In the scientific community, citation counts are the primary currency of prestige and impact. However, the paper "Theory of Citing" by M.V. Simkin and V.P. Roychowdhury offers a provocative challenge to this meritocratic view. By applying a forensic analysis to reference lists, the authors suggest that the "greatness" of highly cited papers may be less a reflection of intrinsic merit and more a result of a massive, unintentional game of "telephone" played by researchers who copy citations without reading the original works.
The Problem: Citation Skew and the Matthew Effect
Citation distributions are notoriously skewed: a tiny fraction of papers receive thousands of citations, while the vast majority receive very few. This phenomenon is often attributed to the "Matthew Effect"—the sociological principle that success breeds more success. Simkin and Roychowdhury investigate the structural mechanism behind this effect, specifically looking for evidence of "citation copying"—the practice of lifting a reference from a bibliography without consulting the source paper.
Methodology: Typos as DNA Evidence
The authors employ a clever forensic approach by treating misprints (typos in volume numbers, pages, or years) as "DNA evidence." If two papers cite the same work and share an identical, non-obvious typo, it is highly probable that one copied the citation from the other rather than from the original source.
To quantify this, they developed the Misprint Propagation Model (MPM) and the Random-Citing Scientists (RCS) Model. These models simulate a process where scientists pick a few papers to cite and then copy a fraction of those papers' references. By comparing the distribution of these "mutated" citations to real-world data, the authors can estimate the ratio of "readers" to "citers."
Datasets and Empirical Evidence
The study utilized detailed citation and misprint statistics from the ISI Database (2002-2003) for 12 high-profile physics papers, including seminal works by Nobel laureates like K.G. Wilson and Kosterlitz & Thouless.
Key Results: The 80/20 Rule of Reading
The core finding of the research is the calculation of the reader-to-citer ratio ($R$). Across the various high-profile papers studied, the median value for $R$ was found to be approximately 0.2. This suggests that only about 20% of researchers who cite a paper have actually read it; the remaining 80% are "non-readers" who simply copied the citation from another bibliography.
As shown in the data above, the most popular misprint in one studied paper appeared 78 times, providing clear evidence of a massive chain of copying. The models also correctly predict the power-law exponents (typically around 2.0) observed in real citation distributions.
Insights: Renown by Chance
The RCS model demonstrates that if all papers are assumed to be of equal quality, the act of copying alone will eventually produce "superstar" papers. In a sample of 18,500 papers from Physical Review D, the model predicted that 44 papers would reach 500+ citations purely by mathematical luck; the actual data contained exactly 44 such papers.
This mechanism also explains other phenomena:
- Literature Aging: Older papers eventually stop being cited as newer papers become the primary "templates" for copying.
- Sleeping Beauties: A dormant paper can suddenly explode in popularity if it is randomly cited by a new, popular paper, triggering a fresh avalanche of copying.
Limitations and Conclusion
While the authors acknowledge potential noise from database errors and the difficulty of accounting for authors who repeat their own typos, the quantitative match between their models and empirical data is striking. The study suggests a need to re-evaluate how we measure scientific prestige, noting that even "nonsense" papers can accumulate thousands of citations through these stochastic processes. Ultimately, the "Theory of Citing" reveals that the scientific record is as much a product of social dynamics and random chance as it is of intellectual progress.
Summary updated