Why peer review is broken — and what's replacing it

By The Arcaneum

Category: Academic Commentary

Tags: peer review, scientific publishing, academic journals, open science, research integrity, retraction crisis

In 2023, more than 13,000 scientific papers were retracted. That number — reported by Ivan Oransky, co-founder of Retraction Watch, the nonprofit that monitors retractions across journals — is not a rounding error or a statistical artifact. It is roughly equal to the total number of papers retracted in the entire decade before it. Something has gone seriously wrong with the mechanism that is supposed to be the immune system of science. Peer review is the process by which experts evaluate research manuscripts before they are published in academic journals. It has been the dominant quality-control system in science since roughly the mid-twentieth century. Most scientists treat it as practically synonymous with trustworthiness: a peer-reviewed paper is, in the common shorthand, a real finding. A preprint that hasn't been reviewed is, by implication, provisional and suspect. That distinction is under enormous pressure right now. Not because science is becoming less rigorous — the evidence, actually, points the other way — but because peer review itself is failing to do the work that was always claimed on its behalf. The problem is structural, economic, and almost certainly irreversible in its current form.

The Numbers Don't Lie, But Journals Often Do

Start with the retractions. As of August 2025, Retraction Watch's database contains more than 55,000 unique retracted papers. That figure comes from a study published on arXiv analyzing the Retraction Watch corpus alongside Web of Science data. Across 42 million articles published between 2000 and 2024, the overall retraction rate is approximately 11 per 10,000 papers — but that average masks wild variation between fields. In electrical engineering and computer science, the rate reaches nearly 32 per 10,000, ten times higher than in physics. Nature — arguably the most prestigious journal in the world — has retracted 32 papers since 2020. Science has retracted 20 since 2019. These are not obscure journals where editorial oversight is minimal. They are the flagships. The reasons for retraction are also shifting. A 2025 study in Accountability in Research found that retractions due to data problems have increased significantly since 2000, with the proportion in 2023 exceeding 75%. These include data fabrication, manipulation of images, and what researchers politely call "discrepancies in data availability" — a phrase that often means the underlying data simply doesn't exist or can't be found. The most alarming driver, though, is industrialized fraud. In 2022, 34 journals had more than 20 retractions in a single calendar year. In 2014, only two did. The culprit in many cases is what researchers call "paper mills" — factories, often operating as businesses, that produce fabricated research for sale. Scientists in countries with strong publish-or-perish pressures pay for authorship on papers they didn't write, and journals with weak reviewing infrastructure let them through. According to the arXiv analysis, over 6,400 retractions between 2024 and 2025 were attributed specifically to fake peer reviews.

Peer Review Was Always a Stopgap

Before we condemn peer review as uniquely defective, it's worth noting what it was designed to do — and what it never was. The modern peer review system took its current form largely in the second half of the twentieth century. It was a reasonable solution to a real problem: as the volume of scientific research grew, journals needed some mechanism to filter quality and provide editorial credibility. The solution they landed on was to ask other scientists to evaluate submissions, usually anonymously, usually without pay. That unpaid part is important. Peer review is essentially a gift economy embedded inside a for-profit publishing industry — and it has generated one of the most extraordinary financial arrangements in modern commerce. The four largest academic publishers — Elsevier, Springer Nature, Wiley, and Taylor & Francis — generated more than $7.1 billion in combined revenue in 2024, according to a paper published on arXiv examining publisher profit structures. Their profit margins exceed 30%. Elsevier alone reported €3.26 billion in revenue in 2022 with a profit margin of 37.8% — higher than Apple, Google, or Microsoft in the same period. The mechanism that makes this possible is straightforward to the point of absurdity: governments fund research through grants. Researchers conduct that research, then write papers about it. They submit those papers to journals. Other researchers review the papers for free, at the journals' request. The journals publish the papers. Universities then pay the journals for access to those papers — papers whose underlying research was already paid for by the public. In some cases, researchers also pay article processing charges (APCs) to publish, which can reach $11,690 for a single paper in Nature Neuroscience. In September 2024, a class-action lawsuit filed against major publishers by a UCLA neuroscientist characterized this arrangement as a "cartel" that has "formed to fix the price of peer review labor at zero." Wiley called the suit without merit. The financial numbers, however, are not in dispute.

The Reviewer Crisis

The gift economy is breaking down. Acceptance rates for peer review invitations have been falling for years. A 2025 study tracking reviewer engagement at the journal PRiMER from 2017 to 2025 found that invitation acceptance peaked at 56.14% in 2020, then fell to 35.71% in 2024 — the lowest since the journal was founded. A separate study of Biological Invasions over 21 years found acceptance rates declining steadily across career stages, with senior researchers accepting only 42.4% of invitations by the end of the study period. Journals have adapted by simply sending more invitations. A 2024 editorial in Acta Odontologica Scandinavica reported that the average number of invited reviewers per submitted manuscript at that journal rose from 2–3 in 2006–2010 to 6–7 in 2019–2024. It is not uncommon, the editor noted, to invite more than 20 researchers to secure 2 who will accept. The reasons researchers decline are documented in survey data. According to figures collected by HighWire Press, over 70% of scholars turn down review invitations because the paper doesn't align with their specialization. Forty-two percent cite being overwhelmed with other commitments. Thirty-nine percent say they lack formal training in peer reviewing. That last figure is worth pausing on. Peer review is a skill — the ability to identify methodological errors, evaluate statistical analyses, detect missing controls, assess whether conclusions are supported by evidence — and it is almost entirely untaught. Doctoral students learn to conduct research; they are almost never formally trained to review it. The entire quality-control system of science rests on a skill that most practitioners acquired informally, if at all.

The Delay Problem

Even when peer review works, it's slow. A 2025 study in Frontiers in Research Metrics and Analytics reported that across 57 health policy journals, submission-to-publication times ranged from 35 days to 353 days, with the peer review phase alone extending to 314 days in some cases. In primary care research, the average submission-to-publication lag is 243 days — over eight months. For the social sciences, arts, and humanities, delays are even worse. A Scopus-based analysis of 2,700 papers across 135 journals found that business and economics papers had an average publishing delay of 18 months from submission, twice the delay experienced in chemistry. These are not merely inconveniences. For early-career researchers, whose hiring, promotion, and grant eligibility depend on their publication records, delays of six to eighteen months can determine whether a career continues. The Frontiers paper that opened with an Italian infectious diseases researcher waiting 380 days just for her paper to be assigned to an editor is not extreme — it is a recognizable description of ordinary academic life. And time delays compound with another structural problem: the one-journal-at-a-time rule. Under current norms enforced by most publishers, researchers must submit to a single journal and wait for rejection before submitting elsewhere. This practice, challenged in the 2024 antitrust lawsuit as anti-competitive, means a paper that would be immediately accepted somewhere can spend years bouncing through sequential rejections. In law and medicine, simultaneous submission is standard. In most of science, it remains forbidden.

What Peer Review Actually Catches

Here is perhaps the most uncomfortable question: does peer review work? The data on this is genuinely mixed, but several findings are troubling. A 2019 study in the Journal of Informetrics found that only 8.1% of peer reviews resulted in a suggestion for rejection — meaning the vast majority of reviews that come back as "revise and resubmit" or "accept" do not fundamentally challenge the work's validity. Peer reviewers, the study found, were particularly poor at identifying plagiarism. The replication crisis — most visible in psychology, where large-scale efforts to reproduce published findings have failed to replicate 50–65% of results — suggests that much of what passes peer review is nonetheless unreliable. This isn't entirely the reviewers' fault: they typically see only a paper, not the underlying data, code, or materials. They can evaluate whether a paper reads coherently, not whether its results are real. There is also the problem of groupthink. Peer review is, by definition, conducted by the people who already work in a field. Novel, paradigm-challenging work is evaluated by the very people whose work it challenges. The history of science is well-stocked with papers that were rejected by peer reviewers because they were too different — and then turned out to be correct. The mechanisms of stomach ulcers, the theory of continental drift, and the idea that the universe's expansion is accelerating all encountered resistance from reviewers before eventual acceptance. This is not an argument against expert evaluation; it is an argument against treating expert consensus as equivalent to truth.

What's Actually Replacing It

The failure of peer review has produced a proliferation of alternatives, some incremental, some genuinely structural. Preprint servers are the most significant development. arXiv, launched in 1991 at Los Alamos National Laboratory and now hosted by Cornell University, hosts over 150,000 new submissions per year in physics, mathematics, and computer science. Physicists have not considered preprint posting a second-class activity for over thirty years. bioRxiv, launched by Cold Spring Harbor Laboratory in 2013 for biology, posted 43,629 preprints in 2024 alone — above its peak during the COVID-19 pandemic. Together, bioRxiv and medRxiv now host over 335,000 preprints. The COVID-19 pandemic accelerated this adoption dramatically. When the world needed to understand a fast-moving outbreak, the months-long peer review process was simply too slow. Preprints provided early data on transmission, vaccines, and treatments in days rather than months. The experience changed the culture, at least in biomedicine. Still, only about 15% of all journal articles currently begin life as preprints — suggesting most fields remain attached to the traditional submission-and-wait model. Overlay journals are a more radical experiment. Instead of hosting papers, they conduct peer review of preprints already posted to arXiv or other servers, then "publish" those preprints with a stamp of approval that links to the reviewed version. Discrete Analysis, a mathematics overlay journal launched in 2015, operates with minimal cost and no subscription fees. The preprint server provides the infrastructure; the journal provides the certification. If this model scaled, it would essentially sever the financial relationship between peer review and commercial publishers — since the review function would no longer require the journal's hosting and distribution infrastructure. Open peer review removes reviewer anonymity, requiring reviewers to sign their reports. The argument for this is accountability: reviewers who know their names will appear alongside their reviews are less likely to be dismissive, more likely to engage carefully, and less able to use the review process to suppress competition. Several journals, including those published by eLife and PLOS Medicine, have moved in this direction. The argument against it — that junior researchers will be reluctant to criticize senior ones — is real, but it applies to the current system too; it just happens invisibly. Review Commons is a service that decouples peer review from individual journal submission. Authors submit their manuscript to Review Commons, which organizes review by a panel of experts. That reviewed manuscript can then be submitted to multiple journals simultaneously — the reviews travel with it, avoiding redundant reviewing of the same paper by different sets of referees. This addresses both the reviewer shortage (papers aren't reviewed multiple times) and the sequential submission problem. Post-publication review platforms like PubPeer allow scientists to comment on published papers after the fact, creating a kind of ongoing public scrutiny that continues beyond the initial review. PubPeer has been instrumental in identifying several major fraud cases — the work that led to the resignation of Stanford's president Marc Tessier-Lavigne in 2023 was initially flagged by post-publication analysis.

The Structural Problem These Solutions Don't Fully Address

None of these alternatives fully resolve the core tension: science needs both rapid dissemination and quality verification, and the current system is too slow for the former and too unreliable for the latter. Preprints solve the speed problem but shift quality assessment entirely to the reader — which works in physics, where most readers are specialists who can evaluate a paper's methodology, but is more difficult in fields with broader, less technically trained audiences. Posting a preprint about a drug's effectiveness or a climate model's predictions to the public without any expert verification creates real risks. Open peer review may improve accountability without improving expertise or speed. Review Commons helps with efficiency but doesn't solve the fact that peer review is unpaid labor extracted from scientists who are already overextended. The economic structure of academic publishing is the root cause of several of these problems, and it remains almost entirely untouched by the alternatives emerging within the system. As long as commercial publishers collect profit margins approaching 40% from a process that costs them little — because the labor is donated and the funding is public — the incentives that produce slow, gatekept, commercially motivated publishing will persist. A 2025 paper published on arXiv concluded bluntly that funding agencies "hold all the cards" and that meaningful change will require mandates from institutions that control research money, not voluntary reform from within the publishing industry. "We've tried that for 30 years," one of the paper's authors told Phys.org, "and it hasn't worked."

What This Means for the Rest of Us

The peer review crisis is usually framed as an internal problem for scientists — a matter of reviewer fatigue and publication delays and retraction scandals that normal people don't need to track. That framing is wrong. Peer-reviewed journals are the primary way that scientific findings enter public discourse, inform policy, guide medical practice, and shape education. When that filtering system fails — when fabricated data passes review, when sound but contrarian findings are blocked, when the delay between discovery and publication is so long that the facts on the ground have changed — the consequences extend far beyond academia. The 13,000 retractions in 2023 didn't happen in a vacuum. Many of those papers were cited in the months or years before retraction. Some were acted on. The distance between a retracted paper and a changed treatment protocol or a misled policy decision is shorter than anyone would like to admit. Peer review is not broken because scientists are dishonest or lazy. Most aren't. It is broken because a process designed for a world of modest submission volumes, small academic communities, and no financial pressure has been scaled, commercialized, and stretched far beyond what it was ever designed to handle. The scientists trying to make it work are doing so for free, in their spare time, inside a system that profits enormously from their labor. The alternatives are real, promising, and already functioning at scale in some fields. Whether they become the norm — or remain the exception — will depend less on whether they work than on whether the institutions powerful enough to mandate change are willing to do so.

Sources and Further Reading

The data in this article is drawn from the following sources:

Retraction Watch (retractionwatch.com) — the primary database for tracking scientific retractions; the 13,000 figure for 2023 was cited by co-founder Ivan Oransky in public remarks. "Prevalence and Trends in Global Retractions Explored Through a Topic Lens", arXiv, November 2025 — source for the 55,000 retracted papers figure, the 11.04 per 10,000 retraction rate, and discipline-specific rates. "Analysis of Scientific Paper Retractions Due to Data Problems", Accountability in Research, July 2025 — source for the 75% data-related retraction figure in 2023. "How Ten Publishers Retract Research", arXiv preprint — source for the 46,087 retractions across ten publishers analysis. "Funding Agencies Can End Profit-First Science Publishing", arXiv / reported in Phys.org, December 2025 — source for the $7.1 billion revenue figure and 30%+ margins for the top four publishers. RELX Annual Report 2022 — source for Elsevier's €3.26 billion revenue and 37.8% profit margin. "Academic Publishers Hit with Antitrust Suit Over Peer Review", Publishers Weekly, September 2024 — source for the antitrust lawsuit details. "When Peer Review Drags On: The Harm to Early Career Researchers", Frontiers in Research Metrics and Analytics, January 2026 — source for submission-to-publication time ranges. "Reviewer Engagement Trends at a Journal: Cause for Concern", PRiMER / PubMed Central — source for the 35.71% acceptance rate in 2024. "The Peer Review Process: Growing Problem in Recruiting Qualified Reviewers", Acta Odontologica Scandinavica, November 2024 — source for the 6–7 invitations per submission figure. "Quantifying Reviewer Declines in Scientific Publishing: 21 Years of Data", Biological Invasions, October 2025 — source for long-term decline in reviewer acceptance rates. "Shortage of Scholarly Peer Reviewers and How to Tackle It", HighWire Press, 2023 — source for the 70% / 42% / 39% decline reasons. "The Effectiveness of Peer Review in Identifying Issues Leading to Retractions", Journal of Informetrics, 2019 — source for the 8.1% rejection suggestion rate. bioRxiv (biorxiv.org) — source for preprint volume data; 43,629 preprints in 2024 figure from Science reporting on the bioRxiv/medRxiv nonprofit transition, March 2025. "In Bid to Expand, bioRxiv and medRxiv Move to Newly Formed Nonprofit", Science/AAAS, March 2025 — source for the 15% of journal articles starting as preprints figure. "The Oligopoly's Shift to Open Access", Quantitative Science Studies, MIT Press — source for APC revenue breakdown by publisher. "Biomedical Paper Retractions Have Quadrupled in 20 Years", Nature, May 2024 — source for the fourfold European retraction rate increase.