📚 Statistical Definitions in Post-16 Destination Data for 16-19 Year Olds | 16 至 19 岁学生毕业去向数据的统计口径
Every autumn, newspapers, school league tables and government briefings publish figures describing what happened to 16-19 year olds after they left compulsory education. The headline numbers look precise - 6.4% not in education, employment or training, 71% progressing to higher education - yet two organisations can publish very different numbers for what appears to be the same group of young people. The difference is almost never arithmetic; it is definitional. This article examines the statistical definitions, or statistical basis, behind post-16 destination data, and shows how to read, calculate and criticise those figures in an A-Level context.
每年秋季，报纸、学校排名表与政府简报都会发布描述 16 至 19 岁学生离开义务教育阶段之后去向的数据。那些头条数字看起来非常精确——6.4% 处于未升学、未就业、未受训状态，71% 升入高等教育——然而两个机构针对看似同一批年轻人，却可能公布截然不同的数字。差异几乎从不来自算术，而来自口径。本文将系统梳理 16 岁后去向数据背后的统计口径，并说明如何在 A-Level 语境下阅读、计算与评判这些数据。
1. Why Statistical Definitions Decide the Headline | 为什么统计口径决定新闻标题
A destination measure is a ratio. Like every ratio it has a numerator (who counts as having reached a destination) and a denominator (who is being described). Change either, and the published percentage changes even though not a single young person has actually done anything different. This is why a statistician's first question about any destination figure should be: "what exactly is being counted, and who exactly is in the base?"
去向指标本质上是一个比率。与所有比率一样，它有分子（谁被算作达成了某个去向）与分母（描述的对象是谁）。只要改动其中任何一项，公布出来的百分比就会变化，即便实际上没有任何一个年轻人做出不同的事情。因此统计学家的第一个问题永远是："究竟在数什么，基数里究竟包含谁？"
ported rate = (count matching the definition) ÷ (population matching the definition) × 100
That single equation hides at least five separate decisions: the age or year-group rule, the reference date, the destination categories, the treatment of missing data, and the treatment of small numbers. Each of the following sections takes one of those decisions and shows its numerical consequences.
这一个公式背后至少隐藏着五个独立决定：年龄或年级规则、参考日期、去向分类、缺失数据的处理方式，以及小样本的处理方式。以下各节逐一讨论这些决定，并展示其数值后果。
2. Defining the Cohort: Who Is in the Base? | 界定统计总体：谁进入分母
The first and most consequential decision is the population base. Published measures of 16-19 destinations differ enormously in who they include, and the base is usually stated in a technical note that readers never see. Four questions separate most measures from one another.
最具决定性的第一个选择是总体基数。公开发布的 16 至 19 岁去向指标在纳入范围上差异极大，而基数通常写在一份读者根本看不到的技术说明里。以下四个问题把大多数指标区分开来。
Age rule versus year group. A cohort may be defined by academic year (all students who were in Year 11 last July) or by age on a fixed date (everyone aged 16, 17 or 18 on 31 August). Students who repeat a year, arrive from abroad or leave early fall into different cells under the two rules.
年龄规则与年级规则。队列可以按学年界定（去年 7 月就读 11 年级的全部学生），也可以按某一固定日期的年龄界定（8 月 31 日已满 16、17 或 18 岁的所有人）。留级生、海外转学生与提前离校生在两种规则下会被分入不同类别。
Institution type. Some measures cover only state-funded mainstream schools; others add independent schools, further education colleges, alternative provision and hospital schools. Adding a sector changes both numerator and denominator.
学校类型。部分指标只覆盖公立主流学校，另一些则纳入私立学校、继续教育学院、替代性教育机构与医院学校。纳入一个板块会同时改变分子与分母。
Residents versus enrolled students. A local authority may count young people who live in its area, while a school counts those on its roll. Mobile students are double-counted in one version and lost in the other.
常住人口与学生注册数。地方政府可能统计居住在本辖区内的年轻人，而学校统计的是本校在读学生。流动学生在一种口径下被重复计算，在另一种口径下则被漏掉。
Leavers versus the whole cohort. A "leavers" measure excludes students who stayed at the same institution, which raises destination rates mechanically because stayers are almost always in education.
离校生与全体队列。"离校生"口径会排除留在原校的学生，这会在机械上抬升去向率，因为留校学生几乎总是处于教育状态。
Consider a concrete effect: a city has 20,000 residents aged 17. If 15,000 attend local institutions but 500 of those are actually resident in a neighbouring borough, then a residency-based base is 20,000 while a roll-based base is 14,500 - a 38% difference in the denominator before any destination is even recorded.
举一个具体的效应：某城市有 20,000 名 17 岁常住居民，其中 15,000 人在本地学校就读，但这 15,000 人中有 500 人实际居住在邻区。此时按常住口径基数为 20,000，按注册口径基数则为 14,500——在记录任何去向之前，分母就相差了 38%。
3. The Reference Date and the Snapshot Problem | 参考日期与快照问题
Destinations are states that change, sometimes week by week. A young person may be unemployed in October, on a training course in January and in an apprenticeship by May. A single snapshot therefore gives a partial picture, and two snapshots give two different partial pictures of the same life.
去向是一种会变化的状态，有时每周都在变。一个年轻人可能 10 月失业、1 月参加培训课程、5 月已进入学徒制。因此单次快照只能给出部分图景，两次快照则给出同一段人生的两幅不同的部分图景。
Most official measures work with two or three collection points, typically an autumn census and a spring census, and then define a "sustained" destination. A sustained destination normally requires participation across at least two consecutive terms or around six months, verified at more than one collection point. Spot measures count a person as being in a destination if they were there on one particular week.
大多数官方指标采用两到三个采集时点，通常是秋季普查与春季普查，然后界定"持续性"去向。持续性去向通常要求在至少连续两个学期或约六个月内保持参与，并在一个以上的采集时点得到确认。而"时点"指标只要某人在某一周处于该状态，就计入该去向。
estination: participation recorded at ≥ 2 collection points spanning ≥ 2 consecutive terms
The statistical consequence is systematic, not random. Sustained measures always produce lower employment and training rates than spot measures, because they filter out short-term and unstable placements. A school that reports 85% employment on a spot basis may report 68% on a sustained basis without a single number being wrong.
其统计后果是系统性的而非随机的。持续性口径下的就业率与培训率总是低于时点口径，因为它过滤掉了短期与不稳定的安置。一所学校在时点口径下报出 85% 的就业率，在持续性口径下可能报出 68%，而这两个数字都没有错。
4. Destination Categories and Their Hierarchy | 去向分类及其层级
Categories must be mutually exclusive and, ideally, exhaustive. In practice they rarely are, which creates the familiar "other" and "not known" columns. The table below summarises the categories used in most 16-19 destination measures and the ambiguity each one carries.
分类必须互斥，最好还能穷尽。实际操作中往往做不到，于是出现了人们熟悉的"其他"与"不详"两列。下表总结了大多数 16 至 19 岁去向指标使用的类别，以及每一类所携带的歧义。
| Category | Typical inclusion | Common ambiguity |
|---|---|---|
| Education | School sixth form, FE college, higher education | Part-time courses, unaccredited provision, distance learning |
| Apprenticeship | Paid apprenticeship with training contract | Whether it is also counted as employment |
| Employment | Paid work above a minimum hours threshold | Zero-hours contracts, casual and gig work, unpaid family work |
| Training | Government-funded training programmes | Whether short courses qualify |
| NEET | Not in education, employment or training | Whether "not known" is folded in |
| Not known | No response or unmatched record | Excluded from the base, or treated as NEET |
| Other / inactive | Moved abroad, in custody, deceased, long-term ill | Often omitted entirely, shrinking the base |
The hierarchy matters too. If apprenticeships are counted inside employment, the employment rate rises without any change in behaviour. Analysts call this double counting; examiners call it a failure of mutual exclusivity.
分类层级同样重要。如果学徒制被计入就业，就业率就会上升，而实际行为毫无改变。分析者称之为重复计算，考官则称之为未满足互斥性。
5. NEET: The Most Misread Acronym | NEET：最容易被误读的缩写
NEET stands for "not in education, employment or training". It is a residual category, defined by what a person is not rather than what they are. That construction makes it highly sensitive to definitional choices at both ends.
NEET 指"未升学、未就业、未受训"。它是一个残差类别，由一个人"不是"什么来定义，而非由他"是"什么来定义。这种构造使它两端都极易受口径选择影响。
First, the hours and duration thresholds. Is a person working four hours a week employed? Is a one-week unpaid work experience placement "training"? Most official definitions require paid work above a minimum hours threshold and recognised training of a minimum duration, but the thresholds differ between publishers.
第一是工时与持续时间阈值。每周工作四小时算就业吗？为期一周的无薪工作体验算"受训"吗？多数官方定义要求高于最低工时阈值的带薪工作，以及达到最低持续时间的受认可培训，但各发布方的阈值并不一致。
Second, the treatment of "not known". If 8% of a cohort cannot be traced, a publisher may (a) exclude them from the base, (b) count them as NEET, or (c) report them separately. The three choices produce three very different headline rates from identical raw data.
第二是"不详"的处理。如果队列中有 8% 的人无法追踪，发布方可以：(a) 将其从基数中剔除；(b) 将其计入 NEET；(c) 单独报告。在原始数据完全相同的情况下，这三种选择会产生三个截然不同的头条比率。
education, and 400 untraced. Excluding the untraced gives a NEET rate of 300 ÷ 4,600 = 6.5%. Including them gives 300 ÷ 5,000 = 6.0% n one convention, but if the untraced are assumed NEET the rate becomes 700 ÷ 5,000 = 14.0%. The same cohort spans 6.0% to 14.0%.
假设某 5,000 人队列中有 300 人确认处于 NEET，400 人处于就业或教育状态，另有 400 人无法追踪。剔除无法追踪者，NEET 率为 300 ÷ 4,600 = 6.5%；若将其计入分母则为 300 ÷ 5,000 = 6.0%；若将无法追踪者全部假定为 NEET，则变为 700 ÷ 5,000 = 14.0%。同一队列在 6.0% 到 14.0% 之间摆动。
6. Data Sources: Administrative, Survey and Longitudinal | 数据来源：行政数据、抽样调查与纵向追踪
Three broad sources feed destination statistics, and each has a characteristic error structure.
去向统计主要来自三类数据源，每一类都有其典型的误差结构。
Administrative records. School censuses, college enrolment returns, tax and benefits records and apprenticeship databases. These are near-complete for the population they cover, but they only see activity that is registered somewhere. Informal work, self-employment and unpaid caring are largely invisible.
行政记录。学校普查、学院注册申报、税务与福利记录、学徒制数据库等。对其覆盖的人群而言几乎是完整的，但只能看到在某个系统中登记过的活动。非正式工作、自雇与无偿照护基本不可见。
Sample surveys. A random or stratified sample of young people is interviewed, giving richer detail on hours, pay and satisfaction. The cost is sampling error, and often a bias towards respondents who are easier to contact.
抽样调查。对随机或分层抽样的年轻人进行访问，可获得关于工时、薪酬与满意度的更丰富细节。代价是抽样误差，并且往往偏向那些更容易联系上的受访者。
Longitudinal cohort studies. The same individuals are followed for years, allowing causal analysis of pathways. Attrition is the main threat: disadvantaged participants drop out of the study at higher rates, biasing later waves.
纵向队列研究。对同一批个体追踪多年，可以用于路径的因果分析。主要威胁是样本流失：弱势参与者的退出率更高，使后续轮次产生偏差。
7. Non-Response, Attrition and Bias | 无应答、样本流失与偏差
Missing data is the single largest source of disagreement between destination measures, because there is no neutral way to handle it. The choice of method is a judgement, and judgements differ.
缺失数据是各去向指标之间分歧的最大来源，因为不存在中立的处理方式。方法选择本身就是一种判断，而判断因人而异。
Missingness can be broadly classed as random with respect to the outcome, or systematically related to it. If non-response is random, dropping missing cases leaves the estimate unbiased. If non-response is concentrated among NEET young people - which is typically the case, since they are harder to trace - then complete-case analysis understates the NEET rate.
缺失可大致分为与结果无关的随机缺失，以及与结果系统性相关的缺失。若缺失是随机的，剔除缺失案例不会使估计产生偏差。若缺失集中于 NEET 年轻人——这通常正是实情，因为他们更难追踪——那么完整案例分析会低估 NEET 率。
mated NEET = (observed NEET + k × untraced) ÷ (observed total + untraced), where 0 ≤ k ≤ 1
Here k is the assumed proportion of untraced young people who are actually NEET. Publishing the value of k, and testing sensitivity to it, is good statistical practice. Omitting it is the most common way that a destination table misleads without lying.
其中 k 是假定的无法追踪者中实际处于 NEET 的比例。公布 k 的取值并对其做敏感性检验，是良好的统计实践。省略 k，则是去向数据表在不撒谎的前提下误导读者的最常见方式。
8. Numerators, Denominators and the Right Kind of Average | 分子、分母与正确的平均方式
Destination rates are usually reported for groups of very different sizes, so aggregating them requires weighted averages. An unweighted mean of institutional rates treats a school of 40 students and a college of 4,000 students as equally important.
去向率通常针对规模差异极大的群体报告，因此汇总时必须使用加权平均。对院校比率取未加权平均，等于把只有 40 名学生的学校与有 4,000 名学生的学院视为同等重要。
Weighted mean = Σ(wᵢ xᵢ) ÷ Σwᵢ
For example, region A has 5,000 learners with a 5% NEET rate and region B has 15,000 learners with a 9% NEET rate. The unweighted mean is (5 + 9) ÷ 2 = 7.0% ct combined rate is (5,000 × 0.05 + 15,000 × 0.09) ÷ 20,000 = (250 + 1,350) ÷ 20,000 = 8.0%. The unweighted answer understates the national figure by a full percentage point.
例如 A 地区有 5,000 名学习者，NEET 率为 5%；B 地区有 15,000 名学习者，NEET 率为 9%。未加权平均为 (5 + 9) ÷ 2 = 7.0%，但正确的合并比率是 (5,000 × 0.05 + 15,000 × 0.09) ÷ 20,000 = (250 + 1,350) ÷ 20,000 = 8.0%。未加权的结果把全国数字整整低估了一个百分点。
When comparing two rates, distinguish percentage points from per cent. A rise from 6.4% to 7.8% increase of 1.4 percentage points, which is a relative increase of 1.4 ÷ 6.4 × 100 = 21.9% tatements are true, and choosing between them is a rhetorical act, not a mathematical one.
比较两个比率时，要区分"百分点"与"百分比"。从 6.4% 升到 7.8%，增加了 1.4 个百分点，相对增幅为 1.4 ÷ 6.4 × 100 = 21.9%。两种说法都成立，选择哪一种属于修辞选择，而非数学选择。
9. Uncertainty: Confidence Intervals and Small Numbers | 不确定性：置信区间与小样本
Destination data are frequently quoted without any measure of uncertainty, which gives an impression of precision that the data cannot support. A proportion estimated from a sample has a standard error, and a confidence interval follows directly.
去向数据常常不带任何不确定性度量地被引用，从而营造出数据无法支撑的精确感。由样本估计的比例具有标准误，置信区间可直接导出。
95% CI = p ± 1.96 × √(p(1 − p) ÷ n)
Take a published NEET rate of 6.4% based on 1,200 traced young people. Then the standard error is √0.064 × 0.936 ÷ 1,200 = √0.00004992 = 0.00707, and the 95% interval is 0.064 ± 0.0139, that is 5.0% to 7.8%. A neighbouring authority reporting 7.1% from a similar sample size is therefore not measurably different.
假设某公布 NEET 率为 6.4%，基于 1,200 名被追踪到的年轻人。则标准误为 √0.064 × 0.936 ÷ 1,200 = √0.00004992 = 0.00707，95% 区间为 0.064 ± 0.0139，即 5.0% 至 7.8%。因此，邻近地区基于相近样本量报告的 7.1% 在统计上并无显著差异。
n: multiply the standard error by √((N − n) ÷ (N − 1)). Second, for proportions close to 0% or 100% ould be used. Suppression rules in published tables usually exist for exactly these cases.
在 A-Level 层面有两个细节值得注意。第一，从小规模有限总体中进行无放回抽样时，应使用有限总体修正：将标准误乘以 √((N − n) ÷ (N − 1))。第二，当比例接近 0
Published by TutorHao | Exam Prep Revision Series | aleveler.com
更多咨询请联系16621398022（同微信）
或截图后用「扫一扫 → 相册」选择本图