IN 2003 YEAR The investigation of Eric Poehlman, an expert on aging and obesity, found that he had falsified data. He was jailed for using invented results to earn grants. The newspapers duly removed his work.
This should have ended its impact on academia. This is not the case. One of his articles, on Women’s Body Composition, has been cited 400 times since it was retracted.
Such wrongdoing is rare: approximately one in 2,500 studies are withdrawn. Yet, items that are shrunk often have a long lifespan.
The scholarship works like building blocks, with each article citing a myriad of studies. This makes it impossible to erase soiling from an unwanted item. Even if the retraction of an article weakens any existing work that refers to it, these studies remain in the books.
A zombie article like Mr. Poehlman’s, which continues to be cited even after its removal, sounds much worse. In fact, this is the norm. To track such incidents, we entered a list of 20,000 withdrawn articles in an archive collected by Retraction Watch, a nonprofit group, in Semantic Scholar, an academic reference database. Of the 13,000 retracted articles that were cited at least once, 84% had a retraction citation.
It only takes one referral for an unwanted study to sneak in. Together, the 20,000 articles in the archives were cited in 95,000 articles after their retractions. In turn, these were cited in 1.65 million other articles.
The retractions have at least reduced the quotes. In the year following a withdrawal, referrals to a typical retracted item declined by approximately 30% and continued to decline thereafter. In contrast, citations of similar articles that were not withdrawn only decreased by 7%.
However, the magnitude of this effect varied by domain. Authors in political science and biology were exceptionally likely to cite retracted work. Education and legal specialists scrupulously avoided such papers. And the covid-19 pandemic has prompted an increase in references to studies of the living dead. Articles mentioning the disease, often produced with unusual haste, were three times more likely than others to cite retracted research.
What reforms could keep the zombies in their graves? No one wants to cite a withdrawn work, but checking the withdrawal status of papers is needlessly tedious. When journals withdraw articles, they could replace online versions with notices indicating the reason for the withdrawal 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; Cross reference; Kaggle; Scimago
This article appeared in the Graphic Detail section of the print edition under the title “C’est une éraatignure”