Facebook provided misinformation researchers with flawed and incomplete data about how users interacted with posts and links.

The research group Social Science One had used the data for the past two years without knowledge of a serious flaw: according to internal emails and interviews with the researchers conducted by The New York Times, the data only included interactions of approximately half of Facebook’s users in the United States, not all of them.

The discovery was made by Fabio Giglietto, a social media researcher from the University of Urbino in Italy, after comparing the data they received with data that Facebook released publicly about the company’s most popular posts.

Even that release was a source of criticism for Facebook, as it after the social media giant banned the personal accounts, apps, and Pages of New York University researchers who were also studying misinformation on Facebook’s platforms.

The mistake “undermined trust researchers may have in Facebook,” said Cody Buntain, a social media researcher at the New Jersey Institute of Technology who was part of the group.

“A lot of concern was initially voiced about whether we should trust that Facebook was giving Social Science One researchers good data. Now we know that we shouldn’t have trusted Facebook so much and should have demanded more effort to show validity in the data.”

The researchers claimed they had lost months of work because of the error, with doctoral degrees at risk. One apparently expressed concern that Facebook was intentionally undermining the research.

Facebook apologised to the researchers “for the inconvenience this may [have caused] and would like to offer as much support as possible.” The company said it was updating the data set, but because of its scale an update would take weeks to be done.