Thomas Edsall at the Huffington Post had a rather interesting piece recently about the Obama campaign’s use of consumer targeting strategies in its voter canvasing efforts:

When volunteers sign up to campaign for Barack Obama, they enter the world of e-politics – guided to a web site with a carefully culled, computer-generated list of people who live nearby. The volunteer is instructed to pick 25 people from the list, preferably people he or she knows – or, better yet, actual friends.

The names have been chosen by slicing and dicing a massive agglomeration of government and commercial data, using datamining technologies which identify the magazines, cars, and cookware specific individuals buy; how often they turn up at the polls; the value of their homes; their membership in organizations running the gamut from the NRA to Planned Parenthood; information customers volunteer when they fill out warranties; shopping histories –Target, Whole Foods,
Ethan Allen, Sports Authority; and on, and on, and on.

When cross-referenced with the results of public opinion surveys and census information, all these pieces of data can ultimately produce demographic-consumer portraits of voters ranging from guaranteed Obama to locks for McCain and multiple shades of grey in between.

Consultants who specialize in datamining contend that, for a campaign willing to pay, they can tell with 90+ percent accuracy whether an individual voter is for Barack Obama, John McCain or, most importantly, on the fence.

This is an enormous component of the Obama campaign’s ground game–otherwise an already unprecendented effort of Democratic local mobilization. It should be a familiar tactic to anyone who uses Amazon.com. Their recommendations are based on a similar approach, but instead of relying on the broad range of information used to predict political disposition, the Amazon engine just uses previous purchase information.

There’s something eerie about both examples. It can be unsettling to think one’s decisions can be quantitatively determined so easily and effectively. Most people want to think of themselves as in control of their own rather unique set of preferences, especially in the case of politics. The idea that how one votes can be determined by other unrelated facts seems to demean the democratic process or at least the participation of that voter. It smacks of a kind of profiling–assuming that African-American from Chicago is going to vote for Obama and that an elderly veteran somewhere in the heartland is going to vote for McCain is, in a basic way, rather prejudiced.

But it’s just playing the odds. Abstracting away from the meaningful content of the data, data mining in these examples is basically about finding correlations within a large set of cases in order to make predictions about other cases. But that dry versions of things doesn’t capture the potential human response to being told that one’s behavior can be reduced to a (probabilistic) function of one’s other attributes. The most objectionable case of this is racial profiling for crime, in which one’s skin color becomes a liability based on the odds generated by the past actions of others. Recently, terrorist profiling based on more sophisticated data analysis–rather than one reductionist variable about identity–has sparked debate about what counts as a legitimate law enforcement practice and what is a way of biasing state action against particular individuals or groups.

But there’s one really critical difference between all of that and what the Obama campaign is doing. Amazon’s recommendation tool, police profiling, and many things in between are interested in what an individual probably will do. The political application of these techniques is more interested in what an individual probably won’t do. The Obama campaign is looking for converts, not just trying to identify existing adherents. This throws a wrench into the entire picture because it relies on a basic assumption that predicted behaviors can be influenced. It’s a reminder that the tendencies data mining strategies can identify are far from being carved in stone. And, convincing one person who the data says "should" do one thing to do the other retroactively changes the data set and makes subsequent predictions different.

Though the Obama campaign picked up this approach to a large extent from consumer targeting, this is an idea that marketing could take back from them. As social media makes more and more data available, there’s potential to do more than just target–it’s possible to start thinking in a data-driven way about different types of users and consumers and how they might interact. The Obama campaign identified effective evangalists and encouraged them to target identified potential converts that they would be most likely to swing. This is an organized word-of-mouth strategy armed with the information resources to lend it amazing precision. It’s a big effort, and going door to door isn’t practical in most cases, but this basic model surely has applications elsewhere.