Source-linked AI summary
Tradition and Innovation in Scientists' Research Strategies
Jacob G. Foster, Andrey Rzhetsky, James A. Evans
TL;DR
The paper asks how scientists balance productivity with risky innovation when choosing research problems. It represents biomedical knowledge as evolving chemical networks and defines research strategies relative to those networks. Strategy frequencies remain stable despite substantial knowledge growth, while risky strategies are rarer, more variable, and associated with both greater failure risk and greater potential recognition.
Problem
The paper addresses limited quantitative evidence about how prevalent risky versus conservative research strategies are and how their risks and rewards compare.
Method
The authors construct evolving chemical networks from 6,455,756 MEDLINE abstracts and classify research projects according to novel chemicals, chemical relationships, and network structure.
Results
Risky strategies are less prevalent but receive more citations on average, show greater citation variability, and are enriched among highly cited articles and prize-winning research.
Takeaways & Limitations
The findings provide quantitative evidence for an essential tension between conservative productivity and risky innovation, with occasional high-impact gambles helping explain observed risk-taking.
Takeaways & Limitations
Chemical annotation is not uniform over time, and the method misses relationships whose chemicals are absent from NLM annotations.
Abstract
from arXiv · showhide
What factors affect a scientist's choice of research problem? Qualitative research in the history, philosophy, and sociology of science suggests that this choice is shaped by an "essential tension" between the professional demand for productivity and a conflicting drive toward risky innovation. We examine this tension empirically in the context of biomedical chemistry. We use complex networks to represent the evolving state of scientific knowledge, as expressed in publications. We then define research strategies relative to these networks. Scientists can introduce novel chemicals or chemical relationships--or delve deeper into known ones. They can consolidate existing knowledge clusters, or bridge distant ones. Analyzing such choices in aggregate, we find that the distribution of strategies remains remarkably stable, even as chemical knowledge grows dramatically. High-risk strategies, which explore new chemical relationships, are less prevalent in the literature, reflecting a growing focus on established knowledge at the expense of new opportunities. Research following a risky strategy is more likely to be ignored but also more likely to achieve high impact and recognition. While the outcome of a risky strategy has a higher expected reward than the outcome of a conservative strategy, the additional reward is insufficient to compensate for the additional risk. By studying the winners of 137 different prizes in biomedicine and chemistry, we show that the occasional "gamble" for extraordinary impact is the most plausible explanation for observed levels of risk-taking. Our empirical demonstration and unpacking of the "essential tension" suggests policy interventions that may foster more innovative research.
Introduction
Scientists choose research problems under competing pressures: reliable productivity favors incremental work, while high standing favors risky originality. The paper frames these demands as an essential tension between tradition and innovation and studies its prevalence and rewards in biomedical chemistry.
- Career advancement favors reliable productivity through incremental contributions, whereas high scientific standing favors important, original contributions through risky new directions.
- This conflict creates an essential tension between productive tradition and risky innovation in research-problem selection.
- Unexpected findings are associated with important scientific contributions, but established methods can encourage their neglect.
- The study quantitatively examines how prevalent strategies yielding unexpected findings are and what risks and rewards they carry in biomedical chemistry.
- The authors measure five broad strategies for selecting chemical relationships, pool articles by publication year, and compare their aggregate frequencies with available opportunities.
- A content-based approach treats chemicals and their relationships as fundamental units, representing research possibilities as a network of scientific knowledge.
Results and Discussion
Biomedical chemistry research strategies remained remarkably stable despite dramatic growth in available chemical relationships. Riskier strategies were rarer and less predictable, but attracted more citations when successful and appeared more often in highly recognized work.
- Network and measurement: 6,455,756 MEDLINE abstracts produced an evolving chemical network that reached 181,078 nodes and 84,709,977 links by 2008.The analysis focused on 1983–2008 to limit effects from the introduction of chemical indexing.
- Prevalence and stability: Repeat statements were six times more frequent than new or jump statements, comprising 85.8% versus 14.2% of observed strategies.New bridges and new consolidations were more common than jumps, at 12.4% versus 1.8%.
- Prevalence and stability: From 1983–2008, the relative frequency of “new” strategies changed little, with jumps rarest, followed by consolidations and bridges.The stable distribution persisted while the number of distinct chemicals increased by an order of magnitude and potential new links expanded substantially.
- Limitations: Non-incremental strategies may be easier to publish in high-prestige outlets when successful, whereas incremental research is relatively easy to publish in lower-impact journals.This publication-selection possibility qualifies interpretations of strategy prevalence in the literature.
- Stability and attention: Pearson’s R = 0.983 indicates that the strategy-selection model predicts observed behavior well, while parameter trends indicate increasing local focus and filtering of new opportunities.The model represents scientists as an abstract agent choosing strategies based on available links and inferred aggregate bias parameters.
- Risk and rewards: A strategy ten times less probable received 2.26 additional citations on average, while surprisal explained 29.2% of variation in mean citations.Surprisal also explained 28.6% of variation in citation standard deviation, indicating that rare strategies had less predictable impact.
- Risk and rewards: Scientists engaged more risk than citation maximization alone would predict, suggesting that the possibility of exceptional impact and recognition helps explain observed risk-taking.The analysis bounds failure probabilities under the assumption that scientists choose strategies only to maximize citations, then compares this with observed behavior.
- Risk and awards: Top-cited articles and prize winners used novel strategies more often, including authors associated with 137 biomedical and chemistry prizes.Elite award winners more frequently introduced new chemicals and new relationships within knowledge clusters.
Conclusions
The study finds quantitative evidence of an essential tension between conservative productivity and risky innovation, while identifying limitations in its chemical-network analysis. It suggests that risky research can produce extraordinary impact but is insufficiently rewarded on average, motivating policy measures to encourage innovation.
- The study finds that scientists’ research choices reflect an essential tension between conservative productivity and risky innovation.
- The analysis is limited by nonuniform chemical annotation, missed unannotated relationships, false co-mention relationships, a coarse taxonomy, and exclusion of unpublished failures.
- Risky research is a gamble whose additional citation payoff is insufficient, on average, to justify its additional risk.
- Linking distant islands of knowledge remains fruitful for science, despite strategic preferences favoring established knowledge.
- Two proposed policy levers are decoupling early job security from productivity and funding scientists or risky projects more aggressively.
Materials and Methods
The study models biomedical research-strategy choices as weighted selections among five network-based possibilities, then estimates parameters and tests how strategy surprisal relates to citations. It also assembles prize-winner publication records to compare recognized scientists with the broader scientific population.
- Strategy-selection model: Five strategies are modeled: jump, new consolidation, new bridge, repeat consolidation, and repeat bridge.The model treats each strategic choice as an independent draw, although article-level choices may be correlated.
- Strategy-selection model: Researchers first choose among jump, new, or repeat, then choose between consolidation and bridge.This factorization defines the probabilities of the five possible events.
- Parameter estimation: Model parameters are estimated by maximum likelihood using the observed counts of each strategy over the six previous publication years.The parameters represent biases for new relationships, repeats, new consolidation, and repeat consolidation; jump bias is fixed at 1.
- Citation analysis: Citation analyses assign each article its citations during the three years after publication and use heteroscedasticity-corrected regressions for 1983–2002.Approximately two-thirds of MEDLINE articles link to citation records; post-2005 coverage is reduced.
- Prize-winner sample: The prize-winner sample comprises 137 biomedicine and chemistry awards, with winner publications identified through MEDLINE author-cluster searches extending up to 30 years before each award.Name disambiguation can create false positives and underrepresent winners with non-English characters.