Persistent geographical biases in global scientific collaboration and citations

Based on a large-scale analysis of 39.35 million publications from 2000 to 2022, this study reveals that while intellectual influence through citations diffuses relatively freely across borders, geographic distance continues to increasingly constrain scientific collaboration, highlighting persistent structural biases that favor domestic networks and the United States while systematically underciting Chinese research.

The Gist

Imagine the world of science as a giant, bustling global marketplace. In this market, researchers are like traders, and their ideas are the goods they trade. For a long time, we hoped that the internet and modern technology would turn this marketplace into a borderless bazaar where anyone could talk to anyone, regardless of where they live.

This paper, however, acts like a detective looking at the receipts and shipping logs of this marketplace. It asks: Is the world actually becoming a borderless bazaar, or are old walls still standing?

Here is the story of what they found, told in simple terms.

1. The "Gravity" of Distance: Why It's Harder to Shake Hands Than to Send a Letter

The researchers looked at two main ways scientists interact:

• Collaboration (Shaking Hands): Working together to write a paper.
• Citation (Sending a Letter): Reading someone else's work and giving them credit.

The Finding:
They discovered that distance still matters a lot for shaking hands. Even with Zoom and email, scientists are much more likely to work with someone in their own city or country than with someone on the other side of the world. In fact, this "distance penalty" has actually gotten worse over time. It's as if the world is getting more crowded, making it harder to reach out to distant neighbors.

The Twist:
However, distance matters very little for sending letters. When it comes to citing (reading and referencing) work, geography barely slows anyone down. A scientist in Tokyo can easily cite a paper from a scientist in Toronto. The "intellectual influence" travels freely, like a digital postcard, but the "physical partnership" is still stuck in traffic.

2. The "Home Team" Bias: We Love Our Own

The study found that scientists have a strong "Home Team" bias.

• Collaboration: Most scientists prefer to work with people in their own country. It's like a local sports team; you know the players, you speak the same language, and you trust them.
• Citation: Scientists also prefer to cite work from their own country, though this bias is slightly weaker than the collaboration bias.

3. The US vs. China: The One-Way Street

The most dramatic finding involves the two scientific superpowers: the United States and China.

• Collaboration: In the past, the US was the most popular partner for everyone. Now, China has become a very popular partner for collaboration. Countries are eager to "shake hands" with China.
• Citation (The Problem): Here is where the bias shows up. While countries are happy to work with China, they are not giving China enough credit in their "letters" (citations).
  • Chinese scientists cite US work constantly.
  • US scientists rarely cite Chinese work.
  • Other countries cite US work heavily but often overlook Chinese work.

The Analogy: Imagine a popular new restaurant (China) that everyone wants to eat at (collaborate with). But when the food critics (scientists) write reviews, they keep praising the old, famous restaurant (the US) and barely mention the new one, even though the new one is serving great food. This creates an unfair system where China is doing the work but not getting the fame.

4. The "Traveling Scholar" Experiment

To test if this was just about personal bias or something deeper, the researchers looked at scientists who moved from one country to another (migrants).

• The Result: When a scientist moves to a new country, they quickly start collaborating with their new neighbors. They make new friends.
• The Catch: But their citation habits don't change much. They still tend to cite the same old famous papers from their home country or the US, regardless of where they live now.

This proves that the bias isn't just about "hating" other countries; it's about structural habits. The system is wired to recognize certain countries (like the US) as the "standard," while others have to work twice as hard to get noticed.

The Big Picture: Why Should We Care?

The authors conclude that technology alone hasn't fixed the problem. Just because we can connect globally doesn't mean we do connect fairly.

• The "Death of Distance" is a Myth: We thought the internet would kill the importance of geography, but for teamwork, geography is still king.
• The "Core-Periphery" Problem: The global science system is like a pyramid. A few countries (the "Core," like the US) sit at the top, getting all the credit and citations. Others (the "Periphery") do the work but get less recognition.

The Takeaway:
If we want science to solve global problems (like climate change or pandemics), we need a fairer marketplace. We need to stop assuming that "global" means "equal." We need to actively fix the biases so that brilliant work from anywhere in the world gets the credit it deserves, not just the work from the usual suspects.

Technical Summary

1. Problem Statement

The study addresses a critical gap in understanding how geography and national structures influence global scientific knowledge flows in the digital age. While scientific ideals suggest a borderless system where knowledge circulates freely, empirical evidence suggests persistent barriers.

• Core Question: How do geographic distance and country-level factors shape international scientific collaboration and citation patterns?
• The Ambiguity: Previous studies often conflate "country bias" with "geographic distance." The authors argue that without controlling for distance and latent city-level tendencies, observed country effects may be spurious.
• The Goal: To disentangle the causal effects of geographic distance from inherent country-specific biases (structural or institutional) in both collaboration (co-authorship) and citation behaviors.

2. Methodology

The authors employed a large-scale bibliometric analysis combined with advanced statistical modeling and a quasi-experimental design.

A. Data Source and Processing

• Dataset: OpenAlex, covering 39.35 million publications from 2000 to 2022.
• Scope: 95 countries and 3,794 cities.
• Pairs Analyzed:
  • 2,444,439 city-level collaboration pairs (undirected).
  • 14,754,699 city-level citation pairs (directed).
• Data Cleaning:
  • City Disambiguation: Used a clustering approach (Constant Potts Model optimized by the Leiden algorithm) to resolve organizations with identical city names but distinct geographic locations (e.g., St. Denis in Paris vs. Réunion).
  • Filtering: Restricted to journal articles; removed self-citations; retained cities with $\ge$50 publications/year and countries with $\ge$3 qualifying cities.

B. Statistical Modeling (Bayesian Framework)

To identify causal country effects, the authors modeled collaboration and citation counts using a Poisson distribution within a Bayesian framework.

• The Model:
  $$\lambda_{ij} = \alpha \cdot n_i \theta_i \cdot n_j \theta_j \cdot d_{ij}^{-\beta} \cdot C_{c_i, c_j}$$
  Where:
  • $\lambda_{ij}$: Expected rate of interaction between city $i$ and $j$.
  • $n_i, n_j$: Publication volume (strength) of the cities.
  • $\theta_i, \theta_j$: Latent collaboration/citation propensity of the cities (controlling for quality/reputation).
  • $d_{ij}$: Geographic distance.
  • $\beta$: The decay coefficient of distance.
  • $C_{c_i, c_j}$: Country-level preference parameter (the key variable of interest).
• Identification Strategy: By explicitly controlling for distance ($d_{ij}$) and latent city tendencies ($\theta$), the model isolates the direct causal effect of the country pair ($C$).
• Priors: Used informative priors (e.g., half-normal for $\beta$) to ensure robust estimation.

C. Mobility Study (Matching Design)

To further validate country effects and rule out individual researcher preferences, the authors analyzed migrant scholars.

• Definition: Scholars who moved from Country A to Country B with at least 5 years of publication history before and after the move.
• Sample: 8,179 identified migrants.
• Matching: Used a propensity score-like matching design to pair migrants with non-migrant scholars based on:
  1. First publication year.
  2. Original or destination country.
  3. Pre-migration publication volume and research topic similarity (using OpenAlex concept vectors).
• Logic: If collaboration patterns shift immediately upon moving to a new country (while citation patterns remain stable), it suggests the "country" acts as a structural constraint or enabler, rather than just personal bias.

3. Key Results

A. Impact of Geographic Distance

• Collaboration: Distance has a strong, intensifying effect. The decay coefficient ($\beta$) for collaboration increased from 0.37 (2000) to 0.48 (2022). Contrary to the "death of distance" hypothesis, physical proximity remains a stronger driver for co-authorship than ever.
• Citation: Distance has a weak, diminishing effect. The decay coefficient is minimal (~0.02) and has declined since 2001. Intellectual influence diffuses more freely across borders than joint production does.
• Domestic vs. International: Within-country collaboration declines much slower with distance than cross-border collaboration.

B. Country-Level Biases

• Domestic Preference: Most countries exhibit a strong bias toward domestic collaboration and citation.
• The US Exception: The US is the primary exception, receiving high citation and collaboration preference globally.
• The China-US Asymmetry:
  • Collaboration: Other countries increasingly prefer collaborating with China (reversing the trend from 2010), but US-China bilateral collaboration remains low/negative.
  • Citation: A stark asymmetry exists. Other countries cite US work significantly more than Chinese work. While citation preference for China has improved slightly, it remains systematically under-cited compared to its collaboration standing. The US rarely cites Chinese papers, while China cites US papers heavily.

C. Mobility Findings

• Collaboration Shift: Upon migration, scholars' collaboration networks shift rapidly to the host country. Collaboration with the origin country drops sharply. This confirms that collaboration is heavily constrained by the current institutional/geographic context.
• Citation Stability: Citation patterns show little change after migration. Migrants continue to cite (and be cited by) international sources similarly to their pre-move patterns, suggesting citation is driven by intellectual lineage rather than current location.

4. Key Contributions

1. Methodological Rigor: The study moves beyond simple correlation by using a Bayesian latent variable model to explicitly control for distance and city-level quality, isolating true "country bias."
2. Decoupling Mechanisms: It demonstrates that collaboration and citation operate under different logics:
   • Collaboration is a process of co-creation, heavily dependent on spatial proximity, trust, and tacit knowledge exchange (physical constraints).
   • Citation is an act of recognition, driven by visibility and intellectual impact, which transcends borders more easily.
3. Evidence of Structural Bias: The paper provides robust evidence that "country bias" is not merely a function of distance. Even when distance is controlled, systematic under-citation of non-Western (specifically Chinese) research persists, pointing to structural inequalities in the global scientific ecosystem.
4. Mobility as a Natural Experiment: The use of migrant scholars effectively separates individual preferences from structural country effects, showing that changing one's location changes collaboration networks but not citation flows.

5. Significance and Implications

• Policy Implications: Digital infrastructure alone cannot solve global scientific inequality. The persistence of distance effects in collaboration suggests that physical mobility, funding exchanges, and institutional partnerships are necessary to overcome barriers.
• Equity in Science: The systematic under-citation of Chinese research, despite rising collaboration, highlights a "core-periphery" structure in global recognition. This limits the diversity of global knowledge production and reinforces the dominance of Western (specifically US) science.
• Future Research: The findings call for a re-evaluation of academic evaluation metrics and funding policies to ensure equitable participation. It suggests that while knowledge recognition (citations) is becoming more global, knowledge production (collaboration) remains deeply local and structurally biased.

Limitations Noted by Authors:

• Reliance on affiliation metadata may miss informal collaborations.
• Equal weighting of all citations/collaborations ignores depth or motivation.
• Statistical models struggled with Health Sciences data.
• Analysis was limited to the top 10% of cities due to computational constraints.