TL;DR
How do arbitrary social conventions—like community jargon or coding styles—win out over their competitors? By analyzing 24 years of arXiv LaTeX source files, this paper reveals that the success of a convention depends heavily on the experience of its adopters and its public visibility. Crucially, the authors find a striking division of labor in collaborations: younger researchers win "invisible" fights (like macro names), while senior researchers dictate highly "visible" choices (like paper titles).
What Problem Does This Paper Address?
Conventions constantly compete for dominance in social and technical ecosystems. However, studying how one convention triumphs over another is notoriously difficult due to two main challenges:
- Substantive Shift: As conventions evolve, their underlying meaning often changes too. For example, if a new political slogan catches on, it might be because it carries a slightly different, more appealing message, confounding purely social network effects. To isolate social dynamics, researchers need neutral variation—instances where the meaning stays exactly the same, but the form changes.
- Local vs. Global Dynamics: Most diffusion models look at global trends (how an idea spreads across a whole population), but conventions are often negotiated locally, face-to-face, when individuals interact or collaborate.
The authors tackle these challenges by finding a perfect digital laboratory: user-defined LaTeX macros in scientific papers.
Methods
In LaTeX, a macro consists of a name (the convention) and a body (the exact code/meaning). For example, \def\Reals{\mathbb{R}} and \def\R{\mathbb{R}} have different names but identical bodies. This provides a perfect control for meaning.
The authors define two main phenomena to track competition:
- Changeovers: Global shifts where one macro name replaces another for the same body over time.
- Fights: Local interactions where two authors who previously used different names for the same macro body collaborate on a new two-authored paper. The name that appears in the joint paper is the "winner."
The authors use logistic regression to predict the outcomes of both global changeovers and local fights. To test the effect of visibility, they extend this "fight" framework from invisible macro names to highly visible paper title styles (e.g., use of colons or question marks) and low-visibility macro bodies (e.g., choosing between \epsilon and \varepsilon).
Datasets
The dataset comes from a complete snapshot of source files from the e-print arXiv, spanning 24 years (July 1991 to September 2015).
From over a million papers, they extracted macros from the subset written in LaTeX, resulting in:
- 583,078 papers containing at least one macro.
- 22,628,300 total macro definitions.
- 2,586,548 unique macro bodies.
- 222,689 unique authors.
Key Insights and Ideas
- The Name Doesn't Matter, the Network Does: Orthographic features of a macro name (length, capitalization) have zero predictive power over whether it will become dominant. Success is driven entirely by social structure and adopter characteristics.
- The Generational Hand-off: Successful conventions typically start among "younger" (less experienced) authors and successfully transition to "older" (more experienced) authors. If a convention stays trapped among less experienced authors, or starts entirely with senior authors, it rarely achieves global dominance.
- The Visibility Caricature: There is a clear power dynamic based on how visible a choice is to the public. In a collaboration, younger researchers win the invisible fights (macro names), while older researchers win the visible ones (paper titles). Senior authors willingly cede control over low-level implementation details but assert authority over public-facing presentation.
Key Results
- Global Changeover Prediction: Using only the experience levels of early adopters, the authors predicted whether a macro would undergo a global changeover with 60% accuracy (compared to a 50% random baseline). Name features yielded exactly 50%.
- Local Macro Fights: Younger authors dominate macro name fights. A model using just the experience of the two authors predicted the winner with 58.2% accuracy, which improved to 67.3% when adding network features (like betweenness centrality) and author flexibility.
- The Visibility Reversal: As the experience gap between co-authors grows, the younger author becomes increasingly likely to win invisible macro name fights. Conversely, in highly visible title style fights, the older author wins 57% of the time, and their win rate increases as the experience gap widens.
Limitations and Future Work
- Domain Specificity: The study is restricted to the arXiv, a community with unique collaboration norms and demographics. How well these findings generalize to other collaborative environments, like open-source software development or corporate settings, remains to be seen.
- Co-authorship Proxy: The paper treats the final submitted LaTeX file as the ground truth for a "fight," but cannot observe the actual iterative editing process or conversations that led to those choices.
- Future Directions: The authors suggest applying this methodology to large source-code repositories (like GitHub) where naming conventions change while code behavior remains constant, and building formal mathematical models of convention evolution based on biological evolutionary theories.
Summary updated