AIMOS 2026 · Wellington · 5-minute lightning talk

Retracted,
but still cited

at least 44,046 times in 2024 alone

Nathan M. Thornhill
Institute for Complexity Science and Advanced Computing · Fort Wayne, Indiana
ORCID 0009-0009-3161-528X

the data, citable and permanent doi:10.7910/DVN/WT8VOA Harvard Dataverse · CC BY 4.0 · 12 files

Retraction is supposed to stop a paper's influence. This counts how often it doesn't.

Every retraction in the Retraction Watch database, joined against the OpenCitations Index. The scripts run on public inputs, so you don't have to take my word for any of it.

the measurement

The numbers

999,842

citations point at a retracted paper

across 46,425 retracted papers

330,834

were made after the retraction date

33.46%. 11,189 could not be dated at all

36,309

retracted papers were cited after retraction

by 300,040 citing works

is the median count per paper

two thirds get five or fewer

621

days is the median gap after the notice

longest in the data: 18,586 days

names in the data

Two of them you already know

Dr. Death covers two men. Only one of them is in this data.

who	cited after	in 2024	retracted	cited before
Andrew Wakefield1 paper, The Lancet 1998	2,941	162	2010	1,589
Paolo Macchiarini13 papers in the Retraction Watch set	522	68	2012–2024	4,801
Christopher Duntsch19 works with a DOI	none	none	not retracted	–

↔ more columns off to the right

Macchiarini's papers have been cited 522 times since the dates they were retracted, 68 of those in 2024 and 64 in 2025. His median gap is 636 days. OpenAlex lists 19 works with a DOI for Duntsch. None of them is in the retraction set.

macchiarini, three most-cited since retraction

paper	cited after	retracted	published	cited before
Tracheobronchial transplantation with a stem-cell-seeded bioartificial nanocomposite: a proof-of-concept study10.1016/s0140-6736(11)61715-7	198	2018	2011	964
Engineered whole organs and complex tissues10.1016/s0140-6736(12)60073-7	134	2018	2012	618
Clinical transplantation of a tissue-engineered airway10.1016/s0140-6736(08)61598-6	82	2023	2008	2,530

↔ more columns off to the right

The 2008 paper collected 2,530 citations before anyone retracted it, and that took until 2023.

figures

The charts from the talk

On a narrow screen a chart pans sideways. The caption carries the same numbers.

post-retraction citations per retracted paper

↔ drag the chart sideways, or just read the caption

The median is three. Two thirds of these papers are cited five times or fewer. Wakefield sits at 2,941, off the right of the chart.

how long after the notice the citation arrives

↔ drag the chart sideways, or just read the caption

Median 621 days. A tenth arrive inside 77 days. A quarter wait longer than 1,568. The longest gap in the data is 18,586 days, just under 51 years.

the same count, eight weeks apart

↔ drag the chart sideways, or just read the caption

Two OpenCitations releases, same method. Every year went up. Citation indexing fills in backwards, so no count here is final.

method

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, 2,556,384,096 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. Each citation is dated against that work's retraction date.
The machine: A mini PC in a spare room. Three workers on four ARM cores, 9,694 seconds. No cluster, no institutional access, no paid API.

the part that matters

How this could be wrong

Every number here is a floor. All of these push the count down.

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.
A year-only citation date resolves to 1 January, which makes every lag read shorter than it was.
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.

the trap that nearly took the whole thing down

RetractionNature has four values, not two. The 8 September export is 66,764 Retraction, 3,713 Expression of concern, 1,509 Correction and 160 Reinstatement. Papers get un-retracted.

Collapse that field to a boolean and you build a list that accuses somebody who was already cleared. An expression of concern is not a retraction either. Only the 66,764 Retraction rows enter the set behind this page.

one release later

No count here is final

The title was submitted in August. OpenCitations shipped a release the morning the talk was accepted, so the join was re-run on it.

44,046

August run

→

51,446

September run

Every earlier year rose too. "At least 44,046" holds as a floor, which is how it was written. The right phrase for 2024 is the worst complete year in this snapshot, not the worst year on record. 2025 already sits at 44,743 with the index still filling in.

reuse

How to cite it

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

Thornhill, N. M. (2026). Citations to retracted works in the OpenCitations Index (July 2026 release), joined to the Retraction Watch database. Harvard Dataverse, V1. https://doi.org/10.7910/DVN/WT8VOA

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