Hybrid Open Access Has 7x the Retraction Rate of Closed Access, While Diamond OA Is Cleaner Than Closed

USE CASE. OA policy researchers (cOAlition S, Plan S, OpenAire), institutional funders (NIH Public Access, Wellcome Open Access policy, EU Horizon Europe), and university librarians making transformative agreement decisions need to know whether OA publishing is associated with elevated retraction rates and which OA model is responsible. The conventional narrative (most loudly visible during the Hindawi/Wiley paper-mill cleanup) treats open access as a single category and concludes OA correlates with retractions. The non-obvious cut is the OA TYPE breakdown. RESULT. Per-OA-status retraction rates across OpenAlex 2020-2025 (per 1,000 papers): HYBRID 5.12 (21,301 retractions / 4,158,908 papers), BRONZE 2.97 (9,066 / 3,049,591), GOLD 2.15 (18,150 / 8,458,759), CLOSED 0.69 (18,759 / 27,333,127), DIAMOND 0.59 (5,336 / 9,066,951), GREEN 0.12 (1,851 / 15,951,240). The hybrid OA rate of 5.12 per 1,000 is 7.4 times the closed-access rate of 0.69 and 8.7 times the diamond OA rate of 0.59. The naive 'all OA combined' rate (1.37 per 1,000, computed from the open_access.is_oa=true subset) is double the closed rate, but this masks the fact that the elevation is driven entirely by hybrid + bronze + gold while diamond and green are at or below the closed-access baseline. STRUCTURAL READING. The hybrid OA model is the dominant revenue model for the largest commercial scholarly publishers (Wiley, Springer-Nature, Elsevier, Taylor & Francis) — authors pay an article processing charge (typically $2,000-$11,000) to unlock a single article in an otherwise subscription journal. Hybrid OA is also the venue where Wiley's Hindawi acquisition disaster played out: the Wiley-acquired Hindawi imprints became hybrid in the Wiley taxonomy after acquisition, and the mass retractions of paper-mill content concentrated in those journals show up as hybrid retractions in OpenAlex. The 5.12-per-1000 hybrid rate is therefore partly a Hindawi cleanup artifact and partly a structural feature of the APC-driven hybrid model where paper-mill submissions exploit the pay-to-publish workflow. Diamond OA (scholar-society journals free to read AND free to publish, e.g., Annals of Mathematics, the Journal of the Statistical Society, ACL/EMNLP proceedings) at 0.59 is BELOW the closed-access baseline. Green OA at 0.12 (researchers self-archiving manuscripts in institutional repositories or arXiv) is the cleanest category by an order of magnitude, but the green classification is partly a measurement artifact: green papers are typically peer-reviewed in mainstream subscription venues whose retraction enforcement is identical to closed papers, and a paper being self-archived doesn't make the retraction more or less likely; the measured rate reflects the green-archive flag being predominantly applied to high-quality, well-vetted papers. The defensible structural finding is: among the THREE categories that represent distinct publishing models (closed, hybrid, diamond), diamond OA has the LOWEST retraction rate, closed is in the middle, and hybrid is the highest by a factor of 7-9. The 'open access = paper mill' framing should be replaced by 'pay-to-publish OA models are paper-mill vulnerable; non-pay-to-publish OA models are not'. CAVEATS. (1) The hybrid rate is partly driven by the Hindawi/Wiley cleanup, which retracted ~10,000 papers in 2022-2024 across former Hindawi imprints; if Wiley had not done that cleanup, the hybrid rate would be substantially lower in the visible OpenAlex retraction count (though the underlying paper-mill problem would still exist). (2) The OA status field can shift over time as a journal's business model changes; OpenAlex assigns a single oa_status per paper based on its current state. (3) Diamond OA is heavily concentrated in mathematics, theoretical CS, and Latin American social science journals; the per-paper rate may also reflect field-specific norms about retraction. (4) Green OA's 0.12 rate is partly a measurement artifact and should not be interpreted as 'self-archiving prevents fraud'.

When a scientific paper is published, it can be 'open access' (free for anyone to read) or 'closed access' (paywalled, you need a subscription). Open access has several flavors. 'Diamond OA' is when scholarly societies publish journals for free — both free to read AND free to publish; these are typically run by university math departments or learned societies, with no money changing hands. 'Hybrid OA' is when you pay a $2,000-$11,000 article processing charge to a big subscription journal (Nature, Cell, Wiley, Springer) to unlock your single article so the public can read it; the rest of the journal stays paywalled. 'Gold OA' is journals that are entirely OA and charge an APC for every article (PLOS ONE, MDPI, BMC). 'Green OA' is when an author posts the manuscript in a free repository like arXiv even though the official version is paywalled. There has been a long-running narrative in scientific publishing that 'open access correlates with retractions' — the idea being that paying-to-publish creates a financial incentive to publish bad papers. I downloaded retraction stats from OpenAlex and computed retraction rates per 1,000 papers for each OA category. The naive open-access-vs-closed split shows OA at 1.37 vs closed at 0.69 — about 2x worse, which is the headline number you see in trade press. But the breakdown by OA TYPE flips this story. Hybrid OA is at 5.12 retractions per 1,000 papers — 7.4 times worse than closed access. Diamond OA is at 0.59 per 1,000 — actually CLEANER than closed access. Green OA (self-archived) is at 0.12, the cleanest of all. Gold OA is in the middle at 2.15. So 'open access correlates with retractions' is wrong on the breakdown. The right framing is: pay-to-publish OA models (hybrid most of all, then bronze, then gold) are paper-mill vulnerable. Free-to-publish OA models (diamond, scholarly society journals) are not. The hybrid OA rate is the worst by a factor of 7-9, and that's the model the big commercial publishers (Wiley, Springer-Nature, Elsevier) make most of their OA revenue from. Why this matters: institutional librarians negotiating transformative agreements with big publishers, OA policy researchers in cOAlition S / Plan S / Wellcome, and NIH Public Access policy administrators all rely on understanding which OA models actually work. The 'OA is bad' framing is wrong for diamond and green; the 'we should mandate OA' framing is wrong if the mandate funds hybrid APCs. The right policy lever is to shift funding from hybrid APCs to diamond support, since diamond OA delivers the public-access goal without the paper-mill exposure of hybrid.

hybrid 5.12 (21,301 / 4,158,908) — pay-to-publish single-article unlock in subscription journals (Wiley, Springer, Elsevier) · bronze 2.97 (9,066 / 3,049,591) — read-only OA without explicit license · gold 2.15 (18,150 / 8,458,759) — fully OA journals with author APC · closed 0.69 (18,759 / 27,333,127) — subscription paywall · diamond 0.59 (5,336 / 9,066,951) — fully OA, no APC, scholar-society model · green 0.12 (1,851 / 15,951,240) — author self-archive in repository (measurement artifact)

Naive (is_oa true vs false): OA 1.37/1000 (55,704/40,685,474) vs closed 0.69/1000 (18,759/27,333,102) → 'OA correlates with 2x retraction rate' · Nuanced (per-oa-status): hybrid 5.12 / bronze 2.97 / gold 2.15 / closed 0.69 / diamond 0.59 / green 0.12 → 'pay-to-publish OA models are paper-mill vulnerable; free-to-publish models are not' · The structural ratio: hybrid/diamond = 8.7x, hybrid/closed = 7.4x

67.6 million papers across all OA statuses · 74,463 retracted papers · system mean 1.10 per 1,000 · OA share of papers 60.0% · OA share of retractions 74.8% (consistent with the OA elevation driven by hybrid + gold + bronze)