📚 Why Different Ranking Systems Give Very Different Results | 为什么不同排名体系给出的名次会差很多
Two families can sit down with the same university, the same publicly available data and the same good intentions, and still end up looking at tables that place that university twenty places apart. This is not a scandal, and it is usually not incompetence either. It is the predictable outcome of different definitions, different weights, different data sources and different questions. Learning to read why the numbers diverge is one of the most useful analytical skills an international applicant can build.
两个家庭面对同一所大学、同一批公开数据、同样善意，却可能看到相差二十名的榜单。这既不是丑闻，通常也不是无能，而是不同定义、不同权重、不同数据与不同问题带来的必然结果。学会解读名次为何分歧，是国际申请者能建立的最实用的分析能力之一。
1. The Puzzle: One University, Many Ranks | 谜题：同一所大学，多个名次
The core puzzle is simple to state. A university is a fixed object with a fixed history, a fixed staff body and a fixed set of outputs. If a ranking were a neutral measurement of quality, every well-constructed ranking should agree within a small margin of error. In practice, global rankings often disagree by ten to forty places for the same institution, and domestic league tables can disagree far more.
这个谜题本身很好表述。一所大学是一个相对固定的对象：固定的历史、固定的师资规模、固定的产出集合。如果排名是对质量的"中立测量"，那么所有构造良好的排名应当在小幅误差内彼此一致。但现实中，同一所大学在不同全球排名中常相差十几到四十名，而各国本土榜单的分歧往往更大。
The explanation is that rankings are not measurements. They are indices, deliberately constructed summaries that compress many different variables into one number. Every step of that compression involves a judgement call, and every judgement call can be challenged.
解释在于：排名并不是"测量"，而是"指数"——人为构造的汇总，把许多不同变量压缩成一个数字。压缩的每一步都包含主观判断，而每一步判断都可以被质疑。
2. What Rankings Are Actually Measuring | 排名究竟在衡量什么
Before comparing tables, ask what each one claims to measure. Some measure research influence. Some measure the undergraduate student experience. Some measure employability. Some try to blend all three and call the result "world university rank", which is where much of the confusion begins.
比较榜单之前，先问：它声称衡量什么？有的衡量科研影响力，有的衡量本科生就读体验，有的衡量就业力，还有的试图把三者混合后称为"世界大学排名"——混淆往往由此开始。
A ranking that measures research output is not wrong; it is simply answering a different question from a ranking that measures teaching satisfaction. If a student uses a research-weighted table to choose an undergraduate programme with heavy seminar teaching, they are using a tool for the wrong job.
衡量科研产出的排名并不"错"，它只是回答了另一个问题。如果学生用科研权重很高的榜单去选一个以研讨课为主、重视教学的本科项目，那就是用错了工具。
3. Weighting: The Hidden Hand Behind Every Table | 权重：每张榜单背后的隐形之手
Suppose two systems use exactly the same six indicators. If one gives research citations 60 per cent of the total weight and the other gives it 20 per cent, the resulting order can be completely different. Weighting is the single most powerful lever in ranking design, and it is also the least visible to the reader.
假设两个体系使用完全相同的六项指标。如果一方给科研引用 60% 权重，另一方只给 20%，最终排序可能完全不同。权重是排名设计中最有力的杠杆，也是对读者最不显眼的一环。
The general form of a weighted composite is easy to write down. Each university receives a standardised score on each indicator, and those scores are then combined.
加权综合分的一般形式很容易写出。每所大学在每项指标上得到一个标准化分数，随后将这些分数加总。
S = Σ (wᵢ × zᵢ) ÷ Σ wᵢ
Here wᵢ is the weight of indicator i and zᵢ is the standardised score. Change a single wᵢ by five percentage points and a mid-table university with an unusual profile can move several places, sometimes in either direction depending on which indicators it happens to be strong in.
其中 wᵢ 是指标 i 的权重，zᵢ 是标准化后的得分。只要把某个 wᵢ 改动五个百分点，一所特点鲜明的中游大学就可能移动好几个名次，甚至方向不定——取决于它恰好在哪项指标上强。
4. Indicator Choice and Data Sources | 指标选择与数据来源
Even with identical weights, two rankings will diverge if they choose different indicators or draw on different data. One system may rely on a survey of academics, another on the Web of Science citation database, another on Scopus. These databases index different journals, with different coverage of non-English scholarship, different subject biases and different lag times.
即便权重相同，只要指标选择或数据来源不同，两个排名也会分歧。一个体系依赖学者问卷，另一个依赖 Web of Science 引文数据库，还有一个使用 Scopus。这些数据库收录的期刊不同，对非英语学术成果的覆盖不同，学科偏向不同，数据滞后也不同。
Consider a concrete example. If one ranking counts publications in indexed journals only, an institution strong in practice-based design, clinical medicine or law may look weak, while a large science faculty will look powerful. The indicator has not measured quality; it has measured the fit between the institution and the database.
举个具体例子：若某排名只统计被索引期刊上的论文，那么在实践型设计、临床医学或法学方面很强的院校可能显得薄弱，而大型理科学院则显得强大。这个指标并没有衡量质量，它衡量的是院校与数据库之间的"匹配度"。
5. Reputation Surveys and the Circularity Problem | 声誉调查与自我循环
Several major rankings give a large share of weight to reputation surveys, in which academics and employers are asked to name the strongest institutions in their field. Reputation correlates with quality, but it also correlates with age, brand advertising, historical wealth and previous ranking positions.
几个主要排名把很大一部分权重交给声誉调查：请学者和雇主提名本领域最强的院校。声誉与质量相关，但同时也与校龄、品牌宣传、历史财富以及往年排名位置相关。
This creates a feedback loop. A university ranked highly this year is more likely to be named next year by respondents who saw the previous table. The survey is not fabricated, but it is partly a mirror held up to the ranking itself, which stabilises the top and makes it hard for a strong but unfamiliar institution to break through.
这就形成反馈回路：今年排名高的大学，更容易被看过上一版榜单的受访者提到。调查并非造假，但它多少是一面照向排名自身的镜子——它让头部位置趋于稳定，也让实力强但知名度低的院校难以突围。
6. Size, Age and Discipline Mix: Apples and Oranges | 规模、校龄与学科结构：苹果与橘子
Institutions differ in ways that no weighting scheme can fully neutralise. A large research university with a medical school, an engineering faculty and a hospital produces far more publications and citations than a small specialist college of music or business, regardless of teaching quality on either side.
院校之间的差异，没有任何权重方案能完全抹平。拥有医学院、工学院和附属医院的大型研究型大学，其论文与引用数量远超小型音乐学院或商学院，这与双方教学质量无关。
Age matters too. Older institutions have larger alumni networks, more accumulated paper stock and more chances to have produced a Nobel laureate. If a system gives 20 per cent of its weight to alumni awards, it is effectively rewarding a century of accumulated history rather than current performance.
校龄同样重要。老校的校友网络更大、累积论文存量更多、产生诺奖得主的概率也更高。如果某体系给校友获奖 20% 权重，它实际上在奖励一个世纪的历史积累，而不是当下表现。
There is also a statistical trap here known as Simpson's paradox: an institution can be strong in every individual discipline yet look mediocre in a combined faculty-level average, simply because it happens to be large in disciplines where absolute citation counts are low, such as mathematics or the humanities.
这里还有一个统计学陷阱，即"辛普森悖论"：一所院校可能在每个单一学科都强，但在院系层面的合并均值上显得平庸——仅仅因为它规模较大的领域恰好是数学、人文学科这类绝对引用量偏低的学科。
7. The Mathematics of Normalisation | 标准化的数学
Raw indicators are measured in incompatible units: citation counts run into the tens of thousands, while survey scores sit between 0 and 100. Before they can be added together, they must be standardised, and the standardisation method itself changes the results.
原始指标的计量单位互不兼容：引用量可以上万，而问卷分数在 0 到 100 之间。要把它们相加，必须先标准化，而标准化方法本身就会改变结果。
The most common approach is the z-score, which expresses each value as a distance from the mean in units of standard deviation.
最常见的方法是 Z 分数，它把每个数值表示为"距均值多少个标准差"。
z = (x − μ) ÷ σ
Where x is the raw score, μ is the mean across all institutions and σ is the standard deviation. Alternatives include min-max scaling, which maps the best performer to 1 and the worst to 0, and percentile ranking, which converts positions into a 0 to 100 scale.
其中 x 为原始分，μ 为全体院校均值，σ 为标准差。替代方案包括极值标准化（把最佳者映射为 1，最差者映射为 0）与百分位排名（把位次转换为 0 到 100 的刻度）。
These choices are not equivalent. z-scores are sensitive to extreme outliers, so one institution with an enormous citation count can compress everyone else towards the middle. Min-max scaling has the opposite weakness: a single outlier can squeeze all other institutions into a narrow band near the bottom. Percentile ranking discards the size of gaps entirely, so a two-point difference at the top is treated exactly like a two-point difference at rank 400.
这些选择并不等价。Z 分数对极端离群值敏感，一所引用量惊人的院校会把其他所有人压向中间。极值标准化有相反的弱点：单个离群值会把其余院校挤到靠近底部的窄带里。百分位排名则完全丢弃差距大小，第 1 名与第 2 名之间的两分差，和第 400 名与第 401 名之间的两分差被同等对待。
8. Bibliometrics and the Citation Snowball | 文献计量与引用滚雪球
Citation-based indicators carry their own systematic distortions. Citation practices vary sharply between fields: a molecular biology paper may collect fifty citations within three years, while a landmark mathematics paper may take a decade to reach twenty. Rankings that do not normalise by field will therefore systematically favour the life sciences, medicine and engineering.
基于引用的指标带有自身的系统性偏差。各学科的引用习惯差异极大：分子生物学论文三年内可能被引五十次，而一篇数学奠基性论文可能要十年才达到二十次。因此，不做学科归一化的排名会系统性地偏向生命科学、医学与工程。
There is also a snowball effect. Highly cited researchers tend to attract more co-authors, more funding and more invitations, which generates more citations. Because the effect compounds, small initial differences can widen over time and produce very different positions in rankings that measure citations per faculty versus those that measure total citations volume.
还存在"滚雪球效应"。高被引学者更容易吸引合作者、经费与邀约，从而产生更多引用。由于这一效应会累积，初始的小差距会随时间放大，导致"师均引用"类指标与"引用总量"类指标给出截然不同的位次。
A further subtlety is that a university's citation score often reflects a small number of very productive research groups rather than the institution as a whole. The indicator is real and quantitative, but the entity it describes may not be the entity a prospective undergraduate is choosing.
更微妙的是，一所大学的引用得分常常只反映少数高产课题组，而非整所院校。指标本身真实且量化，但它所描述的"对象"，可能并不是本科申请者正在选择的那个对象。
9. Missing Data, Imputation and Tie-Breaking | 缺失数据、插补与并列
Not every institution reports every indicator. Some do not return survey responses, some do not publish staff numbers, and some are too new to have alumni records. Ranking organisations must decide what to do with the gaps, and each option has consequences.
并非每所院校都会上报每项指标。有的不回应问卷，有的不公布师资数量，有的太新以致没有校友记录。排名机构必须决定如何处理空缺，而每种处理方式都会带来后果。
Common approaches include imputing a regional or subject-group average, assigning a zero, or excluding the institution from that indicator entirely and re-weighting the rest. Assigning zero is the harshest and can push a genuinely strong university down many places for a purely administrative reason.
常见做法包括：用地区或学科组的平均值插补、直接记为零分、或将该指标整体剔除并对剩余指标重新加权。记零最为严苛，可能仅因为一个行政原因就把一所实力不俗的大学往下推很多名。
Tie-breaking rules matter as well. When composite scores are rounded to one decimal place, dozens of institutions can share the same value, and the published order may then depend on an alphabetical or secondary technical rule. A difference of one or two places near the top of a table is therefore almost never statistically meaningful.
并列处理规则同样重要。当综合分四舍五入到一位小数时，可能有数十所院校数值相同，最终公布的顺序或许取决于字母序或某项次级技术规则。因此榜单前列一两个名次的差异，几乎从来不具统计意义。
10. The Big Four Side by Side | 四大排名并排比较
It helps to see the four most widely consulted global rankings in one place. The figures below are approximate headline weights and are revised regularly, so always check the current methodology page before drawing conclusions.
把最常被查阅的四个全球排名放在一起看会很有帮助。下表中的数字为近似的主要权重，且会定期调整，因此下结论前务必查看当年的方法说明。
| System | What it rewards | Typical skew |
| QS World University Rankings | Academic reputation about 30%, employer reputation about 15%, citations per faculty about 20%, faculty to student ratio about 10%, plus international ratios, sustainability, employment outcomes and research network | Favours large, well-branded, English-speaking and internationally oriented institutions |
| THE World University Rankings | Teaching about 29.5%, research environment about 29%, research quality about 30%, international outlook about 7.5%, industry income about 4% | Favours research-intensive, well-funded universities with strong doctoral activity |
| ARWU (Shanghai) | Staff awards about 20%, highly cited researchers about 20%, Nature and Science papers about 20%, total indexed papers about 20%, alumni awards about 10%, per-capita performance about 10% | Favours science-heavy, older, heavily endowed institutions; largely ignores teaching and the arts |
| US News Best Global Universities | Global and regional research reputation about 25% combined; the remainder is dominated by publications, citations and highly cited paper shares | Favours very large research output and, historically, North American institutions |
The pattern is clear. QS leans on reputation and employability, THE blends teaching with research environment and research quality, ARWU bets heavily on prizes and elite journal output, and US News is largely bibliometric. Four reasonable designs, four different questions, four different tables.
规律很清楚：QS 偏重声誉与就业力，THE 混合教学、科研环境与科研质量，ARWU 重押奖项与顶级期刊产出，US News 则基本是文献计量。四种合理的设计、四个不同的问题，自然得到四张不同的榜单。
11. What This Means for A-Level Applicants | 这对 A-Level 申请者意味着什么
For A-Level students, the practical implication is that a global research ranking is usually the wrong primary instrument. Undergraduate teaching in the UK, for instance, is assessed through a different ecosystem: the Teaching Excellence Framework, National Student Survey outcomes, graduate outcome data, entry standards and continuation rates.
对 A-Level 学生而言，实际含义是：全球科研排名通常不是合适的首要工具。以英国本科教学为例，它由另一套生态来评估：教学卓越框架、全国学生调查结果、毕业生去向数据、录取标准与续读率。
UK domestic league tables such as The Guardian University Guide, The Times and Sunday Times Good University Guide and the Complete University Guide therefore produce orders that look nothing like the global research tables. A small specialist institution with excellent seminar teaching and strong graduate prospects can sit far above a large research powerhouse in a teaching-weighted domestic table, and far below it in a citation-weighted global one. Both results can be correct.
因此，《卫报》大学指南、《泰晤士报》与《星期日泰晤士报》优秀大学指南、以及完全大学指南等英国本土榜单，给出的顺序往往与全球科研榜单大相径庭。一所小型专业院校若研讨课教学出色、毕业生前景良好，可能在偏重教学的本土榜单上远高于某大型研究强校，而在偏重引用的全球榜单上远低于它。两个结果都可能是"对的"。
What actually matters for a specific applicant is narrower and more personal, and it is rarely captured by any composite score.
对具体申请者真正重要的东西更窄、更个人化，而很少被任何综合分捕捉到。
- Course structure and module options | 课程结构与可选模块
- Contact hours and class size | 课时安排与班级规模
- Assessment mix, for example coursework versus closed examination | 考核方式，如论文作业与闭卷考试的比例
- Graduate destinations in your specific subject | 你所选专业的毕业生去向
- Cost, location and student support | 费用、地理位置与学生支持服务
- Accreditation or professional recognition for your intended career | 目标职业所需的认证或行业资格认可
Use rankings to generate a longlist, then use subject-level data and direct contact with departments to make the shortlist.
用排名来生成初选名单，再用学科层面的数据以及与院系的直接沟通来敲定最终名单。
12. How to Read a Ranking Like a Researcher | 像研究者一样读排名
A disciplined reader follows a short checklist every time a new table appears. This takes about five minutes and prevents most of the errors that rankings encourage.
有纪律的读者在每张新榜单出现时都会走一遍简短清单。这大约只需五分钟，却能避免排名最容易引发的多数误判。
- Check the methodology page first, not the results table | 先看方法说明页，而不是结果表
- Identify the top three weighted indicators and ask whether they matter for your goal | 找出权重最高的三项指标，问它们是否与你的目标相关
- Look for the subject-level ranking rather than the overall one | 查看学科排名，而非总榜
- Treat differences of a few places as noise, not signal | 把几个名次的差距视为噪声，而非信号
- Check whether the indicator is absolute or per capita | 确认指标是绝对量还是人均量
- Compare with at least one table built on different principles | 至少与一张原理不同的榜单交叉比对
- Read the year-on-year movement with caution; methodology changes are common | 谨慎解读年度变化，方法调整十分常见
The deeper lesson is about the nature of composite indices. Any single number that compresses dozens of variables must lose information in the process, and the information it loses is exactly the information a student most needs: what it is actually like to study a particular subject at a particular place. Rankings are useful maps, but a map is not the territory, and the person who forgets that is the one who is surprised when four tables disagree.
更深的启示在于综合指数本身的性质。任何把数十个变量压缩成单一数字的做法，都必然在过程中丢失信息，而丢掉的恰恰是学生最需要的信息：在某个地方读某个专业，究竟是怎样的体验。排名是有用的地图，但地图不是领土本身；忘记这一点的人，才会对四张榜单互相矛盾感到惊讶。
Rankings measure what they choose to measure. Always ask who chose, and why.
