Roughly 139 million of Brazil’s 210 million people use the internet on a monthly basis. Of those, 108 million are on Facebook.

As a point of reference, just 43 million people in Brazil watched the World Cup in 2014.

With that kind of ubiquity comes the ability to spread messages rapidly. This is, of course, the basis of Facebook’s business model, but it is also a power that Facebook intends to use for the public good.

In fact, “we really believe that Facebook data can help people make informed policy decisions that would help save lives,” said Molly Jackman, public policy research manager at Facebook. To do that, however, Facebook has to focus on doing research ethically and responsibly and in a privacy-preserving way.

The organization got that opportunity in recent work with Unicef. The humanitarian organization wanted to educate Brazilians on how to protect themselves against the Zika virus, which can particularly affect unborn children, resulting in birth defects affecting brain development. What messages would resonate? Who’s talking about Zika and what are they saying?

That’s something Facebook can help with. “Because of the scale at which Facebook operates,” said Jackman, “we’re able to analyze aggregated Facebook posts to see how people are talking about Zika and what they were thinking. … It’s a responsibility to use data for good, but also to use it in a way that protects the people we’re trying to serve.”

"It’s a responsibility to use data for good, but also to use it in a way that protects the people we’re trying to serve." —Molly Jackman, Facebook

For example, Facebook runs projects like these through two separate review processes. First is a privacy review, to make sure the data is used in a way where individuals can’t be reidentified. “This makes sure,” Jackman said, “the data is stripped of PII and only reported in the aggregate.”

Second is a research review process, triggered particularly with sensitive subjects like health. “This provided guidance on privacy,” as well, Jackman said, “but also just on how to do the research in a way that was ethical and responsible, like a university internal review board.”

In the end, “We found the privacy protections we required were compatible with Unicef’s goals,” Jackman said. “They wanted insights and not individual-level data, so they could take action.”

For example, the data uncovered that 58 percent of those talking about Zika were men, which is fairly counter-intuitive, given that child-bearing women are most affected by the disease in terms of noticeable symptoms. Also notable was that “mosquito” was the most common co-occuring word with Zika. This allowed Unicef to tailor its message. “We were also able,” said Jackman, “to show at a state level the number of conversations in each geography, along with the predominant device types, so they could understand where they were talking about Zika” and cater ads in that direction.

Ultimately, Unicef was able to move the needle. Eighty-two percent of those who saw the campaign reported taking action to avoid zika, as opposed to 79 percent of those who did not see the campaign at all. That is a better result than 90 percent of studies conducted on Facebook ad performance, according to Nielsen Brand Effect norms. The campaign reached more than four million people in the first few hours of its launch. “And we saw strong evidence of action in the comments,” said Jackman. “People saying, ‘Let’s share this message. The more information we have, the easier it is to fight against the mosquito.’”

This is the promise of big data, Jackman said, the ability to uncover trends and insights that would otherwise require expensive and time-consuming surveying, which would probably also be less accurate. Obviously, privacy is a concern, particularly when dealing with matters of health and child-rearing, but “I think the really important point here,” said Jackman, “is that if you went and posted about Zika tomorrow, Unicef wouldn’t be particularly interested. That’s not super helpful. They want to know how people in general are posting about this.”

Editor's note: This piece was edited to correct a misunderstanding of data surrounding campaign effectiveness.