The Scale of Reputation Surveys: Nominating Around 7,000 Institutions a Year | 声誉调查的规模：每年评估约 7,000 所院校的提名

📚 The Scale of Reputation Surveys: Nominating Around 7,000 Institutions a Year | 声誉调查的规模：每年评估约 7,000 所院校的提名

A reputation survey looks almost trivially simple: ask a few tens of thousands of academics and employers which universities they consider excellent, then count the votes. In practice the exercise runs at industrial scale. Before a single institution can be placed in a published table, the organisation behind the survey must build and maintain a candidate universe of roughly 7,000 degree-granting, research-active institutions worldwide, resolve thousands of name variants into single records, weight every response by discipline and region, and then compress that universe into the few hundred positions a league table can actually defend statistically. This article explains how that machinery works, where the 7,000 figure comes from, and what reputation indicators can and cannot tell an applicant.

声誉调查表面上简单得近乎琐碎：请几万名学者和雇主说出他们心中优秀的大学，然后数票。但实际运作是工业级规模的工程。在任何一所院校进入正式榜单之前，调查机构必须先建立并维护一个约 7,000 所、可授予学位且有科研产出的全球候选院校库，把成千上万个名称变体合并成唯一记录，按学科与地区对每一份回答加权，再把这一池子压缩到排行榜在统计上真正站得住脚的那几百个位置。本文将说明这套机制如何运转、7,000 这个数字从何而来，以及声誉指标对申请者意味着什么、又不意味着什么。

A note on sourcing before we begin: the figures quoted below are drawn from publicly published ranking methodologies and annual methodological reports. Exact counts shift every cycle, and any number given here should be read as the order of magnitude that the current design produces, not as a fixed constant.

先说数据来源：下文引用的数字来自各大排名机构公开的方法说明与年度方法报告。具体数值每个周期都会变动，文中的数字应理解为当前设计所能产生的量级，而非固定常数。

1. What a Reputation Survey Actually Measures | 声誉调查究竟在测量什么

A reputation survey measures the beliefs of informed observers, not the performance of an institution. When a senior chemist ticks a university she has never worked at, she is reporting a summary judgement assembled from journal papers, conference networks, former colleagues, doctoral students and press coverage. The resulting indicator is therefore a perception indicator. It has strong face validity for visibility and prestige, and only indirect validity for teaching quality, student support, class size or graduate salaries.

声誉调查测量的是有见识的观察者的信念，而不是院校的实际绩效。当一位资深化学教授勾选一所她从未工作过的大学时，她给出的判断是由论文、会议网络、昔日同事、博士生和媒体报道拼合而成的总体印象。因此这个指标本质上是知觉指标：它对知名度和声望具有很高的表面效度，但对教学质量、学生支持、班级规模或毕业生薪资只有间接效度。

That distinction shapes interpretation. Reputation scores correlate strongly with research volume, institutional age, endowment size and language of instruction. They correlate weakly with the daily experience of an undergraduate. A small, outstanding teaching college that publishes little will rarely surface in a reputational survey, because nobody outside its own region has a reason to type its name.

这一区分决定了如何解读分数。声誉得分与科研体量、建校历史、捐赠基金规模和授课语言高度相关，却与本科生的日常体验相关性很弱。一所教学出色但科研产出很少的小型学院，几乎不会出现在声誉调查中，因为其所在地区之外的人没有理由写出它的名字。

2. Where the 7,000 Figure Comes From | 约 7,000 所这个数字从何而来

The 7,000 figure describes a candidate universe, not a ranking. Ranking organisations assemble this universe from bibliometric databases such as Scopus and the Web of Science, from national registers of recognised degree-awarding institutions, from previous survey cycles, and from open nominations made by respondents themselves. Eligibility screens normally require degree-awarding status plus a minimum research output over a rolling multi-year window. The outcome is a working list in the region of 7,000 institutions: far larger than any published table, yet small enough to be curated and cleaned by hand each cycle.

7,000 这个数字描述的是候选院校池，而不是排行榜。排名机构从 Scopus、Web of Science 等文献计量数据库、各国官方认可院校名录、往届调查周期，以及受访者自己填写的开放式提名中汇总出这个池子。准入筛选通常要求院校可授予学位，并在滚动多年的时间窗内达到最低科研产出门槛。结果是得到一份约 7,000 所院校的工作名单：它比任何公开榜单都大得多，却又小到可以每轮人工清洗与维护。

Two engineering problems dominate this stage. The first is entity resolution: matching ‘Universidad de Buenos Aires’, ‘UBA’ and ‘Univ. Buenos Aires’ to a single record, while keeping genuinely distinct campuses apart. The second is coverage: institutions in lower-income systems are less likely to publish in indexed journals, so they enter the bibliometric pool later, with fewer papers, or not at all. Coverage bias is therefore baked in before a single vote is cast.

这一阶段有两个主导性的工程难题。第一是实体消歧：把 “Universidad de Buenos Aires”、”UBA” 和 “Univ. Buenos Aires” 归并成同一条记录，同时把确实不同的校区区分开。第二是覆盖偏差：低收入体系的院校更难在被索引的期刊上发表论文，因此更晚进入文献计量池、论文更少，甚至完全进不来。也就是说，在第一张选票投出之前，覆盖偏差已经被写进了池子里。

3. The Nomination Funnel: From 7,000 Names to a Published Table | 提名漏斗：从 7,000 个名字到公开榜单

It helps to picture the process as a funnel with four gates. Gate one is the candidate universe of about 7,000 institutions. Gate two keeps only those that receive at least one valid nomination during the cycle. Gate three applies output and eligibility thresholds, removing institutions without enough indexed research to be assessed. Gate four produces the published table of roughly 1,500 to 2,100 institutions, depending on the provider and the year.

把整个过程想象成有四个闸门的漏斗会更清楚。第一道闸门是约 7,000 所的候选院校池；第二道只保留在本周期内至少获得一次有效提名的院校；第三道施加科研产出与资格门槛，剔除没有足够被索引研究成果可供评估的院校；第四道输出约 1,500 至 2,100 所的公开榜单，具体数量取决于机构与年份。

The critical point is that the overwhelming majority of the 7,000 never appear in print. They are nominated occasionally, they are carried forward for the next cycle, and they exist in the dataset mainly to keep the scale of the exercise honest. When a ranking provider says it ‘evaluates’ several thousand institutions, that is what is being described: a screening and nomination universe, not a table with 7,000 rows.

关键之处在于：这 7,000 所中的绝大多数从未出现在纸面上。它们偶尔被提名，被结转到下一周期，在数据集中存在的主要意义是让整个调查的规模保持真实。当排名机构说它”评估”了数千所院校时，说的正是这件事：一个筛选与提名的院校池，而不是一张有 7,000 行的榜单。

4. Sampling Design: Who Gets Invited to Vote | 抽样设计：谁被邀请投票

Respondents are not selected by simple random sampling from the world’s academics. Invitation lists are built from author databases and from previous respondents, and the panels are then balanced by quota: by country or region, by broad discipline, and sometimes by seniority. Quota design protects the survey from being dominated by voices from a handful of large English-speaking systems, which is a genuine improvement in fairness.

受访者并不是从全球学者中做简单随机抽样得到的。邀请名单来自作者数据库和往届受访者，随后按配额对样本结构进行平衡：按国家或地区、按大学科门类，有时还按资历。配额设计可以防止调查被少数几个大型英语国家的意见主导，这在公平性上是实质性的改进。

The trade-off is statistical. A quota sample is not a probability sample, so the textbook margin of error does not strictly apply. If it did, the arithmetic would look like this:

代价是统计上的。配额样本不是概率样本，因此教科书上的误差范围严格来说并不适用。如果硬要套用，算式如下：

MOE ≈ 1.96 × √( p(1 − p) ÷ n )

With n = 100,000 responses and p = 0.5, this gives roughly ±0.3 percentage points for a global share. That number is real but misleading: it describes uncertainty about the aggregate, not uncertainty about any individual institution’s score, which is the quantity a reader actually cares

Published by TutorHao | Exam Prep Revision Series | aleveler.com

更多咨询请联系16621398022（同微信）