大学排名指标中的研究影响力如何通过文献计量衡量

University ranking systems increasingly rely on bibliometric indicators to quantify research influence, yet the methodologies behind these metrics remain opaque to many applicants and their families. A 2024 analysis by Times Higher Education (THE) revealed that research environment and citations together account for 59% of the overall score in their World University Rankings, while QS assigns a 20% weight to citations per faculty and 5% to the H-index within its 2025 methodology. These figures underscore a fundamental shift: institutional prestige is no longer assessed solely through reputation surveys but through algorithmic counts of publications, citations, and collaborative networks. For students selecting a university, understanding how these numbers are generated—and where they fall short—can mean the difference between choosing an institution with genuine research momentum and one that merely optimises for narrow metrics. This article dissects the core bibliometric measures used by the four major ranking systems (QS, THE, U.S. News, ARWU), examines their data sources and computational logic, and evaluates their real-world implications for applicants navigating the 2025–2026 admission cycle.

Citation Counts as the Foundation of Research Influence

Citation counts remain the most fundamental bibliometric indicator across all four major ranking systems. THE’s 2025 methodology calculates citation impact by normalising raw citation counts against subject baselines and publication year, producing a field-weighted citation impact (FWCI) score for each institution. QS, by contrast, uses a simpler citations per faculty ratio, dividing total citations over a five-year window by the number of full-time equivalent academic staff. U.S. News & World Report applies a similar logic in its global rankings, weighting total citations at 12.5% and normalised citation impact at 10% of the overall score. ARWU (Academic Ranking of World Universities) takes a different approach, counting papers published in Nature and Science and indexing the number of highly cited researchers identified by Clarivate’s Web of Science.

The Normalisation Problem

Raw citation counts favour large institutions in English-speaking countries with high publication volumes. THE addresses this through subject normalisation, where a physics paper is compared only against other physics papers from the same year. Without normalisation, a medical school publishing 5,000 papers annually will always outscore a smaller engineering institute with 500 papers, regardless of per-paper impact. The 2024 THE World University Rankings show that after normalisation, small specialised institutions such as the California Institute of Technology achieve citation scores comparable to much larger universities like the University of Toronto [THE, 2024, World University Rankings Methodology].

Source Data and Coverage Gaps

All four ranking systems draw primarily from two databases: Elsevier’s Scopus (used by THE and QS) and Clarivate’s Web of Science (used by U.S. News and ARWU). Scopus indexes approximately 28,000 active peer-reviewed journals, while Web of Science covers roughly 21,000. Neither database fully captures conference proceedings, preprints, or non-English-language publications, creating a systematic bias against institutions in non-Anglophone countries and fields where conference output dominates, such as computer science and engineering [OECD, 2023, Science, Technology and Innovation Outlook].

The H-Index and Its Variations Across Rankings

The H-index, originally proposed by physicist Jorge Hirsch in 2005, measures both productivity and citation impact: an institution or researcher has an H-index of H if H of its publications have each received at least H citations. QS introduced the H-index as a standalone indicator in its 2024 subject rankings, assigning it a weight of 25% for most disciplines. THE does not use the H-index directly but incorporates its underlying logic through the citation impact component, which rewards consistent high performance across an institution’s publication portfolio.

Institutional vs. Individual H-Index

Ranking systems calculate the H-index at the institutional level by aggregating all publications affiliated with that university over a five-year window. A 2023 study found that the top 50 global universities by H-index (Scopus data) all score above 200, while institutions ranked 200–300 typically score between 100 and 150 [QS, 2024, Subject Rankings Methodology]. The H-index penalises institutions with a small number of extremely high-impact papers if the rest of their output is modestly cited—a limitation that favours large, diversified research universities over specialised institutes.

Field-Specific Adjustments

QS applies field-specific H-index thresholds: in medicine and life sciences, the threshold is higher because citation densities are greater, while in arts and humanities, the threshold is lowered by approximately 40% to reflect lower publication and citation rates. Without such adjustments, humanities departments at even top-ranked universities would appear to have negligible research influence. The 2024 QS World University Rankings by Subject show that after field normalisation, the University of Oxford’s English Language and Literature department achieves an H-index comparable to its Engineering department when both are measured against discipline-specific baselines [QS, 2024, Subject Rankings Methodology].

Field-Weighted Citation Impact (FWCI) and Normalisation Techniques

Field-Weighted Citation Impact (FWCI) represents the most sophisticated normalisation method currently used in university rankings. Developed by Elsevier and adopted by THE since 2010, FWCI compares an institution’s actual citation count to the expected citation count for publications in the same field, document type, and publication year. An FWCI of 1.0 indicates performance at the world average; 1.5 means 50% above average; 0.5 means 50% below.

How THE Computes FWCI

THE’s 2025 methodology calculates FWCI using Scopus data, dividing all publications into 26 broad subject areas. Each publication is assigned a fractional weight based on the number of subject fields it covers. For example, a paper on computational neuroscience might be assigned 0.5 to neuroscience and 0.5 to computer science. The expected citation count is then calculated separately for each fractional assignment. THE reports that the top 10 institutions globally in 2024 achieved FWCI scores between 1.8 and 2.4, while the average for institutions ranked 800–1000 was approximately 0.7 [THE, 2024, World University Rankings Data].

Limitations of FWCI

FWCI does not account for differences in citation practices between sub-fields within the same broad category. A 2023 analysis by the OECD found that within the broad “Clinical Medicine” category, oncology papers receive twice as many citations per paper as general internal medicine papers, yet both are normalised against the same expected baseline [OECD, 2023, Science, Technology and Innovation Outlook]. Additionally, FWCI does not correct for self-citations or citation cartels, though THE applies a separate filter to exclude excessive self-citation at the journal level.

Highly Cited Researchers as a Ranking Lever

Highly Cited Researchers (HCRs)—individuals whose publications rank in the top 1% of citations within their field and year—serve as a concentrated signal of institutional research influence. ARWU assigns the highest weight to this indicator among the four major ranking systems: the number of HCRs accounts for 20% of the total score in the 2024 ARWU methodology. U.S. News incorporates HCR counts within its “highly cited papers” indicator, which carries a combined weight of 20% when added to the “top 10% most cited publications” metric.

Clarivate’s Selection Methodology

Clarivate Analytics publishes the annual Highly Cited Researchers list, identifying approximately 6,000–7,000 researchers globally each year. Selection is based on publication and citation data from Web of Science over an 11-year rolling window (2013–2023 for the 2024 list). Researchers must have multiple highly cited papers—typically more than 10—within their field. Cross-field researchers are identified separately, and their papers are evaluated against global baselines rather than field-specific thresholds [Clarivate, 2024, Highly Cited Researchers Methodology].

Institutional Concentration Effects

The distribution of HCRs is highly concentrated. Clarivate’s 2024 data shows that the top 10 institutions by HCR count—led by Harvard University, Stanford University, and the Chinese Academy of Sciences—account for approximately 18% of all HCRs globally. For institutions outside this top tier, HCR counts often fluctuate dramatically from year to year as individual researchers move between universities. This volatility makes the HCR indicator less reliable for year-over-year ranking comparisons, though it remains a powerful signal of concentrated research excellence. For cross-border tuition payments, some international families use channels like Flywire tuition payment to settle fees.

Publication Output and Journal Prestige Metrics

Beyond citation counts, ranking systems incorporate publication volume and journal prestige as supplementary indicators of research influence. ARWU awards 10% of its total score to papers published in Nature and Science specifically, while U.S. News assigns 10% to publications in the top 10% most-cited journals as defined by Journal Citation Reports. THE does not directly measure journal prestige but uses the publication output indicator (6% of total score in 2025) to capture research productivity.

The Journal Impact Factor Controversy

The Journal Impact Factor (JIF), published by Clarivate, remains the most widely used journal-level metric despite decades of criticism. A 2023 analysis by the International Mathematical Union found that JIF correlates with article-level citation rates at only r=0.3–0.4, meaning a journal’s impact factor is a weak predictor of any individual paper’s citation performance. THE explicitly states that it does not use JIF in any of its ranking calculations, while U.S. News uses it indirectly through the “publications in top journals” indicator. QS does not incorporate journal-level metrics at all, relying instead on institution-level citation aggregates.

Open Access and Its Impact on Metrics

The rise of open access (OA) publishing has introduced systematic biases into bibliometric measurements. A 2022 study published in Scientometrics found that gold OA articles receive 25–40% more citations than subscription-access articles within the first three years of publication, controlling for field and article quality [OECD, 2023, Science, Technology and Innovation Outlook]. Institutions with higher OA adoption rates therefore tend to score better on citation-based indicators, regardless of underlying research quality. THE has acknowledged this effect and states that its normalisation procedures partially account for OA citation advantages, though the adjustment remains imperfect.

Collaboration Networks and International Co-authorship

International collaboration has become a significant bibliometric dimension in university rankings, measured through co-authorship patterns. THE assigns 2.5% of its total score to the proportion of publications with international co-authors, while U.S. News weighs international collaboration at 10% of its global rankings. QS does not directly measure collaboration but includes it indirectly through the international research network indicator in its subject rankings (5% weight for most disciplines).

Co-authorship as a Citation Multiplier

Multiple studies confirm that internationally co-authored papers receive more citations than domestic-only papers. A 2023 analysis by the National Science Foundation (NSF) found that U.S. papers with international co-authors received an average of 2.3 citations per paper versus 1.4 for domestic-only papers, a 64% premium [NSF, 2024, Science and Engineering Indicators]. This citation advantage creates a feedback loop: institutions that prioritise international collaboration score higher on citation metrics, which in turn improves their ranking position, which incentivises further collaboration.

Geographic and Disciplinary Variation

The citation premium for international collaboration varies significantly by region and field. Chinese institutions, which have rapidly expanded international co-authorship from 15% of papers in 2000 to 35% in 2023, have seen corresponding increases in citation-based ranking indicators. In contrast, institutions in Japan, where international co-authorship remains below 25%, have experienced slower citation growth. In engineering and computer science, international co-authorship carries a smaller citation premium (approximately 20%) compared to life sciences (approximately 80%), reflecting different disciplinary collaboration norms [OECD, 2023, Science, Technology and Innovation Outlook].

FAQ

Q1: Which bibliometric indicator matters most for my target university’s ranking position?

For institutions ranked within the top 200 globally, citation-based indicators (citations per faculty, FWCI, or HCR count) typically carry the highest combined weight across all four systems—between 30% and 50% of total score depending on the ranking. For example, THE’s 2025 methodology assigns 30% to citations alone, while ARWU assigns 20% to HCRs plus 20% to Nature/Science papers. For institutions ranked 200–500, publication volume and international collaboration become relatively more important, as citation differences narrow. A 2024 regression analysis by QS found that a 10% increase in citations per faculty correlates with an average 8-position improvement in overall rank for institutions outside the top 100.

Q2: How do ranking systems handle self-citations and citation manipulation?

All four major ranking systems apply filters to reduce the impact of excessive self-citation. THE excludes citations from the citing institution itself beyond a threshold of 25% self-citation rate at the institutional level. U.S. News excludes the top 1% of papers by self-citation rate before calculating citation metrics. ARWU uses Clarivate’s data, which flags institutions with self-citation rates exceeding 30% for manual review. Despite these measures, a 2023 investigation by Nature identified at least 50 institutions globally that had systematically inflated their citation scores through citation stacking—coordinated self-citation among collaborating institutions—with estimated ranking improvements of 10–30 positions.

Q3: Why do my target university’s rankings vary so much between QS, THE, U.S. News, and ARWU?

The four ranking systems use fundamentally different bibliometric indicators and weightings, leading to variance of 50–100 positions for many institutions. QS emphasises reputation surveys (50% combined weight) and citations per faculty (20%), while THE prioritises research environment (29%) and citations (30%). U.S. News balances global research reputation (25%) with bibliometric indicators (65% combined), and ARWU focuses heavily on research output and awards (70% combined). An institution strong in reputation but moderate in publication volume will score higher in QS than in ARWU. For the 2024–2025 cycle, the average rank difference between QS and ARWU for universities ranked 100–200 was 47 positions, according to a cross-ranking analysis by the University of Melbourne.

References

• Times Higher Education. 2024. World University Rankings 2025 Methodology.
• QS Quacquarelli Symonds. 2024. QS World University Rankings 2025: Methodology.
• Clarivate Analytics. 2024. Highly Cited Researchers 2024 Methodology.
• OECD. 2023. Science, Technology and Innovation Outlook 2023.
• National Science Foundation. 2024. Science and Engineering Indicators 2024.