Obama, McCain, and Data-Driven Campaigning

by Jesse Farmer on Wednesday, October 29, 2008

On Monday Slate published an article about Obama's text messaging strategy (h/t Andrew Chen) and how it compared to the traditional robo-calling strategy. Politics is getting more quantitative every year and it's great to see the campaigns using techniques like A/B testing to determine what works and doesn't work in political messaging.

A Channel to Voters

Every campaign has several channels to voters: person-to-person contact, phone calls, mailers, etc. You can attach metrics like "dollars per vote" or "votes per contact" to each of these channels, and the campaigns do.

From the Slate article,

Robo-calls are the pyrotechnics of politics: They create a big disturbance, but they don't have a prolonged effect. Numerous studies of robo-call campaigns show that they're ineffective both as tools of mobilization and persuasion — they don't convince voters to go to the polls (or to stay away), and they don't change people's minds about which way to vote. So why do campaigns run robo-calls? Because they're cheap and easy. Telemarketing firms charge politicians between 2 and 5 cents per completed robo-call; that's as low as $20,000 to reach 1 million voters right in their homes.

The campaigns also measure votes-per-dollar. Using the Green and Gerber numbers we get this table:

Voter Contact Methods
Channel Votes-per-contact Dollars-per-vote
Canvassing 1/14 $29
Phonebanking 1/38 $38
Telemarketing 1/180 ???
Mailers (non-partisan) 1/200 ???

Partisan mailers are even farther down the list, and leaflets, emails, and robo-calls showed "no discernible effect" on the electorate. Obama's breakthrough is using text messaging which costs an astonishing $1.50 per vote.

Hierarchy of Personalization and Maturity

The hierarchy is pretty clear: the more personal the contact the more effective it is. Person-to-person contact will always be personal, obviously, and until this campaign text messaging was something you did with your friends and family.

Will this change? Email marketing is less effective nowadays because everyone is used to spam. Text messaging has worked brilliantly for Obama this campaign but as the technique becomes more common there's no way to know if, four years from now, people will still respond the same way.

Bombarding a communication channel with impersonal messages makes the medium itself less personal and therefore less effective.

A/B Testing

How did Green and Gerber calculate the effectiveness of voter contact methods? By carefully measuring the results of their A/B test and using hypothesis testing to determine whether there were actually differences between the control and test groups:

Rather than merely theorizing about how campaigns might get people to vote, Green, Gerber, and their colleagues favor randomized field experiments to test how different techniques work during real elections. Their method has much in common with double-blind pharmaceutical studies: With the cooperation of political campaigns (often at the state and local level), researchers randomly divide voters into two categories, a treatment group and a control group. They subject the treatment group to a given tactic: robo-calls, e-mail, direct mail, door-to-door canvassing, etc. Then they use statistical analysis to determine whether voters in the treatment group behaved differently from voters in the control group.

Personal Experience

I'm an Obama supporter and have been volunteering on-and-of since early this year. Working on the California primary I had a chance to see the campaign's data-driven approach first-hand.

Every state is broken down to the precinct using a system called VoteBuilder. For a village the area might be the whole town, but for a city it could be as small as a few city blocks. Precinct captains or other field workers can slice up the city using queries like "get me a list of voters who are not strong supporters of either candidate and print off their names in walking order."

What's more, Obama's campaign is as much about his brand as it is about finding the votes it knows are there, even if they're in traditionally Republican areas. I canvassed my hometown in Northern Michigan, for example, an area classified as a "persuasion area."

How did the Obama campaign know that this small segment of typically Republican Northern Michigan was persuadable?

It's simple, really: data plus an empirical mindset.

Lessons from Republicans

In a lot of ways the Obama campaign has learned from the Republicans. In the past Republicans have won in large part because they applied their background in quantitative direct marketing to voter mobilization.

And if Obama wins it will be in large part because he absorbed and modernized the data-driven techniques Republicans have been using since the early 90's.