September 18, 2006
Beer AND tacos

There's been a debate raging in baseball literature for some time now about the proper approach to talent evaluation. On the one hand you have the traditionalists, who believe that old-fashioned scouting methods are still the best way to tell if a given player will be a productive major leaguer or not. Numbers don't tell you everything, you have to know what the player is made of and see for yourself what he is capable of. On the other hand are the statheads, who argue that you can successfully project with a high degree of confidence how a player will do in the future based on how he has actually performed in the past. Your eyes can deceive you, and the stats can tell you things that may not be apparent on the surface.

A year or so ago, one of the writers for the Baseball Prospectus addressed this dilemma by saying it's like being asked to choose between beer and tacos. The correct answer, when someone asks you which of those things you want, is to say "Both, you fool!" Each one has its own merits, but together they complement each other, making the total package so much better than the individual components.

That's what I thought of as I read this Observer article about data mining and microtargeting versus traditional get-out-the-vote methods in Texas campaigns. Why must we choose? Why can't we do both?


The premise is simple: In modern politics, candidates for public office are products that must be sold to voters. As in any corporate marketing campaign, it helps immensely to know who the consumers are. In data mining, a campaign adopts the techniques of corporate marketing to compile as much personal and demographic information about as many voters as possible. Voter targeting takes into account everything from which primaries you’ve voted in, to your marriage status, to how long you commute to work, to how often you use FedEx. Each bit of information yields a clue about a voter and voting tendencies. The campaign runs this information through a series of complex, computer-driven statistical analyses. The end result is a surprisingly accurate portrait of who lives in the district, which voters the campaign should pitch to, and how to reach them in a media-saturated culture.

Republicans have used these techniques in recent election cycles to successfully target and turn out conservative religious voters. Democrats in Texas, however, have been slow to catch on. Yet the Democratic Party’s biggest successes in Statehouse races the past two years - Mark Strama in Austin, Hubert Vo in Houston, David Leibowitz in San Antonio, and Howard - all used data mining to some degree.

Not everyone is enamored with the technique. Some of the top Democratic consultants and campaign contributors in Texas believe the effects of voter targeting are overblown, just hype generated by a handful of self-promoters that distracts from more important elements of campaigning. A debate has begun within the party about the effectiveness of targeting voters in small legislative races. In particular, the influential Texas Trial Lawyers Association, the deepest well for Democratic campaign money, has generally refused to provide major support to campaigns that use data mining. Howard, for instance, received hardly any money from trial lawyers in her race. The result is that voter targeting has been slow to gain traction among Democratic candidates. Only about a dozen around the state are using targeting this election cycle.


It's scouting versus sabermetrics all over again. This isn't, or at least it shouldn't be, an either-or choice. Certainly the Republicans, who have a huge head start on the Dems in using this technique nationally, don't see it as such. Of course candidates and messages matter, and of course every effort need to be made to make sure our voters turn out. But why not look for "our voters" in places where they're a distinct minority? Why not look for people who at first glance might not seem like our voters?

I'm going to come back to this article later, because I think there's a lot of meat there for substantive discussion, but for now just read it and say to yourself "What we want is beer AND tacos."

Posted by Charles Kuffner on September 18, 2006 to Show Business for Ugly People | TrackBack
Comments

You ask "why not both?" Let's assume you have limited resources. Using datamining techniques, you can determine 20 voter segments that each will respond to a different degree to your marketing (whatever it may be). 11 segments will never vote for your candidate because they are Republicans. 4 segments will always vote for your candidate because they are hardcore Democrats who always vote. That leaves 5 segments. Now the response rate (i.e., chance that a person in a given segment will vote if they see your ads) varies for each segment between 10% and 90%.

If you have a limited budget, it may make sense to target only those segments where the response rate is greater than 50%. Spending money on a more general GOTV message that goes out to everyone may not be the best way to gurantee the highest turnout for your candidate.

I doubt you have situations quite that cut-and-dried, but one can see how you might--and in that situation, you woul dhave to choose between the data-mining method and the traditional GOTV methods.

Posted by: RWB on September 18, 2006 1:03 PM