The History of Research Papers
How the research paper went from private letters between natural philosophers to the dominant format of scientific knowledge — and what got lost and gained along the way.
Before the Paper: Letters and Books
For most of recorded history, scientific knowledge moved through two channels: books and letters. Books were slow and expensive — the format for established, synthesised knowledge, not live discovery. Letters were fast and personal — the format for “I just found something, tell me what you think.”
The Republic of Letters was the informal network of natural philosophers, mathematicians, and scholars across 16th and 17th century Europe who kept each other informed through correspondence. Galileo, Descartes, Newton, Leibniz, Huygens — they wrote to each other constantly. Discoveries were announced, debated, and sometimes stolen through the post. Priority disputes (who found it first) were vicious because there was no formal mechanism for claiming credit. Newton and Leibniz spent years in a bitter priority war over calculus that was essentially a failure of the informal system.
1665: The First Journals
Two journals appeared in 1665 and changed everything.
Journal des sçavans launched in January in Paris — technically the first scientific periodical, though it covered a broad mix of natural philosophy, law, literature, and church affairs. More newsletter than research journal.
Philosophical Transactions of the Royal Society launched in March in London, edited by Henry Oldenburg. This one mattered more. Oldenburg’s innovation was to solicit, collect, and publish accounts of experiments and observations from Fellows and correspondents — not synthesised knowledge, but reports of specific things done and found. The format was essentially the published letter: informal, first-person, addressed to Oldenburg and through him to the scientific community.
The journal solved the priority problem. Publication date established when you found something. It also solved the dissemination problem — one letter to many readers, rather than one letter to one reader who might or might not pass it on.
The Structure Emerges Slowly
The modern paper’s structure — Introduction, Methods, Results, Discussion — did not appear in 1665. Early papers were narratives. A natural philosopher described what they did, in the order they did it, with digressions and speculations freely mixed in. Newton’s papers on optics read more like letters than like anything in a modern journal.
The shift toward formalised structure happened gradually through the 18th and 19th centuries as disciplines professionalised and journals multiplied. Chemistry led the way — experimental reports needed to be reproducible, which required precise methods description. By the mid-19th century, the structure of the experimental report was fairly settled in the natural sciences: here is what question I asked, here is what I did, here is what I observed, here is what I think it means.
The IMRAD format (Introduction, Methods, Results, and Discussion) became the explicit standard only in the 20th century. The American National Standards Institute formalised it in 1972. Most scientists had been approximating it for decades before that.
Peer Review: Not as Old as You Think
The assumption that peer review is an ancient and fundamental institution of science is wrong. Peer review as a standard requirement for publication is largely a post-World War II phenomenon.
Early journals did have editorial filtering — Oldenburg at Philosophical Transactions exercised judgement about what to publish — but this was editorial discretion, not systematic review by independent experts. The Royal Society itself occasionally circulated manuscripts to knowledgeable Fellows, but this was informal and inconsistent.
Nature, founded in 1869, did not require peer review for most of its early history. Papers were often accepted or rejected by the editor alone, sometimes on the basis of the author’s reputation.
The shift to systematic peer review happened through the 1930s–1960s as the volume of submissions grew past what any single editor could evaluate, and as funding bodies began requiring external review for grants (the NIH instituted study sections in the 1940s). By the 1970s, double-blind peer review was standard across most major journals.
The irony is that peer review became universal at roughly the same time that the critique of peer review became serious. The same volume that made editorial review inadequate also made peer review slow, inconsistent, and gameable by entrenched researchers protecting their fields from inconvenient findings.
The 20th Century: Big Science and the Journal System
The post-war expansion of science funding — primarily in the US through the NIH, NSF, and DOD — produced an explosion of research and researchers. The number of journals multiplied. Commercial publishers, notably Elsevier (founded 1880, but expanding aggressively from the 1950s), Robert Maxwell’s Pergamon Press, and Springer, recognised that journals were a profitable business: the content was provided free by researchers, reviewed free by researchers, and then sold back to universities at rapidly increasing subscription prices.
By the 1980s and 1990s, the “serials crisis” was acute. Journal subscription costs were rising far faster than library budgets. Universities were cancelling subscriptions. Researchers were publishing in journals that their own institutions couldn’t afford to read. The system that was supposed to disseminate knowledge was increasingly restricting it behind paywalls that only wealthy institutions could afford.
arXiv and the Preprint Revolution
The response from physics was characteristically direct: in 1991, Paul Ginsparg at Los Alamos National Laboratory launched an email-based preprint server. Researchers uploaded papers before (or instead of) journal submission. Anyone could read them for free.
This was arXiv. It was initially for high-energy physics, where preprint culture was already established — physicists had been mailing each other paper preprints for decades. The server just made it global and instant.
arXiv expanded to mathematics, computer science, quantitative biology, and quantitative finance. In machine learning, it became the primary venue — papers go up on arXiv the day they’re written, often before any peer review, and the community treats the arXiv timestamp as the priority date. The official journal publication, when it happens, is almost an afterthought.
The preprint model revealed something uncomfortable about the peer-reviewed journal system: in fast-moving fields, the value of peer review was often negative. The review process added months or years of delay while providing inconsistent quality filtering that the community was already performing informally by reading and discussing preprints.
Open Access: The Policy Response
The open access movement formalised the critique of the journal system into a policy agenda. The Budapest Open Access Initiative (2002) defined open access and called for it to become the default for publicly funded research. The argument was simple: if taxpayers fund the research, they should be able to read the results.
Two main routes emerged:
Gold open access: publish in an open access journal (PLOS ONE, founded 2006, was an early large-scale example). The journal is free to read; costs are recovered by charging authors an Article Processing Charge (APC). This shifted the financial burden from readers to authors — and created a new problem: journals with low standards willing to publish anything for the APC fee (“predatory journals”).
Green open access: publish in any journal, but deposit a copy (preprint or accepted manuscript) in an open repository. Most major journals now permit this. PubMed Central, arXiv, and institutional repositories are the main destinations.
Funding bodies began mandating open access: the NIH in 2008, the Wellcome Trust, the Gates Foundation, the European Research Council. The UK Research Councils went further, requiring immediate open access with no embargo. In 2022, the US Office of Science and Technology Policy directed all federal agencies to require open access for federally funded research by 2026.
The commercial publishers adapted by acquiring open access journals, launching hybrid journals (pay to make your article open in an otherwise paywalled journal), and raising APCs. The serials crisis mutated rather than resolving.
The Reproducibility Crisis
In 2011, Bayer healthcare published a review finding that they could reproduce only about 25% of published preclinical studies relevant to their drug development programs. In 2012, Amgen published similar findings — they could reproduce 6 of 53 “landmark” cancer biology papers. In 2015, the Open Science Collaboration published a systematic replication attempt of 100 psychology studies: roughly 40% replicated at the original effect size and significance level.
The reproducibility crisis — the finding that a substantial fraction of published scientific results do not replicate — exposed structural problems in how the research paper system had evolved. Publication bias (journals preferring positive results) meant the literature systematically overrepresented findings that were either false positives or inflated. P-hacking and HARKing (Hypothesising After Results are Known) were common enough to distort fields. Incentive structures rewarded novelty and volume over rigour and replication.
The paper as a unit of scientific communication had developed a serious flaw: it was optimised for persuasion, not verification. A paper makes the best case for a finding — it does not show the failed experiments, the analytic decisions that were tried and discarded, the data that didn’t fit.
Responses have included pre-registration (publishing your hypothesis and analysis plan before collecting data), registered reports (journals committing to publish results regardless of outcome before the study is run), and open data/open materials requirements. Progress is real but slow.
Where It Stands
The research paper in 2025 is an institution under pressure from multiple directions simultaneously. Preprints have decoupled reading from publication. Large language models can summarise and synthesise literature faster than any human reader. Open access mandates are dismantling the subscription model. Reproducibility requirements are changing what a paper must contain.
What has not changed is the basic function: a paper is a public, citable, archivable record of a specific claim and the evidence for it. That function — making findings checkable and attributable — is what the Royal Society was trying to solve in 1665. The format has evolved considerably. The problem it solves has not.