How the Obama campaign used randomised control trials to micro-target its way to victory

by Stephen Tall on November 12, 2012

Identifying and targeting the undecideds: that’s the key to winning elections, isn’t it? Find the people who sit bang-in-the-middle of the political spectrum and bombard them with messages that persuade them to back your side, yes?

Those are two underpinning assumptions of modern political campaigning which have been challenged by the Obama campaign’s use of randomised control trials, as revealed in Sasha Issenberg’s book The Victory Lab. A couple of weeks ago, Sasha wrote an in-depth article for Slate profiling the profound impact of RCTs on the ways in which potential voters were categorised and communicated with: Obama’s Secret Weapon: Democrats have a massive advantage in targeting and persuading voters.

As the scale of Obama’s popular victory becomes apparent — a three million vote advantage isn’t wafer-thin — along with the failure of the Romney campaign’s own polling to predict what would happen, it’s worth revisiting his post to understand some of the key lessons of how the Democrats campaigned smarter.

At its core is the ability to understand — through constant testing and refinement — how to identify the voters who, regardless of their demographics, are persuadable; categorising them according to the issues which matter most to them; and developing micro-targeted messages which will speak directly to those issues.

The UK is — unsurprisingly, given how much more money the Obama campaign has to spend — a long way behind, fixated like the Republicans on Mosaic-style data classification to try and predict voters’ allegiances and touch-point issues. They’re a start, and better than nothing — but how much more powerful it is to put theory into practice in the field in order to test, learn and refine.

It’s a long article, so here are 5 key points…

1) The Democrats have a vital competitive edge: they use RCTs, the Republicans don’t

The most important methodological and conceptual breakthroughs in recent years have originated in the academy, specifically through insights from behavioral psychology and the use of field experiments. Since 2004, myriad advocacy groups and consulting firms on the left have joined forces and launched a series of nominally for-profit private research institutions devoted to campaign tactics. The most impressive among them, the Analyst Institute, was created to link the growing supply of academics interested in running randomized-control trials to measure the efficacy of political communication with the demand of left-wing institutions eager for empirical methods to test their programs. These partnerships have birthed a generation of political professionals—many baptized in the unprecedented pools of data collected by Obama’s 2008 effort—at ease with both campaign fieldwork and the techniques of the social-science academy.

2) How RCTs help the campaign to identify — and shift — voters’ views

The Analyst Institute’s centrality in the left’s research culture has enshrined the use of randomized field experiments as the best tool for measuring what actually moves voters. And the biggest conceptual contribution this body of experimental work has made is to cleanly separate what a voter does in election season into two discrete phases: choosing among candidates and deciding whether to vote. Experiments have shown that giving voters more information about candidates or issues or the stakes of the election does little to adjust their likelihood of casting a ballot. To budge a nonvoter out of complacency, campaigns have learned, they have to use psychological techniques focused on getting someone to do something he or she is not used to doing. There’s one set of tools for changing opinions, and another for modifying behavior.

3) Testing: helping you identify your persuadables

The people who first developed the microtargeting models used in persuasion had assumed, like the rest of us, that voters in the center are the most up for grabs. But in 2006, EMILY’s List ran a series of persuasion experiments that raised doubts about this assumption. The Democratic women’s group sent out mailers on behalf of female gubernatorial candidates in Michigan and Washington, then polled across the entire universe of recipients to gauge the impact of the messages. The voters who’d been assessed as sitting closest to the middle of the road barely budged. In fact, there was significantly more movement among those who were projected to be leaning toward the Republican candidate than among those whose mid-range scores situated them evenly between the two poles. … Whatever those support scores were measuring, it wasn’t exactly susceptibility to persuasion.

4) Testing: helping you categorise your voters

Some voters had simply remained relatively anonymous, with little data about them on file to push them toward one candidate or another, while others existed in demographic categories that did not contribute meaningfully to predictions about their politics. (One example Strauss uses is voters in their 30s, “an age range that neither leans Democrat nor Republican.”) In both of these cases, the middle-range scores gave a misleading indication that a voter was persuadable. … There was a third case, though, in which there could be lots of data available about an individual voter that effectively cancels itself out. These situation resembled the predicament that political scientists have long defined as cross-pressure, where a voter’s choice is complicated by conflicting aspects of his or her identity—the African-American who’s also a Mormon, to take one example. In these cases, a mid-range score seems to quantify precisely the type of ambivalence that makes for a good persuasion target. “We certainly want to talk to voters who are cross-pressured,” Strauss has written.

5) Testing: helping you communicate your micro-targeted messages to these voters

In June, the DCCC ran what Strauss called a “persuasion-microtargeting experiment,” to test Democratic messages on voters in the field. Experiments found pockets of voters who moved in their direction in response to particular appeals: After hearing the party’s message on Medicare, men over the age of 65 increased their support for a generic Democratic congressional candidate three points more than the broader population. The DCCC could build a profile of voters whose opinions it could change, even if the data about them didn’t portray them as perfect centrists. Only through such experiments that try to push voters and wait to see which ones moved can targeters know which voters were actually persuadable, and to what messages. …

… thanks to its experiments, the campaign feels confident enough in its ability to identify persuadable voters that it can direct well-trained volunteers to call them with pre-written scripts. (In an election year when so few voters are at all open-minded about the candidates, true persuasion targets are so dispersed that it is rarely efficient to send volunteers walking among their houses.) The messages are crafted for different kinds of persuadable voters. Obama’s persuasion message for certain female targets threatens a “return to an era when women didn’t have control over own health choices.” Analytics are transforming the role, and value, of volunteers.