Analysts target individual voters

In America, the digital age and data mining mean that someone somewhere knows how you're going to vote, probably before you do, writes Kiwi pollster Stephen Mills.

In August 1955 science fiction writer Isaac Asimov published a short story, Franchise, in which American elections were decided by a single perfectly representative voter selected by the giant computer Multivac.

John Anzalone, a Democratic pollster, in a discussion at his party's convention almost echoed Asimov when he claimed "that the undecided voter is so small and targeting is so [advanced] that it really does come down to this one woman voter in the Columbus [Ohio] media market who works part time, has two kids and likes merlot. It gets to a point where a billion dollars is going to be spent on just this one woman."

It is not, of course, one voter that will decide the 2012 Presidency but it is not all voters either. The contest will be decided by around 1.5 million persuadable voters in the nine swing states.

And through ever- improving data analytics the parties know a lot about them.

Data analysts, engaged in micro-targeting and modelling, are now key players in the Obama and Romney campaigns. Perhaps uniquely in campaign history they do not come armed with their own opinions but rely totally on the numbers. This cycle they are enabling parties, rather than broadly taking aim at stereotype targets such as soccer mums or angry white men, to take dead aim at individual voters.

This advance is getting a lot of media attention and has an air of mystique.

Frank Luntz, a famous Republican pollster, when interviewed waved at a woman at the next table in the cafe and said he can tell who she votes for through her being at this cafe, wearing grey and where she shops. He also outlined how internet searches make voters whose names and addresses are unknown targetable. They are "anonymous in plain sight".

What the analysts are doing is merging data from a variety of sources to enhance voter files. This data is crunched in the first instance to produce two scores for voters - one predicting turnout and one propensity to vote Republican or Democrat.

The sources used vary by state but include public records of whether individuals voted in primaries, voter registration, information purchased from big data companies and gleaned from the "national cookie pool" which tracks searches and website visits.

The information held by the database companies is comprehensive, drawing from retail loyalty programmes, credit card purchases, any digital transactions, smartphone use and many other sources.

Acxiom, based in Arkansas, was the most frequently mentioned company . As one senior Washington communications consultant put it, "they know more about me than I know about myself".

Added to this is the more personal information gathered by party canvassers. As one senior Obama strategist explained data analysis can go only so far. They can find out if the most salient issue for say a 65-year- old grandmother is the special needs of a grandchild only by "actually talking to her".

In the first place this enables highly efficient GOTV (get out the vote) operations. Instead of relying on the law of averages and sending resources into areas of, say, known high partisan voting, parties can go straight to individual households. It means the parties can ignore all those who have high turnout and high partisan scores and concentrate on those who are likely to vote their way but may not turn out.

During the Democrat convention discussion Celinda Lake joked there might be scheduling problems. "They know the doors they are going to knock on in Ohio today and frankly they are going to have to schedule around each other because the Romney knockers are going to be right in front of the Obama knockers."

The data analysis also enables parties to identify their voters among hostile demographics. As explained by Jim St George of campaign software specialist NGP VAN, Republicans can find the small number of African Americans voting their way and Democrats can locate the shrinking number of rural southern white Democrats. If no information is available on voting records, the analysis of all other information can be used to make "best guesses" at a voter's likely behaviour and the issues they are most concerned about.

The enhanced voter files also provide direction for precisely targeted communications and for party canvassers on what issues will move the persuadable voters into the right column.

Talking to a number of political consultants there is an air of scepticism about some of the big claims made for data analytics.

Some may have had at least subliminal commercial motives but not David Radloff of Clarity Campaigns who works on the data analysis frontline. He stresses that "the complex analysis of available data is key to producing accurate models" of voter turnout and choice.

But he was keen to debunk the media's highlighting of specific aspects of consumer shopping habits (such as drink preferences and car ownership) as predictors of voting, suggesting it attracts attention "because it's sexy and makes a good talking point".

He says those types of data are much less important than "the data generated by campaigns and political parties themselves doing the hard work on the ground, and now online, talking to voters and recording which candidate they support".

Despite Radloff's caution on current media hype, there is no doubt data analytics will continue to improve and, unless American privacy laws change, Asimov's scenario may become theoretically viable. It would save billions of dollars.

Stephen Mills is the executive director of Wellington-based UMR Research. He is in the US, following the presidential election campaign.

The Press