PaperlayerView exact revision on arXiv

Source-linked AI summary

The strain on scientific publishing

Mark A. Hanson, Pablo Gómez Barreiro, Paolo Crosetto, Dan Brockington

arXiv:2309.15884v2cs.DL

TL;DR

Rapid growth in journal articles has outpaced growth in researchers able to vet them, creating strain on scientific publishing. Using five metrics to compare publishers, the paper finds that strain is not strictly tied to any one business model, though some behaviors are associated with specific gold open-access publishers.

  • Problem

    Journal-article growth has outpaced the training of researchers able to vet articles, while scientists report being overwhelmed by the volume of new publications.

    Source: §2 · p. 2

  • Method

    The paper compares publishers using five metrics covering article and special-issue growth, turnaround times or rejection rates, and citation behavior termed “impact inflation.”

    Source: §2 · p. 2

  • Results

    Strain is not strictly tied to any one publisher business model, although some behaviors are associated with specific gold open-access publishers.

    Source: §2 · p. 2

  • Takeaways & Limitations

    The study highlights areas needing transparency and actions by publishers, researchers, and funders to respond to strain on scientific publishing.

    Source: §2 · p. 3

    Rejection-rate methods could not be standardized across publishers, so comparisons assume each publisher used a consistent method over time and focus on within-group trends.

    Source: Rejection rates · pp. 8–9

Abstract

from arXiv · show

Scientists are increasingly overwhelmed by the volume of articles being published. Total articles indexed in Scopus and Web of Science have grown exponentially in recent years; in 2022 the article total was approximately ~47% higher than in 2016, which has outpaced the limited growth - if any - in the number of practising scientists. Thus, publication workload per scientist (writing, reviewing, editing) has increased dramatically. We define this problem as the strain on scientific publishing. To analyse this strain, we present five data-driven metrics showing publisher growth, processing times, and citation behaviours. We draw these data from web scrapes, requests for data from publishers, and material that is freely available through publisher websites. Our findings are based on millions of papers produced by leading academic publishers. We find specific groups have disproportionately grown in their articles published per year, contributing to this strain. Some publishers enabled this growth by adopting a strategy of hosting special issues, which publish articles with reduced turnaround times. Given pressures on researchers to publish or perish to be competitive for funding applications, this strain was likely amplified by these offers to publish more articles. We also observed widespread year-over-year inflation of journal impact factors coinciding with this strain, which risks confusing quality signals. Such exponential growth cannot be sustained. The metrics we define here should enable this evolving conversation to reach actionable solutions to address the strain on scientific publishing.

2. Introduction … Description of publisher data

The paper frames rising article volume as a strain on scientific publishing when it outpaces researchers’ capacity to vet work, and examines publisher practices and incentives that may shape that strain. It introduces five metrics and reports that strain is not strictly tied to one publisher business model.

2. Introduction

  • Exponential growth in peer-reviewed articles that outpaces researcher training to vet them defines the “strain on scientific publishing.”Source: §2 · p. 2
  • Growth may reflect inclusivity initiatives or investment in the Global South, which the authors say should be accommodated when it arises from those efforts.Source: §2 · p. 2
  • Growth becomes strain if rigorous vetting slips; paper-mill controversies and mass retractions illustrate this risk.Source: §2 · p. 2
  • The authors assess publisher practices using five metrics: article volume, special-issue share, turnaround times, rejection rates, and citation behaviour termed “impact inflation.”Source: §2 · p. 2, Description of publisher data · p. 4
  • Strain is not strictly tied to any one publisher business model, although some behaviours are associated with specific gold open access publishers.Source: §2 · p. 2
  • The authors call for greater transparency and actions by publishers, researchers, and funders, arguing that existing efforts to address strain are insufficient.Source: §2 · pp. 2–3

3. The triangle of scientific publishing: a conceptual framework

  • Publishers, funders, and researchers interact in a system where output incentives meet quality constraints and researchers’ reliance on journal prestige signals.Source: §3 · p. 3

Description of publisher data

  • The analysis focuses on the last decade, especially 2016–2022, using Scopus data for journals also indexed in Web of Science and publisher-sourced or scraped data.The authors describe this shared-indexing focus as a conservative measure that applies stricter indexing criteria.Source: Description of publisher data · p. 4

Data collection … Web scraping

The study combined publisher and article data from Scimago, Clarivate, publisher sources, and web scraping to compare publication, editorial-time, special-issue, and rejection patterns. Its methods used selected publishers and inclusive article-collection classifications, with data sources and coverage varying across measures.

Data collection

  • Web-scraped data covered selected journals and articles; some counts exceeded Scimago totals because the scrape included publisher journals not indexed, or indexed by only one database.The table summarizes web-scraped data on special-issue shares and turnaround times.Source: Table 1 · p. 5

Publisher and journal-level data: publisher selection

  • The analysis sampled ten publishers spanning long-established mixed-license firms, gold-OA publishers known for special issues, BMC, and PLOS.The sample included Elsevier, Wiley, Springer, Nature, Taylor & Francis, MDPI, Frontiers, Hindawi, BMC, and PLOS.Source: Publisher and journal-level data:… · p. 5
  • Publisher labels followed Scimago corporate branding, while subsidiary brands remained separate to avoid obscuring differences or creating merger-driven changes in publisher totals.For example, merging Hindawi into Wiley would have caused Hindawi to disappear from the 2022 dataset and inflated Wiley’s totals without a change in its existing journals’ behavior.Source: Publisher and journal-level data:… · pp. 5–6

Publisher and journal-level data: data curation

  • Scimago data provided annual article totals and journal metrics for 1999–2022, while Clarivate impact factors covered 16,174 journals across 15 publishers for 2012–2022.Scimago data used the “only WoS journals” filter; Clarivate downloads were limited to 600 journals per publisher, so impact factors represented a subset of those available.Source: Publisher and journal-level data: data… · p. 6
  • Rejection rates came from publisher reports, direct requests, or website scraping; Frontiers’ rates were publisher-wide annual aggregates rather than journal-level data.Sources included Frontiers reports, data supplied by PLOS and Taylor & Francis, and public data for MDPI, Hindawi, and Elsevier.Source: Publisher and journal-level data: data… · p. 6

Article-level data

  • Turnaround time meant submission to formal acceptance, measuring consideration of papers accepted for publication rather than time to first decision.PLOS and Hindawi data came from their article corpora, Wiley shared its corpus on request, and other times were scraped from article pages.Source: Article-level data · p. 7
  • Articles labeled as special issues or comparable collections—such as “Theme Issues,” “Collections,” or “Topics”—were included under an intentionally inclusive classification.Such collections may be handled by guest or in-house editors and vary in format and editorial practice.Source: Article-level data · p. 7

Web scraping

  • Publisher-specific scraping retrieved article HTML or XML and extracted editorial-time and special-issue metadata; incomplete Elsevier special-issue data led to its exclusion from that analysis.Scripts were tailored to website formats; Taylor & Francis and Elsevier turnaround data used CrossRef ISSNs and DOIs, while Elsevier special-issue coverage was too spotty for analysis.Source: Web scraping · pp. 7–8

Global researcher statistics … A few publishers disproportionately contribute to total article growth

Indexed article output rose sharply from 2016 to 2022, with a small group of publishers accounting for most of the increase through different growth strategies. The section also describes data and metric methods, comparability limits, and tools for exploring the data.

Global researcher statistics

  • The main text presents OECD PhD graduate data from 37 countries through 2020; China and India data through 2019 used different parameters and appear only in supplementary material.The supplementary material also includes confirmed and estimated trends beyond OECD countries.Source: Global researcher statistics · p. 8

Rejection rates

  • Rejection rates were defined from accepted, rejected, and total submissions according to publisher data availability, but cryptic factors such as revision policies complicate direct comparisons.Frontiers used 1 − (accepted articles / total submissions); MDPI total submissions were accepted plus rejected articles, while Hindawi’s reported “acceptance rate” method is unspecified.Source: Rejection rates · p. 8
  • Because rejection-rate methods could not be standardised, the authors assumed each publisher maintained a consistent method over time, making within-group directions of change relatively robust.Raw rejection-rate comparisons still carry caveats.Source: Rejection rates · pp. 8–9

Impact Inflation

  • “Impact Inflation” is the ratio of Impact Factor to SJR, or the equivalent Cites-per-doc/SJR ratio, indicating when citations come from disproportionately few citing journals.Unlike Impact Factor, SJR limits the prestige contribution from individual sources, including self-citation or circular citation.Source: Impact Inflation · p. 9

Data Availability

  • Because copyright concerns prevented public release of scraped data and scripts, the authors shared supplementary materials and developed a web app for exploring trends, including publishers omitted from the main text.The supplementary materials include README files explaining how to assemble the data.Source: Data Availability · p. 10

A few publishers disproportionately contribute to total article growth

  • MDPI (~27%), Elsevier (~16%), Frontiers (~11%), Springer (~9.5%), and Wiley (~7.0%) contributed >70% of the increase in articles per year since 2016.After normalising output per journal, MDPI and Frontiers were disproportionately high producers, while Elsevier’s per-journal output increased slightly.Source: A few publishers disproportionately… · p. 10
  • In 2022, ~2.82m articles were indexed versus ~1.92m in 2016, an increase of ~897k per year; article growth outpaced PhDs awarded.The year-on-year article growth rate was ~5.6%.Source: A few publishers disproportionately… · p. 10, Figure 1 · p. 11
  • Elsevier and Springer increased output by distributing articles across more journals, while MDPI and Frontiers increased publications handled by a smaller pool of journals.The paper presents these as two mechanisms promoting the recent exponential increase in total articles.Source: A few publishers disproportionately… · p. 12

Growth in articles published through “special issues” · Decreasing mean, increasing homogeneity of turnaround times · Journal rejection rates and trends are publisher-specific

Special-issue publishing grew substantially at some publishers and was associated with faster, more homogeneous turnaround times. Rejection-rate patterns were largely publisher-specific, with a limited association between special-issue use and lower rejection rates.

Growth in articles published through “special issues”

  • Special-issue growth was publisher-dependent, not an inevitable consequence of open-access publishing or article-processing charges.Similar increases were not seen at gold open-access publishers BMC and PLOS, while Nature, Springer, and Wiley published small proportions of special issues.Source: Growth in articles published through… · p. 12
  • From 2016 to 2022, special-issue shares rose drastically at Hindawi, Frontiers, and MDPI, which published majorities this way, especially MDPI.Special issues are invited by journals or editors and commonly delegate responsibilities to guest editors; the reported data cover eight publishers for which data could be collected.Source: Growth in articles published through… · p. 12, Figure 2 · p. 13

Decreasing mean, increasing homogeneity of turnaround times

  • Special-issue articles were typically accepted faster and in increasingly homogeneous timeframes at publishers adopting that model.The authors report that special-issue articles were processed more rapidly than normal articles for most publishers, and that turnaround-time variance declined at Hindawi, Frontiers, and especially MDPI.Source: Decreasing mean, increasing… · p. 14
  • In 2022, MDPI averaged ~37 days to acceptance, versus Frontiers’ 72 and Hindawi’s 83; other sampled publishers exceeded 130 days.MDPI’s average had remained around 37 days since ~2018; Frontiers and Hindawi saw declines from 2020 to 2022, while other publishers’ times were stable or slightly increasing.Source: Decreasing mean, increasing… · p. 14, Figure 3 · p. 15

Journal rejection rates and trends are publisher-specific

Disproportionately inflated Impact Factor affects select publishers

The section defines impact inflation using Cites-per-doc relative to SJR and finds significantly higher 2022 values for MDPI and Hindawi than for other publishers. Their self-citation patterns differ.

Disproportionately inflated Impact Factor affects select publishers

  • IF increased across publishers in recent years, alongside an exponential rise in references per document from 2018–2021, but impact inflation and self-citation were not strictly correlated with other metrics examined.The paper presents impact inflation as a publicly available-data metric for assessing journal citation behaviour.Source: Disproportionately inflated Impact… · pp. 17, 20
  • Impact inflation is Cites-per-doc divided by SJR, using Cites-per-doc as an IF proxy (R2 = 0.77); self-citation or citation rings can raise the former without raising SJR.SJR limits the prestige credit that can come from any single source.Source: Disproportionately inflated Impact… · p. 18
  • In 2022, MDPI and Hindawi had significantly higher impact inflation than all other publishers, while Frontiers was comparable to other publishers despite similar growth.The comparison covered journals with at least 1,000 annual citations.Source: Disproportionately inflated Impact… · p. 18, Figure 5 · p. 19
  • MDPI journals showed elevated within-journal self-citation rates, whereas Hindawi’s rates were more comparable to other publishers despite its high impact inflation.In a 2021 analysis, about 29% of MDPI’s citations came from other MDPI journals; the authors suggest Hindawi’s high inflation may instead involve a small citation network or paper mills.Source: Disproportionately inflated Impact… · p. 18, Figure 5 · p. 19

6. Discussion

The discussion characterizes scientific-publishing strain as a costly, likely underestimated consequence of rapid article growth shaped by interacting researcher, funder, and publisher incentives. It highlights differences in publisher growth strategies and calls for targeted action to make publishing more sustainable.

  • Nearly one million articles were added per year over six years, imposing writing, reviewing, and fee costs; an indexed-article focus likely understates the strain.Source: §6 · p. 20
  • Growth was widespread but concentrated in a few publishers: older houses paired modest journal and per-journal growth, while some gold open-access firms expanded special issues with faster turnaround and comparable strain.Source: §6 · pp. 20–21
  • Special issues had lower rejection rates and lower, more homogeneous turnaround times; impact inflation (IF/SJR) offers an open-data litmus test for citation gaming.Source: §6 · p. 21
  • MDPI led every metric: article growth +1080%, special-issue share 88%, turnaround 37 days, rejection-rate change -8 percentage points, impact inflation 5.4, and within-journal mean self-citation rate 9.5%.Source: §6 · p. 22
  • Publisher policies appear to drive rejection rates and turnaround patterns, prompting calls to protect editorial independence and transparently report special-issue shares, turnaround times, rejection rates, revenues, costs, and index decisions.Source: §6 · p. 22
  • Funders should curb publish-or-perish incentives through narrative CVs; interventions should target publisher behaviors—not business models—especially special-issue use, via funders, researchers, and ethics bodies.Source: §6 · pp. 22–23

Supplementary figures and tables

Supplementary analyses extend the paper’s findings on publishing growth, turnaround times, and citation patterns, while showing that some trends are robust to alternate datasets and vary by publisher.

Supplementary figures and tables

  • Alternate datasets preserved the recent rise in articles relative to PhD graduates and active researchers, although the UNESCO active-researcher proxy may include non-publishing scientists.The article-to-PhD-graduate and article-to-active-researcher ratios rose substantially since 2019.Source: Figure S2 · p. 33
  • Publishing shifted toward fewer, higher-output journals, with fewer journals producing under one paper weekly and more producing over one paper daily, consistent with the “rise of megajournals” described by Ioannidis et al. (2023).The supplementary analysis also reports a slight decline in journals publishing only a few papers per year.Source: Figure S3 · p. 34
  • Across publishers, special-issue articles generally had shorter turnaround times than regular-issue articles; Springer was the exception, with longer special-issue turnaround times.The extent of the differences may vary with how journals and publishers organize special issues.Source: Figure S7 · p. 38
  • Cites/Doc increased year over year, especially after 2019, while SJR remained relatively constant, supporting the interpretation that Impact Inflation reflects more citations rather than increased network-adjusted prestige.References per document also rose exponentially from 2018 to 2021 and continued increasing through 2022, though growth slowed; the analysis notes COVID-19 cannot fully explain the increase because it began before 2020.Source: Figure S10 · p. 42, Figure S12 · p. 44
  • Within-journal self-citation rose in recent years for MDPI, Frontiers, and Hindawi, while established MDPI journals receiving over 1,000 citations annually had uniquely high rates compared with other publishers.MDPI journals also supplied large fractions of incoming citations to their own journals; MDPI’s 2021 analysis can appear more comparable when it includes many newer, lower-output journals.Source: Figure S14 · p. 46, Figure S16 · p. 48, Figure S17 · p. 50
  • MDPI’s IJERPH submissions fell by 73 percentage points in April 2023 versus March after Clarivate announced its delisting, while submissions to MDPI journals broadly declined that month.A similar submission pattern followed the Chinese Academy of Science’s 2021 Early Warning Journal List trial, which included multiple MDPI journals.Source: Table S2 · p. 49

Supplementary Materials and Methods · Additional data considerations

These supplementary materials explain key data choices and limitations, including how publishers and rejection rates were handled and how the proposed impact inflation metric should be interpreted. They also describe robustness checks and data and code availability.

Additional data considerations

  • Including or excluding Arts & Humanities, about 3% of filtered Scimago articles and 8–11% of OECD PhD graduates, did not change analyses of Figures 1 and 5 or Scimago supplementary figures.Source: Additional data considerations · p. 51

Grouping of publishers per Scimago labels

  • Scimago labels were grouped by publisher brand, but independent imprints such as Cell Press, BMC, and Nature Portfolio remained distinct, limiting interpretation of parent-publisher trends.Elsevier and Springer’s aggregated labels still cover most of their journals, but merging Nature into Springer would obscure differences in special issues, turnaround times, impact inflation, and self-citation. The study retained PLOS, Hindawi, BMC, and Nature as relevant comparisons across publishing models.Source: Grouping of publishers per Scimago… · pp. 51–52
  • Journal size and age were considered in publisher comparisons, and self-citation analyses included only journals receiving at least 1000 citations per year.The filters address differences in article output, younger journals’ growth, and small or niche journals’ limited or specialized opportunities to self-cite.Source: Grouping of publishers per Scimago… · p. 52
  • Rejection rates are not standardized across publishers and are not measures of quality or rigour; they reflect editorial capacity, journal scope and mission, and submissions received.Because publishers may count submissions differently as decisions change, absolute comparisons carry caveats; assuming consistent methods within publishers makes year-to-year directions relatively robust.Source: Grouping of publishers per Scimago… · pp. 52–53
  • Publisher-specific rejection-rate calculations differed: Frontiers used 1 - (accepted articles / total submissions), MDPI submissions were defined as accepted plus rejected articles, and Hindawi reported acceptance rates.Frontiers’ reported accepted and rejected counts did not sum to total submissions, while Hindawi did not provide the underlying calculation method.Source: Grouping of publishers per Scimago… · pp. 52–53
  • MDPI journals have lower SJR relative to IF, reflecting citations concentrated within MDPI and related journals, while Frontiers receives citations from a wider pool.The IF/SJR ratio compares a source-agnostic citation measure with SJR, which limits prestige transfer and accounts for citation networks. The authors do not claim to know why these citation behaviours differ.Source: Grouping of publishers per Scimago… · p. 55
  • A Cites/Doc of 3.00 shifted from somewhat high among publisher comparisons in 2017 to near the lowest average in 2022, motivating relative rather than absolute impact reporting.The authors recommend reporting IF rank or percentile and adapting impact inflation comparisons to relative ranks such as quartiles; impact-inflation trends are robust to using Cites/Doc or IF in 2021 or 2022.Source: Grouping of publishers per Scimago… · pp. 54–55
  • The underlying data cannot be redistributed, but File S2 guides readers to public sources, and the authors said full analysis code would be provided upon article acceptance.Data and code were made available to peer reviewers.Source: Grouping of publishers per Scimago… · p. 56

R packages used

The analysis used R version 4.3.1 and a set of 22 R packages spanning the listed dependencies.

R packages used

  • The analysis used R 4.3.1 with agricolae, emmeans, ggtext, gridExtra, gt, and gtExtras.The passage specifies R version 4.3.1 and package versions.Source: R packages used · p. 56
  • The listed packages also included here, hrbrthemes, kableExtra, magick, MASS, multcomp, and multcompView.Package versions are reported in the source passage.Source: R packages used · p. 56
  • The remaining listed packages were MuMIn, mvtnorm, patchwork, scales, sjPlot, survival, TH.data, tidyverse, and waffle.Package versions are reported in the source passage.Source: R packages used · p. 56

Package citations

The section lists software and statistical resources cited by the paper, including R packages for visualization, data handling, and statistical analysis.

Package citations

  • The references include R packages for graphics, plot composition, and presentation-ready tables.Examples include gridExtra, ggbeeswarm, patchwork, gt, and ggplot2-related tools.Source: Package citations · p. 57
  • Statistical resources include packages for multi-model inference, agricultural procedures, estimated marginal means, and paired comparisons.The bibliography also cites works on multivariate probabilities and simultaneous inference.Source: Package citations · p. 57
  • Other citations cover data cleaning, file management, image processing, and the R language and environment.The list also includes the tidyverse and tools for visualization scales and text rendering.Source: Package citations · p. 57

Themes in Publisher Motivation

Publisher statements present scale, author service and retention, open-access expansion, and research-integrity or open-science initiatives as priorities and sources of growth.

Themes in Publisher Motivation

  • Elsevier said its more than 2,900 journals published over 630,000 articles in 2023, representing over 17% of the world’s scientific articles [3].The company described its work as validating, improving, and disseminating research.Source: Themes in Publisher Motivation · p. 58
  • Frontiers reported 125,000 new articles in 2022 and said 92% of authors and 88% of reviewers rated their experience good or excellent [4, 5].The experience ratings followed more than 150,000 pieces of community feedback.Source: Themes in Publisher Motivation · pp. 58–59
  • More than 1 million total articles marked MDPI’s 2022 milestone; it also reported five consecutive years as the largest OA-article publisher and average annual growth above 50% over that span [7].The statements attribute publication growth to attention to authors’ needs, simple processes, and service.Source: Themes in Publisher Motivation · p. 59
  • 65% of rejected Wiley authors were offered another Wiley publishing option, up from 53% at year-end, as the publisher sought to retain submissions within its portfolio [9].Wiley said most rejections stemmed from poor fit with the first journal and described referrals as a way to find those articles another home.Source: Themes in Publisher Motivation · p. 60
  • 4% underlying revenue growth at Springer Nature was driven by its Nature Portfolio journals and open-access business, while institutions expanded subscriptions [12].The company also reported strong submissions for new Nature titles and a fully open-access portfolio of around 600 journals.Source: Themes in Publisher Motivation · p. 60
  • PLOS doubled institutional partnerships that reduce or remove authors’ open-access costs in 2022 [20] and developed Open Science Indicators to measure practices such as data- and code-sharing [19].The publisher identified financial barriers from the APC model as a major concern for researchers.Source: Themes in Publisher Motivation · pp. 62–63

Scimago data download and assembly instructions

The instructions describe downloading Scimago data year by year, assembling the CSVs into a filtered analysis dataset, and enabling the assembly script in Analysis.R. The resulting data primarily support Figures 1 and 5 and related supplementary figures.

OECD, NSF (2022), Zwetsloot et al. (2021) PhD Data

Supplementing PhD data for India and China did not meaningfully change the overall picture of global PhD growth, while gaps in researcher-per-million reporting may skew global totals. The section also specifies the scope and sources of the doctoral data used.

OECD, NSF (2022), Zwetsloot et al. (2021) PhD Data

  • The OECD dataset covers first-time graduates under age 35 at the doctoral or equivalent ISCED level.The India and China supplements used NSF (2022) and Zwetsloot et al. (2021), respectively; the latter was chosen for its more up-to-date data and projections.Source: OECD, NSF (2022), Zwetsloot et al.… · p. 65
  • Adding India and China data did not meaningfully change the global PhD growth picture, so methodologically inconsistent supplemented figures were kept out of the main text.India data came from NSF (2022), and China data from Zwetsloot et al. (2021).Source: OECD, NSF (2022), Zwetsloot et al.… · p. 65
  • Missing or sporadic country reporting means global researchers-per-million figures may be skewed, particularly from 2019–2020 onward.Canada, New Zealand, Switzerland, and India lacked data from 2019 onward; the United States and China also lacked data from 2020 onward, while the United Kingdom supplied an estimate.Source: OECD, NSF (2022), Zwetsloot et al.… · p. 66

Related papers

Loading 2309.15884v2…