TL;DR
This paper reveals a fundamental dichotomy in scientific and technological advancement: large teams excel at developing and refining existing ideas, while small teams are the primary drivers of disruption and novelty. By analyzing millions of research papers, patents, and software projects, the authors demonstrate that supporting both types of teams is crucial for sustained innovation.
What Problem Does This Paper Address?
The increasing trend towards larger research teams in science and technology has led to a critical question: how do large and small teams differ in the nature of their contributions? While larger teams are often associated with increased output and impact, it's unclear if they foster the same kind of groundbreaking, disruptive innovation that has historically come from smaller, more agile groups. This paper investigates this question, aiming to understand the distinct roles that team size plays in scientific and technological progress.
Methods
The authors employ a multi-faceted approach to quantify "disruption" and "development" in scientific and technological outputs. They analyze citation networks to understand how new ideas build upon or diverge from existing work. Key metrics include:
- Disruption Measure (D): A quantitative score (ranging from -1 to 1) that distinguishes between work that develops existing ideas (negative D) and work that disrupts them by introducing novel concepts (positive D).
- Sleeping Beauty Index: Measures the delay in a work's attention, indicating belated appreciation for disruptive ideas.
- Citation Counts: Used as a proxy for impact.
They also utilize advanced text analysis techniques, such as Doc2vec, to control for topic similarity and isolate the effect of team size on disruption.
Datasets
The study draws upon three large-scale datasets:
- Web of Science (WOS): Over 43 million research articles published between 1900 and 2014, with a focus on the period 1954-2014.
- U.S. Patent and Trademark Office (USPTO): Nearly 5 million patents granted between 1976 and 2014, with a focus on applicant citations from 2002-2014.
- GitHub: Over 15 million code repositories contributed between 2011 and 2014.
Key Insights and Ideas
- Disruption vs. Development Dichotomy: The most striking finding is the consistent pattern that small teams tend to disrupt science and technology with new ideas, while large teams develop existing ones with solutions and refinements.
- Impact and Disruption Diverge with Team Size: While larger teams may produce more impactful work (higher citations), this impact is increasingly developmental, not disruptive. Small teams, even if their work is less immediately impactful, are more likely to be disruptive.
- Search Behavior Matters: Small teams tend to build on older, less popular ideas, which can lead to more novel contributions. Large teams, conversely, focus on recent, high-impact work, leading to more immediate relevance but less disruption.
- Temporal Dynamics: Disruptive work from small teams often experiences a "sleeping beauty" effect, with delayed recognition, while large team developments are recognized more quickly.
- Funding Paradox: Small teams receiving funding from government agencies show less disruptive tendencies, suggesting that funding mechanisms might inadvertently favor development over disruption.
Key Results
- As team size increases, the likelihood of producing disruptive work dramatically declines. For example, as teams grow from one to fifty members, their papers, patents, and software decrease in disruption by 70%, 30%, and 50%, respectively.
- Solo authors are as likely to produce a "hit paper" (top 5% citations) as teams of five, but their articles are 72% more likely to be highly disruptive.
- Ten-person teams are 50% more likely to score a hit paper but tend to develop existing ideas, showing very low disruption.
- The negative correlation between disruption and impact in the short term (≤10 years) turns positive in the long term (>10 years), suggesting that disruptive work, though initially less impactful, can lead to greater long-term influence.
Limitations and Future Work
The authors acknowledge potential biases in their datasets, such as the reliance on journal articles (excluding conference proceedings in some fields) and the unknown representativeness of GitHub for all open-source software projects. They also note that while their models control for author and topic effects, there may be subtle differences in project types that influence disruption.
Future work could explore the mechanisms by which small teams foster disruption, investigate how funding agencies can better support disruptive research, and examine the role of interdisciplinary collaboration in both disruption and development. The paper calls for new investigations into the vital role of individuals and small groups in advancing science and technology, especially in light of the increasing dominance of large teams.
Summary updated