Retracted,
but still cited

Retraction is supposed to end a paper's influence. It often doesn't, and how often depends on how you count.

The same count moves with five choices made while counting, and published studies rarely say which way they went. Every retraction in the Retraction Watch database, joined against the full OpenCitations Index. The scripts run on public inputs, so you don't have to take my word for any of it.

One count, five choices

Every one of the five moves the count, and the two that move it most are definitions: where the line between before and after falls, and which records count as retracted. Ranked by how far each one moved it on this data:

1. 1Where the line between before and after falls Moves it most. 18.7% to 45.0% of dated citations count as after, depending on the rule
2. 2Which Retraction Watch records count as retracted Raises it by over a third. +35.7% if the 2,471 DOIs that were never retracted stay in
3. 3Which release of the index Newer releases raise it, mostly in recent citing years. +20.9% over a year of releases, and one release's files 4.4% short of the citations stated for it
4. 4What counts as one citation Barely moves the total, nearly doubles some of the best-known papers. 2.1% fewer counted as citing works than as index rows; Wakefield: 1,519 against 2,941
5. 5When the retraction list was pulled Barely moves it. +0.13% over 20 days

Published studies mostly leave these choices unstated. Of 27 earlier studies whose methods could be read, 14 say when their citation data were pulled, and two name a dated version of the citation index they counted from. Eleven of them use Retraction Watch, and none of those states a filter on its RetractionNature field.

A count can be set beside another only when both state those five choices, and a figure for a recent year needs its release named, because recent years are the least settled.

1Where the line falls

The rule that separates before from after moves the count more than any other choice here, and recent years most of all, because so much of their citing comes soon after a retraction.

Share is of the citations that carry a usable date. The title was written from the first run, on the March 2026 index, where 2024 came to 44,046 index rows. On the July release it's 50,629 citing works, so "at least" holds under the rule this study uses.

2Which records count as retracted

3Which release of the index

Newer releases raise the count, mostly through recent citing years, and one release's files came up short.

Older citing years moved little over the year, and recent ones rose most. On the same retracted papers, from July 2025 to July 2026, 2000 to 2022 grew 3.7% and 2024 rose 10.0%, for 20.9% in all. The March 2026 release sits between them with a gap. Its files match their published checksums but hold 4.4% fewer rows than the citations OpenCitations states for it, and 9,179 pairs present in both July releases are missing from it. Read from March, 2024 seems to fall and then jump, but 2,628 of the 2,632 missing 2024 citations are back in July 2026.

Held fixed here: the 54,534 retracted-paper identifiers that both 2026 runs targeted, counted as citing works. OpenCitations posted its July 2026 release to Figshare on 8 September 2026, so the join was re-run on it, and the July 2025 release was scanned for this comparison. The release figures and the unfiltered-list figures come from further scans, added to the deposit in version 2.0.

4What counts as one citation

The counting unit barely moves the total and nearly doubles some of the best-known papers. Across all post-retraction citations, citing works come out 2.1% below the dump's raw rows. The index holds Wakefield's paper under two identifiers and writes most of its citations twice, so a raw count of the dump's rows gives 2,941 after retraction against 1,519 citing works. Macchiarini's 13 papers give 522 against 281.

5When the list was pulled

Updating the retraction list barely moved it. On the same index, a Retraction Watch export twenty days newer added 413 post-retraction citations, 0.13%.

One you already know

The hit show Dr. Death put Paolo Macchiarini in front of a lot of people who do not read The Lancet. His papers are in this dataset.

His retracted papers are still being cited. Since the dates they were retracted, his 13 papers have been cited by 281 works, 35 of those in 2024 and 32 in 2025, with a median gap of 619 days. Wakefield is in the table for scale, because that is the case this audience already has a number for.

Macchiarini, three most-cited since retraction

The 2008 paper went fifteen years before anyone retracted it, in 2023, and 1,315 works cited it in that time.

How it was counted

Retraction Watch

72,389 rows in the 8 September export, of which 66,764 are retractions. After DOI cleaning: 60,723 distinct DOIs, 54,985 of them keyed into the index.

OpenCitations

94 zip files, 38.27 GB, 2,556,381,540 data rows as measured here. The release page states 2,559,460,009. All of it was read.

The join

On OMID after DOI resolution. 2,546,370 rows touch a retracted work: 999,842 cite one, and the other 1,546,528 are retracted papers' own references. Each citation of a retracted work is dated against its retraction date and counted once per citing work.

Checking it

The data are open and the scripts ship with the deposit, so anyone can rerun the whole count.

How this could be wrong

What this count can't see.

• Citation data carries no sentiment. Some of these citations are people saying the paper is wrong. I can't separate those and I'm not going to pretend I can. Which is why the distribution matters more than the headline. Nobody writes a rebuttal to a paper that got cited three times.
• An imprecise citation date resolves to the start of its year or month. That puts borderline citations on the before side, so the post-retraction count leans low, and it makes those lags read shorter than they were.
• 9.45% of retracted DOIs (5,738 of 60,723) never keyed into the index, so their citations are missing here entirely.
• OpenCitations only sees open reference lists. Coverage is uneven by publisher and era. A better denominator is still not all of science.
• Retraction notices have their own DOIs. OpenAlex flags 5 more Macchiarini DOIs as retracted than the Retraction Watch set does, and four of those are the notices rather than papers. Counting a notice as a retracted paper inflates the number. The Retraction Watch set decides it here.

How to cite it

The deposit is the citable thing and it will outlast this page. Cite that, not this URL.

CC BY 4.0. Reuse it, including commercially, with credit.

References

Nothing here is my data. It is a join of other people's, and this is who they are.

1. The dataset behind this page Thornhill, N. M. (2026). Citations to retracted works in the OpenCitations Index (July 2026 release), joined to the Retraction Watch database. Harvard Dataverse. V2. CC BY 4.0 https://doi.org/10.7910/DVN/WT8VOA
2. The retracted set and the retraction dates The Center for Scientific Integrity (2026). The Retraction Watch Database. Distributed by Crossref. https://api.labs.crossref.org/data/retractionwatchExport of 8 September 2026, 72,389 rows, retrieved from api.labs.crossref.org/data/retractionwatch. Used with credit under Retraction Watch's stated terms. Retraction Watch does not endorse or comment on work built on it, and no affiliation is implied.
3. The citation index, exact version used OpenCitations (2026). OpenCitations Index CSV dataset of all the citation data. figshare. https://doi.org/10.6084/m9.figshare.24356626.v8v8, the July 2026 Index release. CC0. 94 files, 38.27 GB, 2,556,381,540 data rows.
4. The Index, described Heibi, I.; Peroni, S.; Shotton, D. (2019). Software review: COCI, the OpenCitations Index of Crossref open DOI-to-DOI citations. Scientometrics. 121(2), 1213-1228 https://doi.org/10.1007/s11192-019-03217-6
5. The OMID identifiers the join keys on Massari, A.; Mariani, F.; Heibi, I.; Peroni, S.; Shotton, D. (2024). OpenCitations Meta. Quantitative Science Studies. 5(1), 50-75 https://doi.org/10.1162/qss_a_00292
6. OpenCitations as infrastructure Peroni, S.; Shotton, D. (2020). OpenCitations, an infrastructure organization for open scholarship. Quantitative Science Studies. 1(1), 428-444 https://doi.org/10.1162/qss_a_00023
7. Authorship for the named case only, not the retraction status Priem, J.; Piwowar, H.; Orr, R. (2022). OpenAlex: A fully-open index of scholarly works, authors, venues, institutions, and concepts. arXiv. Preprint arXiv:2205.01833 https://doi.org/10.48550/arXiv.2205.01833

How to reach me

Email reaches me directly. The rest are profiles, and nathanthornhill.com carries the full publication list.

research@nathanthornhill.com

Questions about the method or the data.

• nathanthornhill.com The research record, including every preprint and publication.
• CiteStamp Citation grounding against a typed graph. Where the question at the end of the talk comes from.

• Bluesky
• LinkedIn
• Harvard Dataverse
• Google Scholar
• ORCID
• Newsletter