US election: big data triumph or epic pandering fail?

Peter Aldhous, San Francisco bureau chief

(Image: Sipa Press/Rex Features)

The 2012 US presidential election is a contest powered by geeks. In the age of big data, the Obama and Romney campaigns have each employed teams of analysts, developing algorithms to slice and dice the electorate into segments that can be targeted with specific messages intended to resonate with focused subsets of voters.

But according to the latest research, this "microtargeting" may do the candidates more harm than good. Targeted messages seem to offer little advantage over more general appeals. And if the analysts get it wrong and a campaign's message goes to a member of the wrong group, the candidate is likely to lose votes.

Meanwhile, the first poll to investigate reactions to the use of targeted online political ads has found that the US public deeply dislikes the practice. So in trying to appeal to narrow groups, the candidates could simply be painting themselves as cynical panderers - a reputation that has already been mocked mercilessly in a recurring segment on the satirical television programme The Daily Show.

Campaigns start with public data on voters' registration, gender and past turnout in elections. This may include party affiliation, and in some states also information on race. They then purchase consumer data revealing what cars people own, what magazines they subscribe to, what websites they visit, and so on.

Then comes the clever part: from people who have been polled for their opinions, or have provided information about themselves to campaign websites, analysts derive algorithms to predict from the voter and consumer data the interests and opinions of the majority of people who haven't been contacted. This produces a database that divides the electorate into groups that can be primed with tailored messages - delivered by email, regular mail, or online ads triggered by "cookies" on their computers.

To investigate the effectiveness of such tactics, Eitan Hersh of Yale University and Brian Schaffner of the University of Massachusetts, Amherst, polled a representative sample of voters about a fictional political candidate, asking them how likely they were to vote for him on a scale of 0 to 100 after being shown an example of campaign mail.

These mailings either contained general pledges to work on behalf of constituents or mentioned particular groups, such as Latinos, gun owners, union members or born-again Christians.

The targeted messages had little or no effect on the voting intentions of the intended recipients, but among members of other groups, they caused a marked decline in support - dropping by some 20 to 25 points on the scale.

Voters seem unmoved by candidates' efforts to appeal directly to them but pay more attention to pledges made to other groups, Schaffner concludes. "People think those claims - the ones they see by accident - are the ones the candidate is really going to act on," he says.

Donald Green, a political scientist at Columbia University in New York, cautions that studies on hypothetical candidates may not reflect what happens in the real world. And with the campaigns keeping quiet about the details of their microtargeting, it's possible that their own data may be telling them a more encouraging story.

Still, even the most sophisticated number crunching is bound to result in some errors in assigning voters to target groups, says Schaffner - who next aims to collaborate with actual campaigns to study the effectiveness of microtargeting in real elections.

Whether or not the technique works, it's clear that the public doesn't like the idea of being served tailored online political advertisements. Researchers led by Joseph Turow and Michael Delli Carpini of the University of Pennsylvania in Philadelphia ran a poll that found 86 per cent of respondents don't want to see such ads - compared with 61 per cent who rejected similarly tailored ads for commercial products or services.

What's more, 64 per cent said that they would be less likely to vote for a candidate if they learned that the campaign was buying information about people's online activities, and then sending tailored messages based on this information - a common practice in the current presidential election.

Turow concedes that such opinions won't deter campaigns, if they believe such ads will be effective. After all, people say they dislike TV "attack ads" and other forms of negative campaigning, but these nevertheless seem to work. However, the poll raises the question of whether aspects of political data mining and microtargeting should be restricted, Turow suggests.

Campaigns are sure to argue that such activities are a form of free speech, protected under the first amendment to the US constitution. But Turow says that legal experts have yet to reach a consensus on this point. "It's something that's going to need a lot of discussion," he says.

Source: http://feeds.newscientist.com/c/749/f/10897/s/2409f79d/l/0L0Snewscientist0N0Cblogs0Conepercent0C20A120C10A0Cus0Eelection0Emicrotargeting0Bhtml0DDCMP0FOTC0Erss0Gnsref0Fonline0Enews/story01.htm

safety not guaranteed lifehouse al gore la dodgers lawrence o donnell magic johnson jetblue pilot