Introduction
In the academic world, the order of authors on a research paper serves as the primary currency for professional credit. This ordering significantly influences career milestones, from tenure track appointments to prestigious awards. However, conventions vary drastically across disciplines. Mathematics and theoretical computer science traditionally favor alphabetical ordering, implying equal contribution or following a strict convention. In contrast, fields like biology and psychology utilize contribution-based ordering, where the first author is typically credited with the lion's share of the work.
This research investigates whether these choices are merely cosmetic or if they fundamentally alter the behavior of scientists. By employing a game-theoretic framework, the study explores how authorship norms influence research quality, collaboration patterns, and the prevalence of "free riding."
The Academic Game Model
The authors propose an "Academic Game," a two-player game-theoretic model to simulate research collaboration. In this setup, agents possess a budget of "effort," represented by weights. They can choose to work independently or collaborate.
The value of a project is defined by a convex reward function ($x^d$, where $d > 1$), implying that collaborative effort generates synergy—the total value exceeds the sum of individual contributions. Credit is then allocated according to a fixed vector $[\phi, 1-\phi]$. Under an alphabetical scheme, $\phi = 0.5$, representing an equal split. In contribution-based schemes, $\phi > 0.5$, granting the primary author more credit. The model uses the concept of pairwise stability to predict which collaborations will form and endure.
Empirical Justification
To ground the theoretical model, the authors conducted an empirical analysis of authorship patterns in major Artificial Intelligence conferences between 2013 and 2015.
As shown in the data above, approximately 30% of papers in top-tier venues like IJCAI, AAAI, and NIPS are authored by only one or two individuals. This empirical finding justifies the study's initial focus on a two-player model as a representative and meaningful abstraction of the broader academic ecosystem.
Key Insights: Quality vs. Quantity
The model reveals distinct behavioral incentives under different ordering schemes:
- Effort Matching in Alphabetical Systems: Because alphabetical ordering mandates a 50/50 credit split, it incentivizes authors to match high effort. If one partner commits significant resources, the other must do the same to maximize their own utility. This dynamic fosters the completion of "hard" projects requiring peak effort from all parties.
- "Cheap Talk" in Contribution-Based Systems: Conversely, contribution-based ordering allows for "cross-layer" collaborations. A high-effort researcher might pair with a "light" contributor (providing minimal effort or "cheap talk") because the lead researcher can still claim most of the credit. While this leads to a higher volume of papers and a denser collaboration network, it may result in lower average quality per paper.
- The Free Riding Paradox: Contrary to popular belief that alphabetical ordering invites free riding, the authors demonstrate that steep contribution-based schemes can actually result in a higher "free riding index." When the credit split is heavily skewed, the incentive for the secondary author to contribute meaningfully diminishes significantly.
Stability and Fairness
A critical finding involves the stability of these systems. Under alphabetical ordering, a pairwise stable collaboration structure is guaranteed to exist and can be identified in polynomial time. In contrast, contribution-based systems may suffer from the "stable roommates" problem, where a stable state might not exist at all.
Furthermore, the authors suggest that "rotations"—where frequent collaborators alternate first authorship—can resolve fairness issues. This practice allows teams to achieve high research quality and perfect fairness regardless of the community's official ordering convention.
Limitations and Future Directions
While the two-player model provides foundational insights, it has limitations. Modern "Big Science" often involves massive author lists, and the dynamics of credit in large groups require further study. Additionally, the model assumes a fixed perception of credit and treats effort as a single dimension, ignoring the value of heterogeneous skill sets (e.g., theory vs. implementation). Future research could also explore the impact of diminishing returns (concave reward functions) on large-scale collaborative projects.
Summary updated