Larry Richardson appeared to be an early-career mathematician with potential. According to Google Scholar, he’d authored a dozen papers on topics ranging from complex algebras to the structure of mathematical objects, racking up more than 130 citations in 4 years. It would all be rather remarkable—if the studies weren’t complete gibberish. And Larry wasn’t a cat.
“It was an exercise in absurdity,” says Reese Richardson, a graduate student in metascience and computational biology at Northwestern University. Earlier this month, he and fellow research misconduct sleuth Nick Wise at the University of Cambridge cooked up Larry’s profile and engineered the feline’s scientific ascent. Their goal: to make him the world’s most highly cited cat by mimicking a tactic apparently employed by a citation-boosting service advertised on Facebook. In just 2 short weeks, the duo accomplished its mission.
The stunt will hopefully draw awareness to the growing issue of the manipulation of research metrics, says Peter Lange, a higher education consultant and emeritus professor of political science at Duke University. “I think most faculty members at the institutions I know are not even aware of such citation mills.”
As a general rule, the more a scientific paper is cited by other studies, the more important it and its authors are in a field. One shorthand is the popular “h-index”: An h-index of 10 means a person has 10 papers with at least 10 citations each, for instance.
Inflating a researcher’s citation count and h-index gives them “a tremendous advantage” in hiring and tenure decisions says Jennifer Byrne, a cancer researcher at the University of Sydney. It also drives the business model of shady organizations that promise to boost your citations in exchange for cash. “If you can just buy citations,” Byrne says, “you’re buying influence.”
Enter Larry the cat. His tale began a few weeks ago, when Wise saw a Facebook ad offering “citation & h-index boosting.” It wasn’t the first promo he and Richardson had seen for such services. (The going rate seems to be about $10 per citation.) But this one linked to screenshots of Google Scholar profiles of real scientists. That meant the duo could see just which citations were driving up the numbers.
The citations, it turned out, often belonged to papers full of nonsense text authored by long-dead mathematicians such as Pythagoras. The studies had been uploaded as PDFs to the academic social platform ResearchGate and then subsequently deleted, obscuring their nature. (Wise and Richardson had to dig into Google’s cache to read the documents.) “We were like, ‘Wow, this procedure is incredibly easy,’” Richardson recalls. “All you have to do is put some fake papers on ResearchGate.”
It’s so easy, Wise noted at the time, that a quickly written script to pump out plausible-sounding papers could make anyone highly cited—even a cat. “I don’t know if he was being serious,” Richardson says. “But I certainly took that as a challenge.” And he knew just the cat to beat: F.D.C. Willard. In 1975, theoretical physicist Jack Hetherington added his Siamese to one of his single-author papers so the references to “we” would make more sense. As of this year, “Felis Domesticus Chester Willard” has 107 citations.
To break that record, Richardson turned to his grandmother’s cat Larry. In about an hour he created 12 fake papers authored by Larry and 12 others that cited each of Larry’s works. That would amount to 12 papers with 12 citations each, for a total citation count of 144 and an h-index of 12. Richardson uploaded the manuscripts to a ResearchGate profile he created for the feline. Then, he and Wise waited for Google Scholar to automatically scrape the fake data.
On 17 July, Larry’s papers and 132 citations appeared on the site. (Google Scholar failed to catch one spurious study, Wise notes.) And, thus, Larry became the world’s most highly cited cat. “I asked Larry what his reaction was over the phone,” Richardson told Science. “I can only assume he was too stunned to speak.”
Although Larry’s profile might seem obviously fake, finding manipulated ones usually isn’t easy, says Talal Rahwan, a computer scientist at New York University Abu Dhabi. Earlier this year, he and Yasir Zaki, a computer scientist at the same institution, and their colleagues scanned more than 1 million Google Scholar profiles to look for anomalous citation counts. They found at least 114 with “highly irregular citation patterns,” according to a paper posted in February on the arXiv preprint server. “The vast majority had at least some of their dubious citations from ResearchGate,” Zaki says.
ResearchGate is “of course aware of the growing research integrity issues in the global research community,” says the company’s CEO, Ijad Madisch. “[We] are continually reviewing our policies and processes to ensure the best experience for our millions of researcher users.” In this case, he says, the company was unaware that citation mills delete content after indexing, apparently to cover their tracks—intel that may help ResearchGate develop better monitoring systems. “We appreciate Science reporting this particular situation to us and we will be using this report to review and adapt our processes as required.”
Google Scholar removed Larry’s citations about 1 week after they appeared, so he has lost his unofficial title. However, his profile still exists, and the dubious citations in the profiles that were in the advertisement remain. So, “They haven’t fixed the problem,” Wise says. Google Scholar did not respond to requests for comment.
It’s not the first time somebody has manipulated Google Scholar by posting fake papers. In 2010, Cyril Labbé, a computer scientist at Grenoble Alpes University, invented a researcher named Ike Antkare (“I can’t care”), and made him the sixth most cited computer scientist on the service by posting fake publications to Labbé institutional website. “Impersonating a fake scientist in a cat is very cute,” Labbé says. “If it can be done for a cat, it can easily be done for a real person.”
For that reason, many researchers would like to see less emphasis on h-index and other metrics that have “the undue glow of quantification,” as Lange puts it. As long as the benefits of manipulating these systems outweigh the risks and costs, Wise says, people are going to continue to try to hack them. “How can you create a metric that can’t be gamed? I’m sure the answer is: You can’t.”
