A new independent analysis of the Retraction Watch Database finds that Elsevier, one of the largest scholarly publishers by output, has the lowest normalized retraction rate and by far the highest rate of reinstating retracted articles among the ten publishers with the most retractions on record. The finding comes from a preprint by Jonas Oppenlaender, “How Ten Publishers Retract Research”, and was reported by Retraction Watch on May 11, 2026.
What the study actually measured
The analysis draws on the Retraction Watch Database, covering retractions recorded from 1997 through early 2026, a set of 46,087 records. Oppenlaender identified the ten publishers with the largest raw retraction counts, then normalized each publisher’s retraction count against its total publication volume to produce a rate per 10,000 articles published — a step that matters, because a publisher that simply publishes far more articles than its peers will also rack up more retractions in absolute terms without necessarily having a worse editorial track record. For research administrators evaluating a publisher’s retraction history or advising authors on venue choice, the normalized rate, not the raw count, is the figure that is actually comparable across publishers.
The comparative figures
Per the preprint as reported by Retraction Watch, normalized retraction rates (per 10,000 published articles) across the ten publishers were:
- Elsevier: 3.97
- SAGE: 5.46
- Wiley: 6.21
- Taylor & Francis: 6.50
- Springer Nature: 9.06
- IEEE: 17.70
- PLOS: 26.82
- IOS Press: 283.77
- Hindawi: 320.02
Oppenlaender’s own framing of the spread is worth quoting directly: “Every publisher has their own retraction profile and retraction rates vary by two orders of magnitude. This reflects different editorial cultures and detection strategies, not just different levels of misconduct.” That caveat is the load-bearing part of the study — a low normalized rate is consistent with genuinely stronger pre-publication screening, but it is also consistent with slower or less thorough post-publication detection and correction. The two are not distinguishable from retraction counts alone, and the study does not claim to distinguish them.
Hindawi’s outlier position (320.02 per 10,000) reflects the mass retractions tied to paper-mill infiltration that preceded its 2023 wind-down as a standalone Wiley imprint, a pattern already well documented independently of this study. IOS Press’s similarly elevated rate places it in the same tier. At the other end, Elsevier’s low rate sits below SAGE, Wiley, and Taylor & Francis, all of which cluster in a comparatively narrow band roughly 1.4 to 1.6 times higher.
The reinstatement finding
The more distinctive finding concerns reinstatements — cases where a retraction is later reversed and the article restored to the record. Of 98 reinstated articles identified across the dataset, 86 were originally published in Elsevier journals. Elsevier’s reinstatement rate works out to roughly 1.3% of its own retracted articles being later reinstated; Taylor & Francis was the next closest publisher, at roughly 0.4%. In absolute and proportional terms, Elsevier accounts for the large majority of reinstatements across the entire ten-publisher dataset.
The study does not, per Retraction Watch’s reporting, attribute a single specific mechanism to Elsevier’s reinstatement pattern — whether it reflects a more formalized post-retraction appeals process, a particular editorial policy at specific Elsevier journals, or something else in how individual cases were adjudicated. That is a genuine open question the underlying preprint leaves for further scrutiny, not one this analysis resolves.
A methodological caveat worth naming
The study itself flags at least one publisher-specific limitation: ACM may move some retracted content into a non-public “dark archive” rather than leaving a fully public retraction notice, which could understate ACM’s true retraction count relative to publishers with fully open retraction records. This is a useful reminder that cross-publisher retraction-rate comparisons are only as good as the underlying database’s ability to see equivalent editorial actions across publishers with different archiving and disclosure practices — a structural limitation of the Retraction Watch Database as a comparative instrument, not a flaw specific to this analysis.
Why this matters for research administration
Retraction rate is sometimes used, informally, as a rough proxy for editorial rigor when evaluating a journal or publisher — including in conversations about whether a title’s indexing status is at risk (see CASRAI’s guide on why and how Scopus delists journals, where a journal’s retraction pattern is one input among several databases weigh). This study is a useful corrective to using that proxy too casually: a low normalized retraction rate can reflect strong front-end screening, but as Oppenlaender’s own framing makes explicit, it can just as plausibly reflect less aggressive post-publication detection. Research administrators advising authors on venue selection, or assessing a publisher’s track record for an institutional review, should treat a bare retraction-rate comparison as one data point requiring the same caution CASRAI recommends elsewhere — for instance in coverage of STAT’s investigation into guest-edited special issues, where the structural cause of a retraction cluster mattered more than the raw count.
For a walkthrough of how an individual retraction moves through the editorial process once a concern is raised, see CASRAI’s guide, How a Retraction Actually Happens. For background on the underlying data source this study relies on, see CASRAI’s dictionary entry on Retraction Watch and the practical guide, How to Search the Retraction Watch Database.
