A society journal is often imagined as the scholarly village well: owned or guided by a learned community, watched by editors who know the field, shaped by disciplinary norms rather than purely commercial machinery. A non-society journal, especially one inside a large publisher pipeline, is often imagined differently: faster, broader, more industrial, sometimes more vulnerable to paper mills and special-issue storms.
But does the Retraction Watch database actually support that neat story?
Not quite. The truth is more interesting, with a few trapdoors, some statistical smoke, and one very large IEEE-shaped elephant in the room. š§Ŗš
I analyzed the uploaded Retraction Watch CSV, using 70,589 records with usable original-publication and retraction-notice dates. I classified journals using a high-confidence society-linked publisher rule: publishers such as IEEE, ACS, RSC, ASBMB, ASM, AAAS, NAS, BMJ, Massachusetts Medical Society, American Heart Association, American Medical Association, IOP, AIP, Royal Society Publishing, and similar society or professional-association publishers were marked as society-linked.
Important caveat: this is conservative. If a society-owned journal is published by Wiley, Springer, Elsevier, OUP, or another commercial/university publisher and the society name is not visible in the Retraction Watch publisher field, it is counted here as non-society/unclassified. So the comparison is best read as:
High-confidence society-linked publishers vs everything else in the database.
Also, these are retraction-record counts, not retraction rates. We do not have the total number of papers published by each journal class, country, or subject.
The headline finding
The answer depends on what kind of retraction we are counting.
If we include conference records, society-linked publications appear much faster to retract because IEEE conference proceedings create a huge fast-retraction block. But if we focus on journal-like records by excluding conference abstracts/papers, the pattern flips:
Society-linked journal records have a longer median time to retraction than non-society/unclassified records.
Society-linked records look very fast only when conference records are included. Excluding conference records, society-linked journals have longer median lags.
Calculated from the uploaded Retraction Watch CSV.
The key numbers:
| Comparison | Society-linked | Non-society / unclassified |
|---|---|---|
| All records | 17,670 records, median 0.17 years | 52,919 records, median 1.59 years |
| Excluding conference records | 5,866 records, median 2.81 years | 50,661 records, median 1.66 years |
| Research articles only | 5,251 records, median 2.86 years | 42,612 records, median 1.67 years |
That is the first big lesson: conference proceedings distort the society-journal comparison. Once conference records are removed, society-linked retractions are less frequent in absolute number but often older, slower, and more forensic.
Are fraud-related retractions more common in non-society journals?
This question needs two answers, because “fraud” can mean different things.
I used two reason categories:
Narrow fraud/misconduct: reasons containing terms such as fabrication, falsification, manipulation, misconduct, false/forged authorship, hoax, or criminal proceedings.
Expanded integrity abuse: narrow fraud/misconduct plus paper mill, compromised peer review, rogue editor, peer-review concerns, and computer-generated or computer-aided content.
Here is the pattern for non-conference records.
Percent of non-conference records tagged with each broad reason theme. Categories overlap because one record can have multiple reasons.
Calculated from the uploaded Retraction Watch CSV.
So the answer is a deliciously inconvenient yes and no.
If by fraud we mean classic lab or author misconduct, society-linked records show a higher share:
23.6% of society-linked non-conference records vs 7.0% of non-society/unclassified records.
This is strongly connected to older biomedical and life-science cases where reasons include misconduct, image manipulation, missing original data, or institutional investigations.
But if by fraud we mean industrialized publication fraud, such as paper mills, compromised peer review, rogue editorial behavior, and computer-generated content, the non-society/unclassified group is much higher:
44.5% of non-society/unclassified records vs 9.6% of society-linked records.
So the better conclusion is:
Society-linked retraction records are more enriched for old-style forensic misconduct and image/data problems. Non-society/unclassified retraction records are more enriched for industrial-scale paper-mill and peer-review-abuse problems.
Two different species of rot, two different smells.
Over time: the pace is not similar
Society-linked and non-society/unclassified retractions do not move in parallel. Their waves come from different machines.
When conference records are included, society-linked records explode in 2010 and 2011 because of IEEE conference-proceedings retractions. But for a fairer journal comparison, the plot below excludes conference abstracts and conference papers.
Non-society/unclassified records show a large 2023 spike, while society-linked journal records remain much smaller and more stable.
Calculated from the uploaded Retraction Watch CSV. The year 2026 is partial.
The non-society/unclassified spike in 2023 is enormous. In the data, it is heavily associated with large publisher and journal clusters, especially Hindawi-linked retractions and paper-mill or compromised-review themes.
Society-linked records, after excluding conferences, do not show the same 2023 eruption. They look more like a steady stream of older, field-specific retractions.
So the pace is not similar. The non-society/unclassified group has industrial spikes. Society-linked journals have slower forensic pulses.
Subject patterns: society-linked retractions are concentrated differently
Among non-conference records, society-linked retractions are most visible in biology/life sciences and physical sciences/engineering. But their share is modest in every broad subject group, because most retraction records in the database are classified as non-society/unclassified under the conservative publisher-only rule.
Non-conference records only. Percent shows the share of retraction records in each broad subject group linked to high-confidence society publishers.
Calculated from the uploaded Retraction Watch CSV. Subject categories overlap because records can have multiple subjects.
The median lag by subject also differs:
| Subject group | Society-linked median | Non-society / unclassified median |
|---|---|---|
| Biology / life sciences | 3.56 years | 2.29 years |
| Physical sciences / engineering | 2.00 years | 1.55 years |
| Health sciences / medicine | 2.75 years | 1.63 years |
| Social sciences | 3.27 years | 1.39 years |
| Business / technology / computing | 2.74 years | 1.48 years |
The society-linked side is consistently slower in the journal-like subset. That likely reflects the composition of society-linked retractions: older biomedical, biochemical, cell-biology, cancer, microbiology, chemistry, and physics journals where image manipulation, original-data availability, and institutional investigations are common.
The non-society/unclassified side has many recent paper-mill and compromised-peer-review batches, which can be corrected faster once a publisher-wide investigation starts.
Country patterns: society-linked retractions are not evenly distributed
Country attribution is tricky because one paper can list multiple countries. I counted a multi-country paper once for each listed country. Again, this is not a national retraction rate.
The plot below shows the share of non-conference retraction records that came from society-linked publishers among countries with large record counts.
Non-conference records only. Countries with at least 500 country-paper occurrences are shown.
Calculated from the uploaded Retraction Watch CSV. Multi-country papers are counted once for each listed country.
The pattern is striking:
| Country pattern | Interpretation |
|---|---|
| United States, Japan, South Korea, Italy, France, Canada, Spain | Higher society-linked share among retracted records |
| China, India, Pakistan, Saudi Arabia, Malaysia, Ethiopia | Much lower society-linked share among retracted records |
| Japan and France | Long lags in both society and non-society groups |
| China and India | Non-society/unclassified records are much more numerous and generally faster |
This probably reflects where different countries’ retracted records sit in the publication ecosystem. China and India have large concentrations in non-society journals affected by publisher-wide investigations, paper mills, compromised peer review, and special-issue cleanup. The United States and Japan have more society-linked biomedical, biochemical, medical, and life-science records, many with long-lag forensic issues.
Again, this is not a statement about national scientific quality. It is a map of where retracted papers from those countries appear in the database.
Society-linked journal clusters: slower, older, more forensic
The largest society-linked journal clusters are not dominated by paper-mill megabatches. They are dominated by biochemical, cancer, chemistry, crystallography, society proceedings, and elite general-science journals.
| Society-linked journal cluster | Records | Median lag |
|---|---|---|
| The Journal of Biological Chemistry | 436 | 7.32 years |
| Bioscience Reports | 302 | 3.13 years |
| RSC Advances | 247 | 2.56 years |
| PNAS | 176 | 3.30 years |
| Science | 157 | 1.98 years |
| Journal of Testing and Evaluation | 148 | 2.74 years |
| Acta Crystallographica Section E | 141 | 2.97 years |
| Cancer Research | 114 | 9.22 years |
| Journal of Cell Science | 72 | 6.03 years |
| Journal of Clinical Investigation | 71 | 3.87 years |
This table explains why society-linked non-conference records have a longer median lag. Some of these journals are old, central journals in fields where image-data scrutiny, institutional investigations, and raw-data questions can emerge years later.
The long-lag society-linked pattern is especially visible in journals such as Journal of Biological Chemistry, Cancer Research, Diabetes, Circulation Research, and Journal of Cell Science. These are not usually fast paper-mill cleanup stories. They are often slow forensic stories.
Non-society/unclassified journal clusters: larger, faster, more industrial
The largest non-society/unclassified journal clusters are different beasts.
| Non-society / unclassified journal cluster | Records | Median lag |
|---|---|---|
| Journal of Intelligent & Fuzzy Systems | 1,565 | 1.77 years |
| PLoS One | 1,485 | 4.31 years |
| Journal of Healthcare Engineering | 1,074 | 1.61 years |
| Computational and Mathematical Methods in Medicine | 1,066 | 1.26 years |
| Computational Intelligence and Neuroscience | 1,028 | 1.25 years |
| Security and Communication Networks | 949 | 1.38 years |
| Arabian Journal of Geosciences | 779 | 0.30 years |
| Wireless Communications and Mobile Computing | 762 | 1.16 years |
| Evidence-Based Complementary and Alternative Medicine | 753 | 1.22 years |
| BioMed Research International | 685 | 1.42 years |
| Cochrane Database of Systematic Reviews | 574 | 8.32 years |
| Soft Computing | 566 | 2.67 years |
Here the signature is bigger and more batch-like. Hindawi journals dominate several clusters, with median lags around one to two years. IOS Press, Springer, Wiley, SAGE, PLoS, and others appear strongly too.
The Cochrane Database of Systematic Reviews is an exception: many records there have long lags because review updates, withdrawals, and review-literature corrections follow a different lifecycle.
So non-society/unclassified is not one category. It contains at least two worlds:
- Fast industrial correction: paper mills, peer-review problems, special issues, publisher audits.
- Slow review or biomedical correction: PLoS One, Cochrane, Spandidos, OUP, some clinical journals.
Country-journal clusters: where the map becomes visible
The largest society-linked country-journal clusters:
| Country-journal cluster, society-linked | Records | Median lag |
|---|---|---|
| China, Bioscience Reports | 297 | 3.14 years |
| United States, Journal of Biological Chemistry | 251 | 6.85 years |
| China, RSC Advances | 156 | 2.20 years |
| United States, PNAS | 138 | 3.87 years |
| China, Acta Crystallographica Section E | 138 | 2.97 years |
| China, Journal of Testing and Evaluation | 119 | 2.74 years |
| United States, Science | 101 | 2.17 years |
| United States, Cancer Research | 91 | 10.78 years |
| United States, Journal of Clinical Investigation | 61 | 3.67 years |
The largest non-society/unclassified country-journal clusters:
| Country-journal cluster, non-society/unclassified | Records | Median lag |
|---|---|---|
| China, Computational and Mathematical Methods in Medicine | 997 | 1.24 years |
| China, Journal of Healthcare Engineering | 982 | 1.62 years |
| China, Journal of Intelligent & Fuzzy Systems | 974 | 1.90 years |
| China, Computational Intelligence and Neuroscience | 888 | 1.22 years |
| China, Security and Communication Networks | 857 | 1.36 years |
| China, Arabian Journal of Geosciences | 767 | 0.30 years |
| China, Wireless Communications and Mobile Computing | 740 | 1.17 years |
| China, BioMed Research International | 732 | 1.67 years |
| China, Evidence-Based Complementary and Alternative Medicine | 705 | 1.17 years |
| India, Journal of Intelligent & Fuzzy Systems | 429 | 1.56 years |
| United Kingdom, Cochrane Database of Systematic Reviews | 321 | 8.90 years |
| United States, PLoS One | 285 | 7.04 years |
This table captures the whole story in miniature.
Society-linked clusters: fewer records, longer lags, often biochemical, biomedical, chemistry, or elite multidisciplinary journals.
Non-society/unclassified clusters: larger record counts, many fast-lag computational, medical-engineering, special-issue, and publisher-audit clusters, plus a few long-lag exceptions such as Cochrane and PLoS One.
What about publisher differences?
For non-conference records, the largest society-linked publishers are:
| Society-linked publisher | Records | Median lag |
|---|---|---|
| Royal Society of Chemistry | 539 | 2.40 years |
| American Society for Biochemistry and Molecular Biology | 452 | 7.09 years |
| American Chemical Society | 418 | 1.68 years |
| Portland Press | 331 | 3.25 years |
| American Association for Cancer Research | 233 | 8.67 years |
| American Society for Microbiology | 211 | 4.28 years |
| AAAS | 187 | 1.97 years |
| IOP Publishing | 177 | 1.56 years |
| National Academy of Sciences | 176 | 3.30 years |
| IEEE | 164 | 0.30 years |
| BMJ Publishing | 153 | 1.83 years |
| American Heart Association | 150 | 6.34 years |
The largest non-society/unclassified publishers are:
| Non-society / unclassified publisher | Records | Median lag |
|---|---|---|
| Hindawi | 11,524 | 1.31 years |
| Elsevier | 6,951 | 1.70 years |
| Springer | 5,152 | 1.46 years |
| Wiley | 4,145 | 2.56 years |
| Springer Nature Publishing Group | 2,849 | 1.97 years |
| Taylor and Francis | 1,855 | 2.15 years |
| IOS Press | 1,743 | 1.73 years |
| SAGE Publications | 1,722 | 2.00 years |
| PLoS | 1,610 | 4.38 years |
| Spandidos | 1,004 | 5.68 years |
| Oxford University Press | 959 | 4.71 years |
This is not a clean “society good, non-society bad” picture. It is more like:
Society-linked journals have fewer but older and more forensic retraction records.
Non-society/unclassified journals have many more records, including very large recent batch events.
The most important hidden variable: article type
Conference records radically change the story. IEEE alone creates a large society-linked block in the full dataset, and many of those records are fast. That is why all society-linked records have a median lag of 0.17 years, but society-linked non-conference records have a median of 2.81 years.
So any society vs non-society comparison must specify:
- Are conference proceedings included?
- Are only research articles included?
- Are expressions of concern included?
- Are corrections and reinstatements included?
- Are paper-mill mass retractions included?
- Are publisher-hosted society journals identifiable?
Without these distinctions, the analysis becomes statistical soup with a DOI garnish.
The core answer to the user’s questions
Are retractions more common in non-society journals?
In this dataset, yes in raw counts, but not as a rate. Non-society/unclassified records dominate: 52,919 valid dated records vs 17,670 society-linked records. But because we do not know total publication output, we cannot say non-society journals have a higher probability of retraction.
Is fraud more common in non-society journals?
It depends on the fraud definition.
| Fraud definition | Society-linked | Non-society / unclassified | Interpretation |
|---|---|---|---|
| Narrow fraud/misconduct | 23.6% | 7.0% | Higher in society-linked records |
| Expanded integrity abuse including paper mills and peer review | 31.6% | 50.7% | Higher in non-society/unclassified records |
Are some subjects more common in society vs non-society retractions?
Yes. Society-linked retractions are more concentrated in biology/life sciences and physical sciences/engineering, but still represent a minority of records in every broad subject. Non-society/unclassified records dominate health, computing, social sciences, environmental sciences, and paper-mill-heavy clusters.
Have retractions proceeded at a similar pace over time?
No. Society-linked records show conference-driven spikes in 2010 and 2011 if all records are included. After removing conference records, society-linked journal records are relatively smaller and steadier. Non-society/unclassified records show huge recent spikes, especially 2023, driven by publisher-wide cleanup and paper-mill-related patterns.
Are retractions from some countries more common in society vs non-society journals?
Among retracted records, yes. The United States, Japan, South Korea, Italy, France, Canada, and Spain have higher society-linked shares. China, India, Pakistan, Saudi Arabia, Malaysia, and Ethiopia have lower society-linked shares and much larger non-society/unclassified concentrations.
Are articles from some countries in some specific journals retracted more often?
Yes, in database-count terms. China dominates many large non-society journal clusters in Hindawi and related computational/medical-engineering journals. The United States dominates several society-linked biomedical clusters such as Journal of Biological Chemistry, PNAS, Science, and Cancer Research. India appears strongly in non-society computational and fuzzy-systems journals.
Final interpretation: two retraction cultures
The database suggests two broad retraction cultures.
1. The society-linked correction culture
Smaller in count, slower in median lag, enriched for:
- image concerns,
- narrow misconduct/fraud labels,
- older biomedical and life-science cases,
- institutional investigations,
- long-tail forensic corrections,
- prominent society journals.
This is the slow detective novel of retraction: gels, blots, institutional committees, missing raw data, correspondence, old claims, careful excavation.
2. The non-society/unclassified correction culture
Much larger in count, faster in many recent clusters, enriched for:
- paper mills,
- compromised peer review,
- computer-generated content,
- journal/publisher investigations,
- special-issue failures,
- large batch retractions,
- strong country-journal clusters.
This is the factory-floor version of retraction: conveyor belts, audit tools, publisher sweeps, metadata anomalies, suspicious review networks, and whole stacks of papers falling at once.
Neither world is innocent. They fail differently.
Society journals do not magically prevent fraud. Non-society journals do not automatically produce lower-quality science. But the failure modes differ. Society-linked retractions look older, deeper, and more forensic. Non-society/unclassified retractions look larger, faster, and more industrial.
That is the real lesson: the retraction landscape is not divided between saints and sinners. It is divided between different publishing ecologies, each with its own vulnerabilities.
Some errors hide in old blots.
Some arrive in paper-mill batches.
Some wear the badge of a learned society.
Some ride through a mega-publisher pipeline.
Science corrects itself, yes, but the correction machinery has many engines. Some are scalpels. Some are bulldozers. š¬š
No comments:
Post a Comment