STEVE EMBER: Welcome to THIS IS AMERICA. I'm Steve Ember.
BARBARA KLEIN: And I'm Barbara Klein. This week on our program, we look at how political campaigns use data mining to try to predict how people will vote.
(MUSIC)
STEVE EMBER: Andrew Drechsler is a data miner. He works with millions of pieces of information. He looks for details about people -- what neighborhood they live in, what they buy, what they like to do on their weekends.
Mr. Drechsler's job is to collect enough details to form a sense of how people think. That also makes him a little like a storyteller.
ANDREW DRECHSLER: "One art that we've sort of been working on perfecting over the years is telling a story with information."
Data miners look for connections in the ways that people behave. What they find can help stores decide which items to put closer together. It can help advertisers target their messages. And it can help political campaigns know where to look for votes.
BARBARA KLEIN: The use of data mining in politics is nothing new. Michael Kazin, a Georgetown University history professor, gives an example in his book "A Godly Hero." The example involves William Jennings Bryan.
Bryan ran for president of the United States in eighteen ninety-six, nineteen hundred and again eight years later. His supporters wrote him letters -- thousands and thousands of letters. His brother Charles read them all.
Professor Kazin writes: "Charles jotted down every bit of information he could find about a correspondent: party affiliation, job, religion, even income. He updated the file constantly for the next thirty years and used it to send out regular mailings to the Bryan network. The index contained some 200,000 names in 1897 and grew to half a million by 1912."
STEVE EMBER: In other words, Charles was data mining. He was trying to learn what kind of people supported his brother for president. Then he used the data to guide the campaign.
Did it work well? Apparently not well enough. Bryan lost every time.
These days campaigns hire professional data miners. They usually work for one party, either the Democrats or the Republicans.
Andrew Drechsler is vice president of Strategic Telemetry, a company in Washington that works with Democrats. In two thousand eight, Strategic Telemetry worked for Barack Obama's presidential campaign. Mr. Drechsler says his company began by looking for the easiest data to find.
ANDREW DRECHSLER: "To be quite honest, when we first started working for Obama we were only using census data."
BARBARA KLEIN: The Census Bureau provides basic information about an area. It could be a small neighborhood or a whole state. The population data includes things like racial and ethnic percentages, average income, education levels and how many people are married or have children.
Data miners then match this information with public voting records of people who live in that area. Mr. Drechsler says this information is known as the voter file.
ANDREW DRECHSLER: "When we get the voter file it's usually first name, last name, address, phone number, and then vote history, is the typical information in the voter file."
Voter files will not say who a person voted for, but they will say whether or not the person voted.
Data miners also buy information from companies.
ANDREW DRECHSLER: "You know, what type of magazine do people read, what kind of car do they drive, do they rent, do they own, do they have pets, what sort of pet do they have?”
Each detail is a data point, and by the time data miners are finished, they can know a lot about a person.
ANDREW DRECHSLER: "So we will have close to a thousand data points on the voters."
(MUSIC)
BARBARA KLEIN: Data mining sounded like fun, so we decided to try it ourselves. We went out on the street here in Washington and found a woman sitting on a bench. We asked her age.
WOMAN: "Thirty-three."
Where she lives.
WOMAN: "Maryland."
What she buys at the grocery store.
WOMAN: "I always buy this herbed mixed green salad."
And a few other questions.
REPORTER: "Do you have any pets?"
WOMAN: "I don’t."
REPORTER: "And you said you’re married?"
WOMAN: "Yes."
REPORTER: "Kids?"
WOMAN: "No kids."
REPORTER: "Do you have a car?"
WOMAN: "Yes."
REPORTER: "What kind?"
REPORTER: "It’s a Lexus."
And now for our big question. Based on this information, what would a political data miner think?
REPORTER: "Do you think they think that you’re a Democrat or a Republican?"
WOMAN: "Based on what I said? Um, wow. Based on what I said, I might be a Republican."
REPORTER: "And are you, can you tell me?"
WOMAN: "No, I’m not."
STEVE EMBER: So why does she say a data miner might think she was a Republican? Is it because she drives a Lexus, an expensive car? Judging people's politics by the kind of cars they drive may not be meaningless.
Americans in the top twenty percent of income are more likely to vote for Republicans -- more likely, but not always. Andrew Drechsler says data mining is not as simple as finding one important piece of information.
ANDREW DRECHSLER: "There's not one silver bullet. There’s not, 'If we can just find the cat owners or the bourbon drinkers, we can win this election.'"
Those examples may sound silly. But look at it another way. Why would a candidate for local dog catcher want to waste time and money reaching out to cat owners?
BARBARA KLEIN: Andrew Drechsler says instead of focusing on one person or one data point, data miners look for patterns across large groups. They are not trying to learn everything about you. They are looking for ten thousand people like you. To do that, they use math, maps and machines.
Mr. Drechsler's office is so small, if he sneezed, his two employees would probably have to go home sick.
ANDREW DRECHSLER: "Our office is relatively small. But, it’s, in our server room we have power -- our IT director one time said we have enough power to power a company of three thousand employees.”
Let’s say there are four million people in a state -- a state like Kentucky. Mr. Drechsler and his employees cannot talk to all of them. But they can use an automated telephone service to call ten thousand of them and ask who they plan to vote for. The data miners add those responses to the other data points they have about those people.
From there, they use their computer power to crunch the numbers. The goal is to predict how the people they did not talk to -- the other three million nine hundred ninety thousand -- will vote.
ANDREW DRECHSLER: "Whatever question that we ask we can model and show the likelihood of every voter in the state supporting -- er, responding to that question as if we did talk to everybody.
STEVE EMBER: So how do campaigns use that information? Mr. Drechsler stands up and gets a framed map off the wall.
ANDREW DRECHSLER: "Our offices are no-thrills offices, but this is just a -- we do have a couple maps and campaigns that we’re particularly proud of."
The map shows Iowa in the Midwest, the American heartland. Iowa traditionally votes at the beginning of the presidential nominating season.
Some areas of the map are colored in purple. These are the areas that Mr. Drechsler's team thought would support Barack Obama. The gray areas on the map are those they thought would probably not support him. Campaign officials used these maps to help them decide where to advertise and send volunteers.
ANDREW DRECHSLER: "This is just a good example of a visual that we often give to campaigns just showing -- mapping -- where the support is, generally speaking. A lot of times campaigns know that, but it just helps illustrate where they need to focus and where, frankly, they don’t need to focus."
Dianne Bystrom at Iowa State University studies political campaigns. She says that in Iowa, the Obama campaign used data to target people who did not usually vote in the political meetings called caucuses. By bringing in new voters, she says, Mr. Obama was able to win that state.
DIANNE BYSTROM: "And that's what really won him the caucuses. He turned out more than the traditional Democratic caucus-going base in the state of Iowa."
BARBARA KLEIN: Ms. Bystrom is director of the Carrie Chapman Catt Center for Women and Politics at Iowa State. She points to the way Ronald Reagan's campaign used data in nineteen eighty when he ran for president. She says the data helped his Republican team target messages to women based on the issues most important to them.
That year, women voted at a greater rate than men for the first time. That change helped close the “gender gap” and send Ronald Reagan to the White House. Professor Bystrom says stories like this show the value of data, even if some people might be concerned about personal privacy.
DIANNE BYSTROM: "For the campaigns, I think they're very powerful. For the voter, and the consumer, I think as technology gets more sophisticated, it gets a little creepy."
BARBARA KLEIN: Daniel Kreiss is an assistant professor at the University of North Carolina.
Professor Kreiss says campaigns will increasingly combine information about what people do online and offline in their everyday life.
DANIEL KREISS: "What I think is starting to happen now and is genuinely new -- although there were the first steps in this direction in two thousand eight -- is the increasing ability of campaigns now to sync their general voter databases with online user data that they’re getting from other sources."
STEVE EMBER: Privacy is not the only concern that some people have about political data mining. Professor Kreiss says some scholars worry that it will harm democracy. They say highly targeted campaigning means fewer people will hear a candidate’s message.
Andrew Drechsler, the data miner in Washington, thinks just the opposite.
ANDREW DRECHSLER: "I would take that argument and turn it around. You are talking to those who are most interested and most likely to participate and those who should really have the information."
(MUSIC)
BARBARA KLEIN: Our program was written by Kelly Nuxoll. We welcome your comments at voaspecialenglish.com. You can also find a transcript and MP3 of our program and a PDF version for e-readers. I'm Barbara Klein.
STEVE EMBER: And I'm Steve Ember. Join us again next week for THIS IS AMERICA in VOA Special English.
