📚 Ranking vs Rating: Complementary Relationships in Educational Evaluation | 排名与评级的互补关系：谁最强与谁在哪方面强
Every year students, parents and teachers ask two questions that look identical but are not: "Which is the best?" and "What is each one actually good at?" The first question demands a ranking; the second demands a rating. Confusing the two is one of the most expensive mistakes in educational decision-making, because a ranking compresses a rich profile into a single line, while a rating expands a single entity into a full profile. This article explains the mathematical and practical relationship between the two instruments, and shows how to use them together.
每年，学生、家长和老师都会问两个看起来一样、其实完全不同的问题："哪个最好？"和"各自究竟强在哪里？"第一个问题需要排名，第二个问题需要评级。把两者混为一谈，是教育决策中最昂贵的错误之一：排名把丰富的画像压缩成一行，评级则把一个对象展开成一张完整画像。本文将解释这两种工具在数学与实践层面的互补关系，并说明如何配合使用。
1. Two Questions, Two Instruments | 两个问题，两种工具
A ranking is an ordinal instrument. It places objects in a sequence from best to worst and answers a comparative question: who is strongest relative to the others inside this particular comparison.
排名是一种序数工具。它把对象从最好到最差排成一个序列，回答的是一个比较性问题：在这次特定的比较中，谁相对于其他人最强。
A rating is a cardinal instrument. It assigns a value, grade or profile to a single object on its own terms, and answers a descriptive question: who is strong in what, and by how much.
评级是一种基数工具。它按对象自身的标准，给出一个数值、等级或画像，回答的是描述性问题：谁在哪些方面强，强多少。
Both instruments are legitimate, and neither is complete on its own. A ranking without a rating tells you the order but not the reason. A rating without a ranking tells you the profile but not the position.
两种工具都合理，但单独使用时都不完整。只有排名没有评级，你知道顺序却不知道原因；只有评级没有排名，你知道画像却不知道位置。
Typical examples in education illustrate the division of labour clearly:
教育领域中的典型例子能清楚展示这种分工：
- Ranking instruments: QS World University Rankings, THE World University Rankings, national school league tables, A-level class position lists.
排名工具：QS 世界大学排名、THE 世界大学排名、各国学校排行榜、A-Level 班级名次表。 - Rating instruments: the Teaching Excellence Framework (Gold / Silver / Bronze), Ofsted inspection grades, rubric-based coursework feedback, CEFR language levels.
评级工具：英国教学卓越框架（金 / 银 / 铜）、Ofsted 督查等级、基于评分量规的作业反馈、CEFR 语言等级。
The key insight is that the two instruments answer complementary halves of the same decision: selection versus fit.
关键洞见在于：两种工具回答的是同一个决策的两个互补侧面——筛选与匹配。
2. Ordinal vs Cardinal: The Core Mathematical Difference | 序数与基数：核心数学差异
In measurement theory, data are classified into nominal, ordinal, interval and ratio scales. Rankings live on the ordinal scale; ratings normally live on the interval or ratio scale.
在测量理论中，数据被分为名义、序数、区间和比率四个尺度。排名属于序数尺度；评级通常属于区间或比率尺度。
On an ordinal scale, only the order is meaningful. The statement "University A is ranked 3rd and University B is ranked 4th" tells you nothing about the size of the gap between them.
在序数尺度上，只有顺序有意义。"A 大学第 3 名、B 大学第 4 名"这句话，完全没有告诉你两者之间的差距有多大。
On an interval scale, differences are meaningful. The statement "University A scores 92 and University B scores 88" allows you to say that A is four points ahead.
在区间尺度上，差值有意义。"A 大学得分 92、B 大学得分 88"这句话，允许你说 A 领先 4 分。
A rank can always be derived from a rating, but a rating can rarely be recovered from a rank. This asymmetry is the mathematical root of the complementarity principle.
排名总是可以由评级推导出来，但评级几乎无法从排名中还原。这种不对称性，正是互补性原则的数学根源。
R(xᵢ) = 1 + #{ j : xⱼ > xᵢ } and z = (x − μ) ÷ σ
The first expression converts a score into a rank. The second converts a raw score into a standardised rating, where μ is the mean and σ is the standard deviation.
第一个表达式把分数转成名次；第二个表达式把原始分转成标准化评级，其中 μ 是平均值，σ 是标准差。
A crucial consequence: averaging ranks is mathematically dubious unless the underlying scores are roughly equally spaced. Yet many informal "average ranking" tables do exactly this, which is why their results can be misleading.
一个关键推论是：除非底层分数大致等距，否则对名次取平均在数学上是可疑的。然而许多非正式的"平均排名"表恰恰这么做，这也是其结果可能误导人的原因。
3. What a Ranking Reveals - and What It Hides | 排名告诉你什么，又掩盖了什么
The strength of a ranking is compression. It reduces dozens of indicators to one ordered list, producing instant comparability and a clear headline for decision-makers.
排名的长处是压缩。它把几十个指标压缩成一张有序清单，带来即时的可比性和清晰的头条结论。
A ranking is excellent for shortlisting, for setting thresholds ("top 50 only"), and for tracking movement over time. It is fast, memorable and easy to communicate.
排名非常适合初筛、设定门槛（"只看前 50"）以及追踪随时间的变化。它快速、好记、易于沟通。
However, a ranking systematically destroys information:
但排名会系统性地破坏信息：
- It hides distance: rank 1 and rank 2 may differ by 0.1 points or by 30 points.
它隐藏距离：第 1 名和第 2 名可能只差 0.1 分，也可能差 30 分。 - It hides the criteria: the weights behind the list are often invisible.
它隐藏标准：榜单背后的权重往往不可见。 - It hides distribution: a tightly clustered field and a widely spread field look identical.
它隐藏分布：密集的群体和分散的群体看起来一模一样。 - It creates cliff effects: small score changes can cause large rank jumps near ties.
它制造断崖效应：在分数接近处，微小变化会造成名次大跳。 - It invites Goodhart's law: when a measure becomes a target, it ceases to be a good measure.
它招致古德哈特定律：当一个指标成为目标，它就不再是好指标。
In short, a ranking answers "who is strongest" but it is silent on "strong in what".
简言之，排名回答"谁最强"，却对"强在哪些方面"保持沉默。
4. What a Rating Reveals - and What It Costs | 评级告诉你什么，代价是什么
A rating preserves dimensionality. Instead of one number, you obtain a profile across several criteria - teaching quality, research output, student satisfaction, graduate outcomes, cost of living.
评级保留了维度。你得到的不是一个数字，而是跨多个标准的画像——教学质量、科研产出、学生满意度、毕业去向、生活成本。
This makes ratings diagnostic and actionable. If a rating shows that a school scores highly on pastoral support but low on stretch for high attainers, the response is specific rather than general.
这使评级具有诊断性和可操作性。如果评级显示某校在 pastoral 关怀上得分很高、但在拔尖学生的拔高上得分很低，那么应对措施就是具体而非笼统的。
Ratings are also criterion-referenced, which makes them more stable than rankings. A school can be rated "Good" without any reference to what other schools achieve.
评级还是标准参照的，因此比排名更稳定。一所学校可以仅凭自身表现被评为"良好"，而无需参照其他学校。
The costs of ratings are equally real:
评级的代价同样真实：
- No single headline: profiles are harder to compare at a glance.
缺少单一结论：画像难以一眼比较。 - Weighting disputes: any composite rating embeds value judgements.
权重争议：任何综合评级都嵌入了价值判断。 - Scale drift and grade inflation: over time, "Good" may quietly mean less.
尺度漂移与等级膨胀：随着时间推移，"良好"的含义可能悄悄缩水。 - Ceiling and floor effects: excellent and merely adequate candidates hit the same top band.
天花板与地板效应：优秀与尚可的申请者可能落在同一最高档。 - Self-report bias in satisfaction-based indicators.
以满意度为基础的指标存在自我报告偏误。
A rating answers "strong in what" but it does not, by itself, tell you who is strongest overall.
评级回答"强在哪些方面"，但它本身并不能告诉你谁总体最强。
5. The Complementarity Principle | 互补性原则
The complementarity principle states that rankings and ratings are not competing methods but sequential stages of one decision process.
互补性原则指出：排名与评级不是互相竞争的方法，而是同一决策过程中的先后阶段。
Stage one uses ranking to reduce a large choice set to a manageable shortlist. Stage two uses rating to compare the shortlisted options on the criteria that actually matter to the individual.
第一阶段用排名把庞大的选择集压缩成可管理的候选清单；第二阶段用评级，在真正与个人相关的标准上比较这些候选对象。
Rank to shorten → Rate to choose → Rank again with personal weights
The third step matters more than most students realise. A ranking built on national average preferences is not the same as a ranking built on your preferences.
第三步比大多数学生意识到的更重要。基于全国平均偏好构建的排名，并不等于基于你个人偏好构建的排名。
Information loss is also complementary. If a ranking hides distance, the rating restores it. If a rating hides overall position, the ranking restores it. Used together, the two instruments recover most of what either one loses alone.
信息损失也是互补的。如果排名隐藏了距离，评级把它还原；如果评级隐藏了总体位置，排名把它还原。两者合用，可以找回单独使用时丢失的大部分信息。
6. Multi-Criteria Decision Analysis | 多准则决策分析
Multi-criteria decision analysis (MCDA) is the formal framework that unites rankings and ratings. You list criteria, assign weights, score each option, and compute a weighted total.
多准则决策分析（MCDA）是统一排名与评级的正式框架。你列出标准、分配权重、为每个选项打分，然后计算加权总分。
(wᵢ × sᵢ) with Σ wᵢ = 1 and 0 ≤ wᵢ ≤ 1
Here wᵢ is the weight of criterion i, sᵢ is the score on that criterion, and S is the composite score. Sorting options by S produces a personalised ranking.
其中 wᵢ 是第 i 项标准的权重，sᵢ 是该标准上的得分，S 是综合得分。按 S 排序，就得到了个性化排名。
| Criterion 标准 | Weight 权重 | Option A | Option B | Option C |
| Course strength 专业实力 | 0.35 | 9 | 7 | 8 |
| Teaching quality 教学质量 | 0.25 | 6 | 9 | 7 |
| Cost 成本 | 0.20 | 5 | 6 | 9 |
| Location 地点 | 0.20 | 8 | 5 | 6 |
| Composite 综合 | 1.00 | 7.15 | 6.85 | 7.55 |
Notice that Option A wins on the single most heavily weighted criterion, yet Option C wins overall. This is precisely the difference between "strongest in one dimension" and "strongest in the composite".
注意：A 在权重最高的单项标准上胜出，但 C 在总体上胜出。这正是"某一维度最强"与"综合最强"的区别。
7. Weighting: The Hidden Author of Both Systems | 权重：两种体系背后的隐形作者
Every published ranking is a weighted composite. The weights are chosen by the publisher, not discovered in nature. Changing them changes the order.
每一个公开排名都是加权综合。权重由发布方选择，而非从自然中发现。改变权重就改变顺序。
This means that a ranking is a point of view expressed as a number. Reading a league table without reading its methodology is like reading a verdict without reading the law.
这意味着排名是一种用数字表达的观点。只看排行榜不看方法论，就像只读判决不读法律。
Sensitivity analysis is the remedy. Recompute the composite under two or three different weight sets and see whether the top choice changes.
敏感性分析是解药。用两组或三组不同权重重新计算综合分，看首选是否改变。
| Weight set 权重方案 | Winner 胜出者 | Interpretation 解读 |
| Reputation-heavy 重声誉 | Option A | Best for prestige-driven goals 适合追求声誉的目标 |
| Teaching-heavy 重教学 | Option B | Best for classroom experience 适合课堂体验优先 |
| Cost-heavy 重成本 | Option C | Best for budget constraints 适合预算受限 |
When the winner is stable across all weight sets, the decision is robust. When it flips, the decision is a genuine value trade-off and should be made deliberately, not by default.
当胜出者在所有权重方案下都稳定时，这个决策是稳健的；当它翻转时，说明这是真正的价值权衡，应当有意识地做出，而不是听凭默认。
8. Normalisation and Z-Scores: Why Rank Swaps Happen | 归一化与 Z 分数：名次为何会互换
Criteria are measured in different units - percentage points, money, star ratings. Before combining them you must normalise, or the criterion with the largest raw numbers will dominate by accident.
各项标准的计量单位不同——百分点、金额、星级。合并之前必须先归一化，否则原始数值最大的那项标准会意外地主宰结果。
Two common methods are min-max scaling and standardisation. Min-max rescales to the range 0 to 1; standardisation expresses each score as a number of standard deviations from the mean.
两种常用方法是最小-最大缩放和标准化。前者把数值重新缩放到 0 到 1 区间；后者把每个分数表达为距平均值多少个标准差。
z = (x − μ) ÷ σ min-max: x' = (x − xmin) ÷ (xmax − xmin)
Z-scores are preferable when outliers exist, because a single extreme value can crush the rest of the distribution under min-max scaling.
存在离群值时，Z 分数更可取，因为在最小-最大缩放下，一个极端值就能把其余分布压扁。
Rank swaps between two published league tables are usually caused by three things: different weightings, different normalisation methods, and different indicator definitions.
两份公开排行榜之间的名次互换，通常由三个原因造成：权重不同、归一化方法不同、指标定义不同。
To measure how similar two rankings are, use Spearman's rank correlation coefficient, where dᵢ is the difference in ranks for item i and n is the number of items.
要衡量两份排名的相似程度，可使用斯皮尔曼等级相关系数，其中 dᵢ 是第 i 个项目的名次差，n 是项目数量。
ρ = 1 − (6 × Σ dᵢ²) ÷ (n × (n² − 1))
A value of ρ close to 1 means the two systems largely agree. A value near 0.6 means a third of the ordering is essentially noise-driven.
ρ 接近 1 意味着两套体系大体一致；接近 0.6 则说明排序中有相当一部分本质上是由噪声驱动的。
9. A Practical Protocol for Reading Any League Table | 阅读任何排行榜的实用流程
Before you use a ranking, run through a short checklist. It takes ten minutes and prevents years of regret.
在使用任何排名之前，先走一遍简短检查清单。花十分钟，可以避免多年的后悔。
- Who published it, and what is their business model?
谁发布的？其商业模式是什么？ - Which indicators are included, and with what weights?
包含哪些指标？权重是多少？ - Are the indicators absolute, or relative to institutional size?
指标是绝对值，还是相对于机构规模的比值？ - What is the time lag between data collection and publication?
数据收集与发布之间有多长的时滞？ - How large are the gaps between adjacent ranks?
相邻名次之间的差距有多大？ - Is there a rating report that explains the position?
是否有评级报告解释这个位置？
If a ranking cannot answer these questions, treat it as a marketing document rather than a measurement instrument.
如果一个排名无法回答这些问题，就把它当作营销材料，而不是测量工具。
The rule of thumb is simple: use the ranking to decide what to investigate, and the rating to decide what to believe.
经验法则很简单：用排名决定"要调查什么"，用评级决定"该相信什么"。
10. Case Study: A-Level Subject and University Choices | 案例：A-Level 选科与大学选择
Consider a student choosing between two sixth forms with nearly identical league-table positions. The ranking says they are equivalent; the rating says they are not.
假设一名学生在两所排行榜位置几乎相同的高中之间选择。排名说两者等价，评级说并非如此。
- School X rates highly on A* and A grades in Mathematics and the sciences, with strong progression to Russell Group universities.
X 校在数学与理科的 A* 和 A 比例上评级很高，升入罗素集团大学的比例也高。 - School Y rates highly on value-added measures, meaning students enter with lower GCSE profiles and improve substantially.
Y 校在增值指标上评级很高，说明学生入学时 GCSE 成绩较低，但进步显著。
The "who is strongest" question is unanswerable at the level of the ranking, because both occupy the same position. The "strong in what" question is answerable, and it is the one that should drive the decision.
"谁最强"这个问题在排名层面无法回答，因为二者位置相同；"强在哪些方面"可以回答，而且它才应该驱动决策。
The same logic applies to examination boards. Comparisons between AQA, Pearson Edexcel, OCR and Cambridge International are largely rating questions about specification structure, assessment style
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