Electoral stakes high in data mining
Does Big Data threaten to play a Big Brother-like role in American electoral politics? Or are concerns about the micro-targeting of voters that arose during the 2012 campaign making a mountain out of a molehill -- albeit a very busy and well-financed one?
That was the stark range of opinion voiced last week at the University of Pennsylvania's Annenberg School for Communication, where more than 30 data experts, scholars, and political operatives gathered to discuss the largely invisible ways Big Data tools are altering politics.
During last year's presidential campaign, conference organizer Joseph Turow, an Annenberg professor, found that 86 percent of Americans did not want to see "political advertising tailored to your interests" -- a common description of "behaviorally targeted" ads that have stirred controversy not for their content, but because of how they find their way to your computer or smartphone screen.
For advertising cars or travel, data miners and ad networks don't have to know exactly who you are. But campaigns have other needs -- among them to identify supporters, motivate them to get to the polls and encourage donations and volunteering.
And campaigns have access to other kinds of data, including voter files and off-line data, such as purchases or donations tied to a credit card or a frequent-shopper program.
Turow, who cosponsored last week's conference with University of North Carolina professor Daniel Kreiss, already knew that many Americans were skeptical about targeted ads. But he wanted to know how data mining has changed campaigns, and what the mix of Big Data and politics portends.
"We're only at the beginning of this era," Turow said.
There was anything but consensus, though the split wasn't partisan. Republican and Democratic campaign operatives largely agreed on one thing: that they weren't all that great, at least so far, in using the data at hand.
"The challenge in voter modeling is that people don't take that action very often," said Rayid Ghani, chief data scientist in last year's Obama campaign. He said rare decisions, such as what kind of house somebody will buy, are very tough to predict.
Still, the stakes in campaigns are enormous -- particularly in America's winner-take-all elections, where a swing of a few hundred thousand votes in a handful of states is often enough to tip a national election, said Zeynep Tufekci, a fellow at Princeton's Center for Information Technology Policy.
"The campaigns just keep saying this is business as usual -- there's nothing to see," said Tufekci, also a UNC professor. "My concerns are about where this is going."
One question raised Friday centered on transparency -- and on whether voters would be reassured or repelled if they knew they were targeted, say, by knowledge that they read a particular magazine.
Ethan Roeder, the Obama campaign's data director, said the question missed the point both in how much campaigns might know and its value.
"The way that that conversation would actually go is, 'So we see from the voter file that you're a male and that you're 49 years old, and there's a 60 percent chance that you're a Republican because of your voting history. And we've also noticed that in your neighborhood the median income is $117,000, and there's an education level around college age. We also happened to notice . . . we don't know why, but people that are close to the lake in your area tend to be more supportive of the environmental issues than other folks. And there's this environmental subscription that we also got from a consumer data set that we bought. We don't know if we trust it or not, but we're using it as sort of an indicator just at a small percentage,' " he said.
After the conference, Roeder told me: "I don't think campaigns want to keep that information from individuals. I just don't think there's a good way to share it."
Especially not when, from both parties' perspectives, the overriding goal is to win.