A retraction is a dated, public statement that a result should no longer be treated as knowledge. This is what happens to the citations afterwards.
scroll
Every thin arc is one citation, leaving the paper the day it was published and landing in the year it was made. The stamp falls on the year the journal declared the result invalid — pale before, red after.
What it shows — this paper was cited 950 times after it was retracted, against 936 before. Half its citation life happened after it was withdrawn.
3.4 million biomedical papers, placed by what they are about. Neighbours share a subject; the red ones were retracted.
What it shows — the red is not spread evenly. Against a baseline rate of 0.057%, the worst neighbourhood runs at 10.7%, or 188× the average, and half of all retractions sit in 4.3% of the map. Two thirds of neighbourhoods have none at all.
Everything so far counted papers citing the retracted paper directly — authors who at least had it in their hands. The tall columns are the ring beyond: papers citing one of those, whose authors have no reason to know. Same scale; direct citation is the slivers along the bottom.
What it shows — direct citation falls after retraction, from 307 a year to 12. Second-hand citation keeps climbing, from 2,926 in 2018 to 9,318 by 2024. The year before retraction there were 10 second-hand citations for every direct one. By 2026 there are 396.
The correction reaches the first ring. It never reaches the second.
63% is partly an artefact: the median paper spent two years valid and five retracted, so it simply had longer. Divide by the time available instead. One dot per paper — rate before across, rate after up — and the diagonal is where nothing changed.
What it shows — most papers fall below the line, so retraction does slow them. The median lands at 0.62×: about a third off the rate. But 63 of 205 sit above it, cited faster dead than alive.
Retraction slows the spread by about a third. It does not stop it.