Source-linked AI summary
The first-mover advantage in scientific publication
M. E. J. Newman
TL;DR
The paper asks whether preferential-attachment models correctly predict a strong, persistent first-mover advantage in scientific citations. It compares those predictions with citation data across fields and finds substantial support in several cases, while identifying later papers that outperform expectations and cautioning that the evidence is not always clean.
Problem
Preferential-attachment theory predicts that early papers receive much higher citation rates indefinitely, but the paper tests whether this first-mover effect is observed across scientific fields.
Method
The authors derive publication-order citation distributions from the citation model and compare theoretical predictions with empirical citation data from multiple fields.
Results
The predicted first-mover advantage is quantitatively substantiated by empirical citation data in at least some areas, with observed effects sometimes similar in magnitude to theory.
Takeaways & Limitations
A paper’s publication date can account for a large part of citation variation, while a small number of relatively late papers receive substantially more citations than theory predicts.
Takeaways & Limitations
Field comparisons are constrained by incomplete or ambiguous data, uncertain field boundaries and start dates, and citation practices that may not apply to older fields.
Abstract
from arXiv · showhide
Mathematical models of the scientific citation process predict a strong "first-mover" effect under which the first papers in a field will, essentially regardless of content, receive citations at a rate enormously higher than papers published later. Moreover papers are expected to retain this advantage in perpetuity -- they should receive more citations indefinitely, no matter how many other papers are published after them. We test this conjecture against data from a selection of fields and in several cases find a first-mover effect of a magnitude similar to that predicted by the theory. Were we wearing our cynical hat today, we might say that the scientist who wants to become famous is better off -- by a wide margin -- writing a modest paper in next year's hottest field than an outstanding paper in this year's. On the other hand, there are some papers, albeit only a small fraction, that buck the trend and attract significantly more citations than theory predicts despite having relatively late publication dates. We suggest that papers of this kind, though they often receive comparatively few citations overall, are probably worthy of our attention.
The first-mover advantage
The paper models citation accumulation as preferential attachment and derives how citation rates depend on publication order. The model predicts that early papers gain a lasting advantage because their initial citation lead is amplified over time.
- Model assumptions: Each paper cites m previous papers, selected in proportion to their existing citations plus a positive constant r.The constant r allows newly published papers with no citations to receive citations.
- Model solution: The model can be solved exactly for the citation distribution in the large-network limit using a master-equation method.Its citation distribution has a power-law tail, pk ∼ k−α, consistent with empirical citation data.
- Publication-order formulation: Publication order is represented by t = i/n, and a density πk(t) describes the expected fraction of papers receiving k citations at that publication position.This approach avoids the continuous-time approximation while allowing publication rates to vary over actual time.
- The first-mover advantage: The model predicts that average citations become arbitrarily large as t →0, so early papers should receive far more citations than later papers even after accounting for exposure time.The resulting first-mover advantage is expected to persist indefinitely and to concentrate the highly cited tail among the earliest papers.
- Interpretation and qualification: Early papers can receive citations largely regardless of content because they are initially among the few relevant publications, after which preferential attachment amplifies their lead.The authors also note that paper content affects attention, so preferential attachment is not a complete citation model.
Comparison with citation data
Testing is restricted to within-field citation networks, with early-foundation data being difficult to obtain. In the network-theory example, observations closely match the model and show a strong first-mover effect.
- The model cannot represent the entire citation network because most citations occur between papers in the same field.
- The analysis therefore focuses on citations within individual fields, using data that describe a field from its earliest foundation.Such data can be difficult to obtain.
- The network-theory data set covers 2,407 physics and related-area papers published from June 1998 to June 2008.It excludes review articles and includes five early, well-cited papers together with papers citing them.
- Publication order, rather than publication date, is used to measure time when comparing observed and predicted citation rates.
- The data agree well with theory and show a first-mover effect with magnitude and duration similar to the model’s prediction.
- 101 citations versus 26: the first 10% of papers averaged 101 citations each, while the second 10% averaged 26.The most recent 10% averaged 0.08 citations each.
Highly cited papers
The data broadly support a first-mover advantage, but later papers retain a longer citation tail than preferential-attachment theory predicts, including a small set of unusually highly cited papers.
- Early publication dates produce more highly cited papers than theory predicts, while some papers receive substantially more citations than expected.The excess highly cited papers are few enough to have little visible effect on the overall figure but lie well outside the expected range.
- Citation counts normalized by publication-date averages identify highly cited papers across the full ten-year period rather than concentrating them among the earliest papers.The network-theory analysis uses each paper’s z-score above the mean citation count for papers published around the same date.
- A later 2006 paper reached 7.2 standard deviations above its date-adjusted mean despite having only 63 citations.Another later paper reached 6.5σ with 233 citations, showing that raw citation totals can favor older papers.
- The later papers’ citation tail remains relatively long after excluding the earliest papers, contrary to the model’s stronger truncation prediction.The observed tail is diminished, but it is neither well described by the predicted exponential nor consistent with the model’s expected short tail.
- Preferential attachment predicts that later papers should receive very many citations only rarely because most available citations flow to early papers.The model’s later-paper distribution becomes exponentially truncated at a characteristic scale.
Other examples
Additional examples show that first-mover effects vary with whether a literature constitutes a genuinely new field. Strange-matter papers broadly match theory, whereas adult neural stem-cell papers do not; such analyses may test claims about field formation, though data limitations remain.
- Strange matter: A 24-year strange-matter dataset shows strong first-mover advantage and generally agrees with theoretical predictions.A citation bump around 1999–2001 is attributed to presumed scientific developments.
- Interpretation: When a supposed new field is only a branch of an established subject, its earliest papers are cited according to their position within the larger field.Under this condition, the model does not predict a first-mover effect.
- Adult neural stem cells: Adult neural stem-cell papers show no discernible first-mover effect, with citations increasing roughly linearly with paper age.The literature appears to form part of a larger neural-stem-cell community rather than a new field.
- Interpretation: Citation analyses could independently test whether a publication or discovery created a new scientific field, but the data do not always support firm conclusions.Field boundaries, completeness, starting dates, and historical citation practices can complicate analysis.
Conclusions
The paper concludes that theories predicting a strong first-mover advantage are quantitatively supported by empirical citation data in at least some areas.
- Conclusions: Empirical citation data quantitatively substantiate the strong first-mover advantage predicted by citation-process theories.The conclusion is limited to at least some scientific areas.
- Conclusions: The observed evidence supports a strong first-mover advantage in at least some areas.
- Conclusions: The paper’s conclusion concerns empirical citation data and theories of the scientific citation process.