Parties Square Off In a Database Duel

This fall, thousands of people such as Linda Houston -- who live in a county that Democrat Al Gore won by 4,156 votes in the 2000 presidential election -- will receive "customized" appeals from the parties, courtesy of the databases.

Those whom computer models have identified as, say, education voters may get a knock on their door from a teacher, who will talk up Sen. John F. Kerry's ideas about education. Senior citizens concerned about Medicare or Social Security might get a phone call on that topic from President Bush's volunteers, and perhaps campaign literature highlighting the president's views. Groups, such as the Democratic-leaning Emily's List and the conservative National Rifle Association, have their own database-driven efforts, piggybacking on the two major parties' electronic files in an effort to find and motivate voters.

Because of programs that sift and cross-reference reams of data, database jockeys are starting to discover some surprising behavioral nuggets in their info-mountains. By analyzing lists of Democratic donors and consumer data last year, Malchow found that people who live in households without a call-waiting feature on their phones are more likely than average to respond to a political fundraising pitch.

Malchow is not sure why, but he theorizes that people who do not have call waiting are older and have a more leisurely lifestyle -- the kind of people who might take a few moments to think about and contribute to a political cause.

Room for Error

The DNC's database team has used modeling programs to project the top issues for groups of voters based on common personal characteristics. For example, the DNC estimates that health care is the top priority of 940,000 people in Ohio. It has also projected where these people live among the state's 88 counties, providing a valuable road map for campaign advertising.

Laura Quinn, the DNC's technology guru, cautions that the identity of each "health care voter" is based on statistical probability -- that is, it is a likely identification, not an absolute one (for competitive reasons, the DNC will not detail the cluster of attributes that marks someone as a health care voter).

Predicting voters' thinking this way is hardly foolproof. As a rule, more information about a person helps improve the accuracy of assessing which beliefs they hold. "You could take all the obvious things about a person and still not screen out the important variable," said Doug Kelly, one of the architects of DataMart and the DNC's donor database, known as "Demzilla."

"I might live across the street from a guy. We're the same age, we have the same approximate house value, same family size, same education, maybe even the same minivan. But he's a Republican and I'm not," Kelly said.

Even so, Quinn said, this kind of "statistical oddsmaking" is more reliable than the broad assumptions made about voters before. "We're no longer just speaking about 'women voters' or 'minority voters,' " she said. "The closer we can get to the real circumstances of people's lives, the better. What is more telling about a person is not how they feel about President Bush, but how they live and what they say about themselves."

In other words, Houston's address is not just a household on a precinct map; in the database, Houston may show up as many things: a Caucasian, a woman, a mother, a grandmother, a homeowner, a Democrat, a resident of a mixed urban neighborhood of renters and homeowners that turns out for liberal candidates by a 2 to 1 margin.

A Complicated Task

Consumer marketers have been profiting from this kind of information for many years. Supermarket chains review data on purchase patterns, collected through "shopping clubs," for clues about how to "micro-target" new products to shoppers. Internet companies such as Amazon.com keep records on what customers bought to offer them deals on products they have shown interest in before.

But predicting which book or brand of breakfast cereal someone might buy is easier than figuring out how millions of people will vote -- or even whether they will vote -- several months before Election Day.

The task is complicated by the fact that accurate data on many voters are not readily available. Information in state voter files -- the foundation of the national databases -- varies by state. And every state protects the most basic and compelling political fact -- whom someone voted for. The only way to find out is to ask people directly, and that is an expensive and time-consuming job.

The parties are not even sure who is a Democrat or a Republican. Since only about a third of Ohio's voters are registered with a party, the vast majority of the state's electorate has no clearly marked partisan "trail."