IN 2003 AN investigation into Eric Poehlman, an expert on ageing and obesity, found that he had faked data. He was imprisoned for using made-up results to win grants. Journals duly withdrew his work.

The Economist Today

Hand-picked stories, in your inbox

A daily email with the best of our journalism

This should have ended his impact on academia. It didn’t. One of his articles, on the composition of women’s bodies, has been cited 400 times since it was retracted.

Such wrongdoing is rare: around one in 2,500 studies is retracted. Yet papers that do get retracted often have long afterlives.

Scholarship works like building blocks, with each paper citing myriad studies. This makes expunging the taint of a junk article impossible. Even though retracting a paper weakens all existing work that has referred to it, those studies remain on the books.

A zombie article like Mr Poehlman’s, which keeps getting cited even after it is withdrawn, sounds much worse. In fact it is the norm. To track such mishaps, we fed a list of 20,000 withdrawn papers in an archive amassed by Retraction Watch, a non-profit group, into Semantic Scholar, a database of academic references. Of the 13,000 retracted papers that were cited at least once, 84% had a post-retraction citation.

It takes only one reference for a junk study to burrow in. Together, the 20,000 papers in the archive were cited in 95,000 articles after their retractions. In turn, these were cited in 1.65m further papers.

Retractions did at least put a dent in citations. In the year after a withdrawal, references to a typical retracted article fell by around 30%, and continued declining after that. In contrast, citations of similar articles that were not retracted fell by only 7%.

However, the size of this effect varied by field. Authors in political science and biology were unusually likely to cite retracted work. Those in education and law avoided such papers most scrupulously. And the covid-19 pandemic caused an increase in references to undead studies. Papers mentioning the disease, often produced with unusual haste, were three times likelier than others were to cite retracted research.

What reforms might keep the zombies in their graves? No one wants to cite withdrawn work, but checking papers’ retraction status is unnecessarily tedious. When journals withdraw articles, they could replace online versions with notices stating the reason for the retraction, and notify administrators of research databases. They could also use tools like Semantic Scholar to ensure that references remain valid.

Sources: Retraction Watch; Semantic Scholar; CrossRef; Kaggle; Scimago

This article appeared in the Graphic detail section of the print edition under the headline “’Tis but a scratch”