The Wall Street Journal – Data Science Lab Conducted Tests on Users With Little Oversight
By REED ALBERGOTTI
Thousands of Facebook Inc. FB -0.24% users received an unsettling message two years ago: They were being locked out of the social network because Facebook believed they were robots or using fake names. To get back in, the users had to prove they were real.
In fact, Facebook knew most of the users were legitimate. The message was a test designed to help improve Facebook’s antifraud measures. In the end, no users lost access permanently.
The experiment was the work of Facebook’s Data Science team, a group of about three dozen researchers with unique access to one of the world’s richest data troves: the movements, musings and emotions of Facebook’s 1.3 billion users.
The little-known group was thrust into the spotlight this week by reports about a 2012 experiment in which the news feeds of nearly 700,000 Facebook users were manipulated to show more positive or negative posts. The study found that users who saw more positive content were more likely to write positive posts, and vice versa.
Facebook Chief Operating Officer Sheryl Sandberg said Wednesday during a trip to India that the study was “part of ongoing research companies do to test different products” and was “poorly communicated.”
The company said that after the feedback on the study, “We are taking a very hard look at this process to make more improvements.”
Until recently, the Data Science group operated with few boundaries, according to a former member of the team and outside researchers. At a university, researchers likely would have been required to obtain consent from participants in such a study. But Facebook relied on users’ agreement to its Terms of Service, which at the time said data could be used to improve Facebook’s products. Those terms now say that user data may be used for research.
“There’s no review process, per se,” said Andrew Ledvina, a Facebook data scientist from February 2012 to July 2013. “Anyone on that team could run a test,” Mr. Ledvina said. “They’re always trying to alter peoples’ behavior.”
He recalled a minor experiment in which he and a product manager ran a test without telling anyone else at the company. Tests were run so often, he said, that some data scientists worried that the same users, who were anonymous, might be used in more than one experiment, tainting the results.
Facebook said that since the study on emotions, it has implemented stricter guidelines on Data Science team research. Since at least the beginning of this year, research beyond routine product testing is reviewed by a panel drawn from a group of 50 internal experts in fields such as privacy and data security. Facebook declined to name them.
Company research intended to be published in academic journals receives additional review from in-house experts on academic research. Some of those experts are also on the Data Science team, Facebook said, declining to name the members of that panel.
A spokesman said Facebook is considering additional changes.
Since its creation in 2007, Facebook’s Data Science group has run hundreds of tests. One published study deconstructed how families communicate, another delved into the causes of loneliness. One test looked at how social behaviors spread through networks. In 2010, the group measured how “political mobilization messages” sent to 61 million people caused people in social networks to vote in the 2010 congressional elections.
Many of Facebook’s data scientists hold doctoral degrees from major universities in fields including computer science, artificial intelligence and computational biology. Some worked in academic research before joining Facebook.
Adam Kramer, the lead author of the study about emotions, said in a 2012 interview on Facebook’s website that he joined the company partly because it is “the largest field study in the history of the world.” Mr. Kramer, who has a doctorate in social psychology from the University of Oregon, said that in academia he would have had to get papers published and then hope that someone noticed. At Facebook, “I just message someone on the right team and my research has an impact within weeks, if not days.”
Much of Facebook’s research is less controversial than the emotions study, testing features that will prompt users to spend more time on the network and click on more ads. Other Internet companies, including Yahoo Inc., YHOO +0.72% Microsoft Corp.MSFT -0.24% ,
The recent ruckus is “a glimpse into a wide-ranging practice,” said Kate Crawford, a visiting professor at the Massachusetts Institute of Technology’s Center for Civic Media and a principal researcher at Microsoft Research. Companies “really do see users as a willing experimental test bed” to be used at the companies’ discretion.
Facebook’s team has drawn particular interest because it occasionally publishes its work in academic journals that touch on users’ personal lives, including the study about positive and negative posts.
“Facebook deserves a lot of credit for pushing as much research into the public domain as they do,” said Clifford Lampe, an associate professor at the University of Michigan’s School of Information who has worked on about 10 studies with Facebook researchers. If Facebook stopped publishing studies, he said, “It would be a real loss for science.”
Dr. Lampe said he has been in touch with members of the Data Science team since the controversy erupted. “They’ve been listening to the arguments and they take them very seriously,” he said.
Mr. Ledvina, the former Facebook data scientist, said some researchers debated the merits of a study similar to the one that accused users of being robots but there was no formal review, and none of the users in the study were notified that it was an experiment.
“I’m sure some people got very angry somewhere,” he said. “Internally, you get a little desensitized to it.”
Write to Reed Albergotti at email@example.com