检验。

4. Choosing Outcome Measures That Actually Move | 选择真正会变化的成效指标

A-Level outcomes are coarse. A single grade band may represent a wide range of raw marks, and a student may improve substantially in understanding without crossing a boundary. Evaluators should therefore collect both a fine-grained attainment measure and a set of secondary outcomes.

A-Level的成绩是粗粒度的。一个等级区间可能对应很宽的原始分范围，学生的理解可能显著提升却未跨越分数线。因此评估者应同时收集细粒度的成绩指标和一组次要成效指标。

Domain 领域	Measure 指标	Why it matters 重要性	Risk 风险
Attainment 成绩	Raw mark or UMS on unseen past papers 未见过的真题原始分或UMS	Sensitive to small gains 对小幅度提升敏感	Test familiarity 测试熟悉度干扰
Grade 等级	Final A-Level letter grade 最终A-Level字母等级	Policy-relevant 与政策直接相关	Boundary noise 分数线噪声
Progress 进步	Value-added score 增值分数	Controls for intake 控制生源差异	Model dependence 依赖模型设定
Destination 去向	Offer met, university enrolled 是否达con、是否入学	Ultimate purpose 最终目的	Long lag, small n 滞后长、样本小
Non-cognitive 非认知	Self-efficacy, exam anxiety scales 自我效能、考试焦虑量表	Explains mechanisms 解释作用机制	Self-report bias 自陈偏差
Cost 成本	Cost per student hour 每生每小时成本	Enables comparison 便于横向比较	Hidden overheads 隐性管理成本

Predicted grades should never be used as the primary outcome. They are forecasts produced by teachers who know which students are receiving tutoring, and they are therefore contaminated by expectation effects.

预估成绩绝不应作为主要成效指标。它们是由知道哪些学生正在接受补习的教师作出的预测，因而会受到期望效应的污染。

5. Effect Size, Value-Added and Standard Error | 效应量、增值与标准误

A-Level grades are on an ordinal scale, so raw differences between groups are hard to interpret. Standardised effect sizes allow comparison across subjects, cohorts and programmes.

A-Level等级属于有序尺度，因此组间原始差异难以解释。标准化效应量可使不同科目、年级和项目之间具备可比性。

d = (M₁ − M₂) ÷ SD_pooled

nchmark in education is that d ≈ 0.40 represents one additional year of typical progress, though this benchmark is contested.

其中M₁为受资助组的成效均值，M₂为对照组的均值，SD_pooled为合并标准差。教育研究中常用的基准是d ≈ 0.40相当于额外一年的常规进步，但这一基准存在争议。

With small sixth-form cohorts, precision matters more than the point estimate. The standard error of d is approximated by the following expression.

在小型六年级学院中，精度比点估计更重要。d的标准误可由下式近似给出。

SE_d = √[ (n₁ + n₂) ÷ (n₁ × n₂) + d² ÷ (2 × (n₁ + n₂)) ]

The 95% , the evaluation has not demonstrated an effect, regardless of how encouraging the raw grade figures look.

95%置信区间则为d ± 1.96 × SE_d。若该区间包含零，则无论原始成绩数字多么令人鼓舞，该评估都未能证明存在效应。

Value-added measures offer a complementary view. For each student, compare actual attainment with the attainment predicted from prior GCSE performance, ALIS-type baseline tests and subject difficulty.

增值指标提供了互补视角。对每位学生，将其实际成绩与依据先前GCSE表现、ALIS类基线测试和科目难度预测的成绩进行比较。

VA = Attainment_actual − Attainment_predicted

Finally, plan the sample before collecting data. Under standard assumptions the minimum detectable effect size is approximately as follows.

最后，应在收集数据前规划样本量。在标准假设下，最小可检测效应量大致为下式。

MDES ≈ 2.8 ÷ √n

Per arm, for 80% power at the 5% significance level. Adjusting for a strong baseline covariate with correlation r reduces this further by a factor of √1 − r². With only 40 students per arm, MDES is around 0.44, which is larger than most plausible tutoring effects; this single calculation explains why so many small sixth-form evaluations end inconclusively.

每臂样本量，在5%显著性水平、80%检验力下适用。若纳入相关系数为r的强基线协变量，可再乘以√1 − r²进一步降低。若每臂仅40名学生，MDES约为0.44，已大于多数可能的补习效应；这一简单计算就解释了为何许多小型六年级评估最终得不出结论。

牛津AQA国际考试教程 AS&A级自然地理原版进口图书教材牛津大学出版社

¥378.0

📱 去京东购买

京东精选

6. Dosage, Attendance and Implementation Fidelity | 剂量、出勤与实施保真度

Tutoring is not a binary treatment. A student who attends three sessions and a student who attends thirty are both nominally "funded", and pooling them will dilute any real effect. Evaluators should record hours actually received, not hours offered.

补习并非二元处理。参加三次课的学生和参加三十次课的学生在名义上都属于"受资助"，把二者合并会稀释真实效应。评估者应记录实际接受的小时数，而非计划提供的小时数。

A simple dose-response model can be estimated as follows.

可估计如下简单的剂量—反应模型。

ΔGrade = β₀ + β₁ × Hours + β₂ × Prior + β₃ × Attendance + ε

A positive and statistically significant β₁ provides supporting evidence for a causal dose effect, though it remains vulnerable to the objection that more motivated students attend more sessions.

若β₁为正且统计显著，则支持存在因果性剂量效应，但仍需警惕"更有动力的学生出勤更多"这一反驳。

Fidelity should be monitored against a written protocol. The following checks are the minimum viable set.

实施保真度应依据书面方案进行监测，以下是最低限度的检查清单。

Session logs signed by tutor and student, with start and end times.
由导师和学生签署的课时记录，含开始与结束时间。
A minimum dosage threshold, typically 12 to 15 hours per student per term, below which data are analysed separately.
设定最低剂量门槛，通常为每生每学期12至15小时，低于此值的数据应单独分析。
Independent observation of a sample of sessions against a fidelity rubric.
依据保真度评分表对部分课时进行独立观察。
Curriculum alignment checks to confirm that sessions target the examined specification.
课程一致性检查，确认补习内容对准考试大纲。
Recording of tutor qualification and subject specialism.
记录导师资历与学科专长。

Intention-to-treat analysis should always be reported alongside any per-protocol analysis. The former answers the policy question "what happens if we fund this?", while the latter answers the delivery question "what happens if students actually attend?".

意向性分析应始终与符合方案分析一并报告。前者回答政策问题"如果我们出资会怎样"，后者回答实施问题"如果学生真的来上课会怎样"。

7. Subject-Specific Patterns | 分学科规律

Effects are rarely uniform across subjects, and an evaluation that pools all A-Level subjects into one average will hide the most useful findings.

各科效应很少一致，若把所有A-Level科目合并为一个平均值，会掩盖最有价值的发现。

In mathematics and the quantitative sciences, tutoring tends to show clearer short-term gains. Procedural fluency, multi-step problem solving and the correction of entrenched algebraic errors are all amenable to targeted instruction, and improvement is relatively easy to detect on unseen questions.

在数学与定量理科中，补习往往呈现更明显的短期提升。程序性熟练度、多步骤解题以及对根深蒂固代数错误的纠正，都适合通过针对性教学改善，且提升在未见过的题目上较易被检测。

In essay-based subjects such as History, English Literature and Politics, progress is slower and harder to measure. Gains often appear in argument structure, use of evidence and evaluative judgement, which mark schemes reward inconsistently between examiners.

在历史、英语文学、政治等论文类科目中，进步较慢且更难测量。提升常体现在论证结构、证据运用与评价性判断上，而评分标准在不同阅卷人之间的给分一致性较低。

In modern foreign languages and in subjects with a non-examined assessment component, tutoring may shift coursework quality but not examination performance, or vice versa. Evaluators should separate the components.

在现代外语以及含非考试评估部分的科目中，补习可能改变课程作业质量却未改变考试成绩，或反之。评估者应把各部分分开处理。

A useful rule of thumb is that the same 20 hours of tutoring will typically produce a larger measured effect in a subject where marks are densely distributed and questions are objectively marked, and a smaller measured effect where marking is holistic and cohorts are small.

一条有用的经验法则是：同样20小时补习，在分数分布密集且题目客观评分的科目中通常产生更大的可测效应；而在整体评分、考生规模较小的科目中，可测效应更小。

8. Grade Boundaries, University Offers and the Threshold Effect | 分数线、大学录取与阈值效应

At 16-19, the outcome of interest is usually not a continuous score but a threshold: did the student achieve the grades required by their offer? This makes the relationship between tutoring and outcomes distinctly non-linear.

在16-19岁阶段，真正关注的成效通常不是连续分数，而是阈值：学生是否达到了录取条件所要求的成绩？这使补习与成效之间的关系呈现明显的非线性。

A student sitting two marks below a boundary gains an entire grade from a very small improvement. A student sitting fifteen marks above the same boundary gains nothing measurable from the same tutoring, even if their understanding has improved. This is the threshold effect, and it means average marks can understate or overstate grade-level impact depending on where students sit relative to boundaries.

一名距分数线差两分的学生，只需极小提升即可获得整整一个等级；而一名高出同一分数线十五分的学生，即便理解力提升，也无法在成绩上体现。这就是阈值效应，意味着平均分的变化会因学生与分数线的相对位置而高估或低估等级层面的影响。

UCAS tariff points make the stakes concrete. At A-Level the standard tariff is as follows.

UCAS积分可让这种利害关系变得具体。A-Level的标准积分如下。

Grade 等级

Tariff points 积分

广告 / Sponsored