Source-linked AI summary
Principles of scientific research team formation and evolution
Staša Milojević
TL;DR
The paper addresses the lack of a model for how scientific team sizes form and evolve. It combines Poisson formation of core teams with productivity-based cumulative growth of extended teams, reproducing 50 years of empirical distributions and decomposing authorship modes directly from data.
Problem
Scientific teams have grown substantially, but few models explain their basic property—team size—or its evolution across fields.
Method
The model combines Poisson-distributed core teams with extended teams that add members in proportion to aggregate productivity, then validates simulations against astronomy data.
Results
154,221 simulated articles reproduce the emergence and increasing prominence of the power-law tail, the low-k hook, and the shift toward two- or three-author peaks.
Takeaways & Limitations
The analytical decomposition estimates authorship-mode contributions directly from data and fits team-size distributions across astronomy and other fields.
Takeaways & Limitations
The model’s cumulative-advantage growth is best reproduced using members’ aggregate productivity as lead authors rather than productivity including coauthorships.
Abstract
from arXiv · showhide
Research teams are the fundamental social unit of science, and yet there is currently no model that describes their basic property: size. In most fields teams have grown significantly in recent decades. We show that this is partly due to the change in the character of team-size distribution. We explain these changes with a comprehensive yet straightforward model of how teams of different sizes emerge and grow. This model accurately reproduces the evolution of empirical team-size distribution over the period of 50 years. The modeling reveals that there are two modes of knowledge production. The first and more fundamental mode employs relatively small, core teams. Core teams form by a Poisson process and produce a Poisson distribution of team sizes in which larger teams are exceedingly rare. The second mode employs extended teams, which started as core teams, but subsequently accumulated new members proportional to the past productivity of their members. Given time, this mode gives rise to a power-law tail of large teams (10-1000 members), which features in many fields today. Based on this model we construct an analytical functional form that allows the contribution of different modes of authorship to be determined directly from the data and is applicable to any field. The model also offers a solid foundation for studying other social aspects of science, such as productivity and collaboration.
Empirical team-size distributions
Astronomy team-size distributions changed from predominantly Poisson-like small teams in the 1960s to distributions with a power-law tail of very large teams. The model attributes this evolution to Poisson core-team formation combined with cumulative-advantage growth of extended teams.
- Empirical change: Average astronomy team size grew from 1.5 authors in 1961-1965 to 6.7 in 2006-2010, while the distribution also changed shape.The later period includes teams with several hundred authors, unlike the earlier period, which had no teams larger than eight authors.
- Empirical change: 1961-1965 team sizes were well described by a Poisson distribution, consistent with paper production governed by a Poisson process.The Poisson rate λ represents the characteristic number of authors needed for a study.
- Empirical change: Recent distributions contain a power-law tail, indicating dynamics fundamentally different from a simple Poisson process and consistent with cumulative advantage.Cumulative advantage is dynamic because system properties depend on the previous state.
- Small-team structure: 90% of recent astronomy articles still have fewer than ten authors, where the power law breaks down into a small-team hook.A single Poisson distribution cannot fit this region because matching the two-author-to-one-author ratio would shift the predicted peak to five authors.
- Interpretation: The proposed model combines Poisson core-team formation and cumulative-advantage growth to explain two principal modes of knowledge production.The model is intended to answer how earlier Poisson-like distributions evolved into recent power-law-tailed distributions.
Model of team formation and evolution
The model represents each lead author as associated with Poisson-drawn core teams and extended teams that add members according to aggregate productivity. Simulations reproduce astronomy’s evolving team-size distributions and support two authorship modes, while also predicting several author-centric distributions.
- Model structure: The model simulates team formation over time by assigning each lead author a Poisson-sized core team and an expandable extended team.Core-team size is governed by a characteristic rate λ.
- Model structure: Extended teams add members in proportion to the aggregate productivity of current members, creating cumulative advantage in team growth.New members are selected from existing members’ core teams or from a general pool when needed.
- Additional assumptions: “Core +1” teams add one member to a Poisson draw to reproduce the excess of two-author papers, especially among newer authors.The model also includes repeat-publication and retirement assumptions, although retirement is not essential for reproducing the empirical distribution.
- Validation: 154,221 simulated articles reproduce the emergence of the power-law tail, the low-k hook, and the shift toward two- or three-author peaks.The strongest mismatch is a bump near k = 200 in 2006-2010, attributed to several FERMI collaboration papers.
- Validation: The model also predicts productivity, collaborator, and team-per-author distributions, including formation of the giant component in the early 1970s.These predictions extend validation beyond team-size distributions.
- Interpretation: A power-law collaborator tail can reflect authors belonging to cumulatively growing extended teams, rather than only preferential attachment by star scientists.In the single-appearance limiting case, F_C(n) = (n + 1)F(n + 1).
- Interpretation: A constant extended-team publishing propensity of p_ext = 0.3 reproduces the empirical distribution across the 50-year period.The result suggests that extended-team growth, rather than changing propensity alone, made this authorship mode increasingly conspicuous.
- Authorship modes: The model distinguishes core and extended authorship modes, with core teams dominating articles containing fewer than ten authors.Figure 3 also requires standard core and “core +1” teams to reproduce the distribution accurately.
Analytical decomposition of team-size distributions
The paper decomposes team-size distributions into two Poisson core-team components and an exponentially truncated power-law extended-team component, enabling authorship-mode contributions to be estimated from empirical data. Applied across fields and decades, this functional form fits observed distributions and shows that rising mean team size is primarily associated with extended teams.
- Functional form: The analytical decomposition represents standard core and “core +1” teams with Poisson functions and extended teams with a truncated power law.The six-parameter expression assigns separate roles to the two Poisson rates, power-law slope, exponential truncation, and component normalizations.
- Functional form: 57% of recent astronomy articles are attributed to standard core teams, 12% to “core +1” teams, and 31% to extended teams.These shares come from integrating the fitted components for astronomy’s most recent distribution.
- Cross-field application: All examined field distributions are well described by the sum of two Poisson functions and a truncated power law, supporting application of the analytical description across fields.The fields include mathematics, ecology, literature, social psychology, and arXiv; literature has too few points for a KS test.
- Cross-field application: In mathematics, extended teams account for 9% of articles and average 2.9 members versus 1.8 for core teams, while ecology’s increase mainly reflects larger standard core teams.The field-specific trends differ in both the prominence and size of extended teams.
- Evolution over time: Astronomy’s overall mean team-size increase primarily reflects rapidly growing extended teams rather than core teams.Core teams grew linearly from 1.1 to 3.2 members, while extended teams grew exponentially to a recent mean of 11.2 members.
Implications and conclusions
The model concludes that scientific team formation is multimodal, combining small core teams with extended teams that produce the power-law tail. This framework connects team-size distributions to broader questions about collaboration, communication, and research evaluation.
- Conclusions: Team formation is multimodal: core teams produce the distribution’s hook, while expanding extended teams produce its power-law tail.The extended-team mode is associated with research requiring expertise or resources outside the core team.
- Conclusions: The authors state that the model explains the evolution of scientific team sizes as manifested in research-article author lists.The conclusion concerns observable article teams rather than every aspect of research-group organization.
- Broader connection: The hook-and-tail structure of team sizes resembles the two-mode structure proposed for citation distributions.The cited comparison distinguishes direct citations from indirectly accumulated citations subject to cumulative advantage.
- Broader implications: Team-size distributions underpin concepts of scientific collaboration and team science and may inform research evaluation.The paper presents its team-formation principles as potentially useful for studying collaboration and communication.