Skip to main content

Originally published Sunday, August 17, 2014 at 8:00 PM

  • Share:
  • Comments
  • Print

Researchers grapple with ethics of studying users’ online data

With social media and other companies sitting on a gold mine of personal information, researchers are puzzling out how they may be able to use that data in their studies. One researcher likens it to the point at which chemistry first got the microscope.

The New York Times


Scholars are exhilarated by the prospect of tapping into the vast troves of personal data collected by Facebook, Google, and a host of startups, which they say could transform social-science research.

Once forced to conduct painstaking personal interviews with subjects, scientists can now sit at a screen and instantly play with the digital experiences of millions of Internet users. It’s the frontier of social science — experiments on people who may never even know they are subjects of study, let alone explicitly consent.

“This is a new era,” said Jeffrey Hancock, a Cornell University professor of communication and information science. “I liken it a little bit to when chemistry got the microscope.”

But the new era has brought some controversy with it. Hancock was a co-author of the Facebook study in which the social network quietly manipulated the news feeds of nearly 700,000 people to learn how the changes affected their emotions. When the research was published in June, the outrage was immediate.

Now Hancock and other university and corporate researchers are grappling with how to create ethical guidelines for this kind of research. In his first interview since the Facebook study was made public, Hancock said he would help develop such guidelines by leading a series of discussions among academics, corporate researchers and government agencies like the National Science Foundation.

“As part of moving forward on this, we’ve got to engage,” he said. “This is a giant societal conversation that needs to take place.”

Scholars from the Massachusetts Institute of Technology and Stanford University are planning panels and conferences on the topic, and several academic journals are working on special issues devoted to ethics.

Microsoft Research is a prominent voice in the conversation. It hosted a panel last month on the Facebook research with Hancock and is offering a software tool to scholars to help them quickly survey consumers about the ethics of a project in its early stages.

Facebook, which has apologized for its experiment, declined to comment further, except to say, “We’re talking with academics and industry about how to improve our research process.”

Mostly in-house

Much of the research done by the Internet companies is in-house and aimed at product adjustments, like whether people prefer news articles or cat videos in their Facebook feeds or how to make Google’s search results more accurate.

But bigger social questions are studied as well, often in partnership with academic institutions, and scientists are eager to conduct even more ambitious research.

The Facebook emotion experiment was in that vein. The brainchild of a company data scientist, Adam D.I. Kramer, but shaped and analyzed with help from Hancock and another academic researcher, Jamie Guillory, it was intended to shed light on how emotions spread through large populations.

Facebook deliberately changed the number of positive and negative posts in the subjects’ news feeds over a week in January 2012, then looked at how the changes affected the emotional tone of the users’ subsequent Facebook posts.

Voting study

In another well-known experiment, Facebook sent voting reminders to 61 million American users on Election Day in 2010. Some users also saw a list of their friends who said they had already voted, and the researchers found that the specific social nudge prompted more of those people to go to the polls. The study prompted some to suggest that Facebook had the power to sway election results.

Such testing raises fundamental questions. What types of experiments are so intrusive that they need prior consent or prompt disclosure? How do companies make sure that customers have a clear understanding of how their personal information might be used? Who even decides what the rules should be?

Existing federal rules governing research on human subjects, intended for medical research, generally require consent from those studied unless the potential for harm is minimal. But many social-science scholars say the federal rules never contemplated large-scale research on Internet users and provide inadequate guidance for it.

For Internet projects conducted by university researchers, institutional-review boards can be helpful in vetting projects.

Facebook don’t face such formal reviews.

New rules

Sinan Aral, a professor at MIT’s Sloan School of Management who has conducted large-scale social experiments with several tech companies, said any new rules must be carefully formulated.

“We need to understand how to think about these rules without chilling the research that has the promise of moving us miles and miles ahead of where we are today in understanding human populations,” he said. (Aral does some data analysis for The New York Times Co.)

Mary Gray, a senior researcher at Microsoft Research and associate professor at Indiana University’s Media School, who has worked extensively on ethics in social science, said that too often, researchers conducting digital experiments work in isolation with little outside guidance.

She and others at Microsoft Research spent the past two years setting up an ethics-advisory committee and training program for researchers in the company’s labs who are working with human subjects. She is now working with Hancock to bring such thinking to the broader research world.

“If everyone knew the right thing to do, we would never have anyone hurt,” she said. “We really don’t have a place where we can have these conversations.”

Gray advocates a simple litmus test for researchers: If you’re afraid to ask your subjects for their permission to conduct the research, there’s probably a deeper ethical issue that must be considered.

For Hancock, solutions could include an opt-in process for projects that involve big changes in an Internet user’s experience, and a debriefing system to inform users about smaller tests after the fact.

Companies won’t willingly participate in anything that limits their ability to innovate quickly, he said, so any process has to be “effective, lightweight, quick and accountable.”

While some would say the risks of the Facebook study were obvious, Hancock said the researchers didn’t realize that manipulating the news feed, even modestly, would make some people feel violated.

He learned otherwise from hundreds of anguished and angry emails he received after the work was published. “They said: ‘You can’t mess with my emotions. It’s like messing with me. It’s mind control.’ ”

Four weeks for 99 cents of unlimited digital access to The Seattle Times. Try it now!

Also in Business & Technology

News where, when and how you want it

Email Icon

Relive the magic

Relive the magic

Shop for unique souvenirs highlighting great sports moments in Seattle history.



The Seattle Times

The door is closed, but it's not locked.

Take a minute to subscribe and continue to enjoy The Seattle Times for as little as 99 cents a week.

Subscription options ►

Already a subscriber?

We've got good news for you. Unlimited content access is included with most subscriptions.

Subscriber login ►
The Seattle Times

To keep reading, you need a subscription upgrade.

We hope you have enjoyed your complimentary access. For unlimited access, please upgrade your digital subscription.

Call customer service at 1.800.542.0820 for assistance with your upgrade or questions about your subscriber status.

The Seattle Times

To keep reading, you need a subscription.

We hope you have enjoyed your complimentary access. Subscribe now for unlimited access!

Subscription options ►

Already a subscriber?

We've got good news for you. Unlimited content access is included with most subscriptions.

Activate Subscriber Account ►