The new science of winning campaigns
Posted Tuesday, Nov. 6, 2012, at 3:22 PM
Posted Tuesday, Nov. 6, 2012, at 11:55 AM
"I Voted" stickers at Battlefield High School in Gainesville, Virginia. The AFL-CIO thinks it can detect election fraud if the number of reported votes cast doesn't match regularly occuring patterns in nature.
Photo by KAREN BLEIER/AFP/Getty Images
When election returns start flowing into the AFL-CIO’s boiler room tonight, analysts will be looking to see where Big Labor’s preferred candidates can claim victory. But they’ll also be monitoring whether the digits in vote totals follow a natural logarithmic distribution, part of a newly-developed election-fraud detection system that AFL officials say they modeled on the math-heavy protocols that credit-card companies and the IRS use to catch cheats.
“What this is doing is looking at the numbers,” says Mike Podhorzer, the AFL’s political director. “Not whether the candidates are hitting the numbers we thought they were going to hit but looking at the numbers themselves.”
The system, designed by senior political strategist Matt Lackey, is based on Benford’s law, a mathematical principle that states that the first digits of numbers are distributed steadily in nature. In most organically produced sets of numbers, from baseball statistics to the areas of rivers, the ratio of whose (for example) first digit is 7 to those whose first digit is 4 should remain constant. A math and economics major before dropping out of Purdue in his senior year, Lackey says he has had Benford’s law on his mind since college. “It shows up in math textbooks once it gets to the point where it’s all weirdos taking those classes,” he explains. Lackey entered politics after stumbling into a job as Barack Obama’s Indiana voter-file manager during the 2008 campaign.
Lackey analyzed past election returns, and realized that the total number of votes cast per precinct adhered to the same natural progression. There were roughly three times as many precincts whose number of total votes cast started with a 2 as with a 6. Over the last couple of months, Lackey wrote code that could automatically assess the ratio of first digits appearing in the real-time election returns that the AFL receives from the Associated Press. “It became something to think about while the servers were working on other stuff,” he says.
Tonight, Lackey and his colleagues in the AFL’s analytics department will be checking on charts that should appear as a clean slope if the digits in vote totals occur in the same ratios they are expected to appear in nature. “We can detect when things are or are not natural,” says Lackey. “If you’ve got a big analytics shop, it’s not hard to get them to do one more thing, if it’s looking at a graph and saying does it look like a slalom?”
When the IRS sees numbers that don’t follow that slope, it puts up red flags calling into question whether the returns that produced them were honest. The AFL will search for places where abnormal patterns may be clustered, like those in a county or municipality where a common election authority may have tampered with the results. “You’re not going to pick up an individual voting twice,” Podhorzer says. “The thing that most election administrators don’t realize is that the returns are going to follow a natural pattern. If they’re not following a natural pattern it’s because someone changed them.”
Any abnormalities they find will feed into the AFL’s media and legal departments, who will use it to draw attention to areas where fraud may have occurred. “It doesn’t mean it’s airtight proof that there’s something the matter. It provides a warning system that before winners are decided, have a reason to look at that particular precinct or county for some kind of reason it was that way,” says Podhorzer. “It’s a burglar alarm. The fact that it was tripped doesn’t mean someone is coming in through the ductwork.”
Posted Saturday, Nov. 3, 2012, at 10:38 AM
One of the anti-Obama designs that emerged from Affinova's "survival of the fittest" tests. The individual elements—the message, photo, hook line—were developed by Gridiron Communications and pitted against one another before a panel of online consumers, then reassembled as a single mailer
A few weeks ago, I poked some fun at a PR stunt from the Boston-based packaging consultancy Affinnova, which claimed that its "repurposed algorithms devised for genetic research" had tested potential Romney running mates and determined that Condi Rice would have been the most valuable addition to his ticket. It seemed like the classic example of corporate marketers exploiting a media appetite for “Big Data” stories about politics without any idea of the problems campaigns try to solve.
Now the company seems onto a more productive collaboration with political professionals. The Republican data firm Vlytics and Indiana-based mail vendor Gridiron Communications approached Affinnova in September to see if those so-called evolutionary algorithms could help improve the process by which campaigns develop direct-mail pieces. Now they are often generated through the worst design-by-committee impulses, where consultants and campaign staff arrive with holistic concepts and then litigate the particulars. “In mail meetings, everyone comes in with a few ideas and then they tweak crappy ideas,” says Scott Tranter of Vlytics. “Everyone is just going from their gut.” Why not start by identifying all the individual parts and systematically test each element against other options to develop the best result?
That’s the way Affinnova helps corporate clients, like Procter & Gamble and Microsoft, shape their product packaging or (in the case of credit-card marketing) the offers themselves. The company reduces a design to its essential components — a tag line, bottle color, image, slogan — and then solicits as many as dozens of different possibilities for each. Those elements are reassembled in various combinations into complete designs juxtaposed against side by side before an online panel of consumers who pick the one they think works best. The winning ones are then pitted against one another in what Affinnova vice president Kris Green calls a “survival of the fittest” competition.
To test whether this approach would work for political design, Gridiron’s Chris Faulkner developed individual components for an imaginary anti-Obama mailer, and sent all of them to Affinnova for testing. The company screened its panel for undecided and leaning voters in battleground states and, over four weeks, showed around 500 of them the elements in various combinations. “A lot of the time we are guessing about which images of President Obama work best on moving voters,” Faulkner says.
Perhaps the most radical assault on traditional consultant expertise came in the way Affinnova used its panel: as would-be experts rather than as voters. Respondents weren’t asked which one they liked most, or placed in a survey experiment to see if their views of Obama changed in response to certain pieces, but were asked which design they thought was “most likely to change opinion against Barack Obama.”
It is an indirect formulation that Affinnova uses on behalf of commercial clients, to adjust for the fact that not all respondents are necessary target customers: If you were running this company, which would you launch? Such a “gamified question,” as Affinova’s Mayur Kshetramade describes it, can “make them more invested in this decision-making process, so they think their decision has a lot of value,” he says. “You get consumers and voters interested in the process and they’re more thoughtful.”
It is a peculiar way of detaching voters from the decision about what might change their opinion, and the political professionals involved in the project say it’s just a tentative foray into new methodologies for testing mail. “It’s definitely not how traditional pollsters ask the question,” says Tranter. “It sounds good, it definitely makes sense, but do I know it works? No, I don’t.”
Posted Thursday, Nov. 1, 2012, at 7:02 PM
A Barack Obama poll watcher observes the voting January 26, 2008 in Denmark, South Carolina.
Photo by Stephen Morton/Getty Images
The Huffington Post reported earlier today on Project ORCA, a smartphone-based real-time poll-monitoring system developed by the Romney campaign for deployment on Tuesday:
Project ORCA will rely on 34,000 volunteers in swing states on Election Day, in an effort to keep track of who is voting at key polling places. Romney staffers will use the data to help them target their get-out-the-vote efforts before the polls close, in hopes of gaining an edge over Obama's grassroots operation.
"There's nothing that the Obama data team, there's nothing that the Obama campaign, there's nothing that President Obama himself can do to even come close to what we are putting together here," Romney Deputy Political Director Dan Centinello said Wednesday night in a training call for Project ORCA volunteers, which The Huffington Post called into.
Such unfounded “We will bury you”-style bluster is becoming a hallmark of the Romney campaign style. In this case, the two campaigns seem to have developed similar, parallel systems, with one big difference: Republicans appear quicker to have their poll-watchers create work for lawyers.
Obama has developed his own smartphone-based real-time poll-monitoring system, code-named Gordon, after the man who reputedly killed magician Henry Houdini. In 2008, Obama’s campaign had built a similar system, named for Houdini, because it was designed to automatically “disappear” voters on the campaign’s list of turnout targets as soon as they showed up at the polls. By late morning, the call volume from volunteers assigned to enter touch-tone reports through interactive voice-response protocols overloaded the campaign’s phone lines and crashed the system. It was perhaps the most staggering tech flop of the year for Obama.
While ORCA keeps an IVR option open for volunteers who lack data-equipped phones, Gordon works on smartphones only. “There was no appetite for going down that road again,” an Obama official says. Generally, however, Democratic operatives have grown skeptical that there are many tactical adjustments that can actually be made during election day based on imbalances in turnout. Typically campaigns will patch surrogates into talk-radio stations in areas where they feel they need a boost, and are able to edit phone-center call lists so that canvassers are more likely to be getting out the vote of those who haven’t yet. But it’s often too late for a wholesale redirection of resources to underperforming areas if there’s not already a local field infrastructure in place.
“We're not putting a whole lot of stock in this stuff this cycle—we're doing bellwether reporting through Gordon but we're not going to attempt to cover like every single polling location in our target areas,” the Obama official explains. “We figure those bodies are put to better use working the doors.”
One crucial way in which the two election-day systems differ is that ORCA allows poll-watchers who spot potential problems to “press a yellow button on their phone to instantly report them to the Romney campaign legal team,” according to Huffpo’s Amanda Terkel. Obama has established a separate system, called LBJ—a salute to the Democratic president who enacted the Voting Rights Act as well as an abbreviation of “Lawyers Bound for Justice”—that so-called voter protection counsels can use to report legal issues.
“It's cool that they built their election-protection stuff right into the app,” one Democratic data manager says of ORCA, “but i don't know if you'd want to turn every volunteer into an amateur lawyer.”
Posted Thursday, Nov. 1, 2012, at 9:33 AM
The election recount room for the Senate race between Al Franken and Norm Coleman at the Ramsey County election office on November 19, 2008 in St. Paul, Minnesota.
Photo by Cory Ryan/Getty Images
Let’s move on from last week’s vogue of electoral-college deadlock scenarios—which would throw the election of the president to the House and the vice president to the Senate—and toward what is shaping up as 2012’s post-election legal nightmare: a North Dakota recount.
Both Democrats and Republicans are busy preparing to wake up Wednesday having to wage legal fights in both North Dakota and Montana, which share small electorates and very close Senate races. North Dakota has the prospect of being a much more open-ended mess if lawyers for candidates Rick Berg and Heidi Heitkamp start challenging the eligibility of those who cast ballots: The state is the only one with no voter registration.
Any North Dakotan who arrives at a polling place (or early-vote location) with an ID card showing him or her to be over 18 years old and a resident of the local precinct is handed a ballot approved by a judge of elections. Anyone who does not have ID, or one with a local address on it—or is challenged by an election monitor—can complete an affidavit ballot attesting to his or her eligibility. While the state retains a central file of who has participated in past elections, voters’ IDs are not checked against it or any other external data source.
But these affidavit ballots are not segregated, as provisional ballots are elsewhere. Like standard ballots, they are filled out by hand and then run through an optical scanner on the spot, according to Cass County auditor Mike Montplaisir, and counted immediately. Even though recounts are automatically triggered when a North Dakota election is determined by a margin of less than one-half of 1 percent—or within 2 percent if requested by the losing candidate—they have never in the past involved questions of individual eligibility. “It’s never got into challenging voters, because if a vote has been cast and a vote has been counted you can’t determine which votes have been cast which way," says Montplaisir, who oversees elections in the state's largest county.
Imagine Heitkamp v. Berg having all of the legal machinations of the epic eight-month Minnesota recount that eventually sent Al Franken to the Senate, laced with the partisan vote-fraud paranoias that have hovered over this election cycle. This time, however, it is Democrats who are imagining buses of ineligible voters streaming to polling places, particularly arriviste oil-field workers in the booming northwest corner of the state who could cast affidavit ballots even if not they’re not permanent North Dakota residents.
“If you are challenged at the time, all you need to do is sign something saying you are a qualified elector,” says one Democratic operative familiar with the party’s plans in case of a post-election legal battle. “If we can prove that an affidavit ballot was inappropriately cast we have no recourse.”
Posted Wednesday, Oct. 31, 2012, at 11:10 AM
MoveOn is mailing 12 million registrants in battleground states a “Voter Report Card” scoring their participation in their past five elections and comparing it to the neighborhood average, Politico’s James Hohmann reported in Morning Score. “MoveOn will then run online ads to draw attention to the project and the fact that outsiders can figure out whether you voted or not,” Hohmann write.
Here’s a look at one of the mailers. It arrived yesterday at the home of a New Hampshire woman who is registered as “Undeclared” (how the state classifies those without party affiliation) but otherwise likely resembles a Democratic turnout target.
The mailer’s design marks the latest twist on the political world’s adoption of what behavioral psychologists call “social pressure.” Experiments have shown that letting citizens know that their voting histories are publicly available—and that as a result they can be monitored and judged based on how often they cast a ballot—is the most potent known tool for driving people to the polls. The most effective version of this technique, in which researchers tried to shame non-voters by threatening to out them before their neighbors, was tested in Michigan in 2006 and documented in this paper by Don Green, Alan Gerber, and Chris Larimer. Much of the focus in the GOTV research world in the last several years has focussed on how to channel the psychological power of that threat into something more positive, while still letting non-voters know they’re being watched..
In late September, MoveOn sent a message to supporters looking to raise money for what appears to be this mail program. “Two weeks ago, we ran a secret test of a new voter turnout method in a state primary election. There were 170,000 voters in the study,” the group wrote. “We just got back the results, and our new method was 3.7 times more effective, dollar for dollar, than the best techniques used by campaigns today.”
By all accounts, we’re seeing that technique in mailboxes now. (My assumption is that the 3.7-times figure compares this method to standard canvassing, mail and phone practices, not necessarily to other versions of social-pressure mail, which are in use elsewhere.) The groundbreaking insights: soften your surveillance with smiling children.
Posted Tuesday, Oct. 30, 2012, at 1:15 PM
Ralph Reed, founder and chairman of the Faith and Freedom Coalition, speaks at Herman Cain's Revolution on the Hill Tax Day Rally at the U.S. Capitol April 16, 2012 in Washington, DC.
Photo by Win McNamee/Getty Images
Ralph Reed writes in response to my story yesterday, which cited the example of a phone-banking Obama supporter and regular Democratic primary voter in Virginia who has received mail from Reed’s Faith & Freedom Coalition as part of the group’s voter-contact effort to mobilize ten million socially conservatives to vote this year. Reed says it’s unfair to use that as an example of his targeting missing its mark.
“We could have easily dropped them out of the voter education program---their party and voting history is a matter of record,” he writes. “But for strategic reasons we felt it important to contact Democrats who are voters of faith. We're not a Republican campaign committee, and we believe it's a good investment both short-term and long-term to keep them in the program.” I asked him to elaborate on that point:
As to why contact Democrats, because we are a faith-based issues advocacy organization that seeks to mobilize, train, educate, and equip people of faith to be more effective citizens, in particular evangelicals and faithful Roman Catholics. Especially if you're seeking to move and motivate Catholics on issues, you're dealing with a community that is still about one-third Democrat (that's frequently Mass-attending Catholics).
We're not the RNC or the a GOP candidate committee. So the point in your piece that we may have communicated with a Democrat in VA may be a critique of a Romney or RNC — of that I'm not sure. We communicate with voters based on issues, not party or candidate identification. That goes for evangelicals as well. After all, Obama will get about 20% of the evangelical vote, which is one out of every five evangelicals. Are we not to communicate with them either? I think we should, and we do.
Reed notes that the group keeps approximately two million Democrats in its “evangelical/Catholic voter file.”
Posted Tuesday, Oct. 30, 2012, at 8:39 AM
Photograph by Mandel Ngan/AFP/Getty Images.
I’ve been quiet here recently, largely because I’ve been busy working on a story that went up yesterday. It brings together a few things I’ve been wrestling with throughout this election year, specifically trying to come up with a solid systems and culture explanation for why the left has such a monumental tactical advantage over the right today.
That analytical edge is probably most acutely felt in the way the two sides think about their ability to measure and model which voters are actually persuadable. For several years now, lefty campaign researchers have been aware of the limits of traditional voter targeting to answer that question, but it was a matter mostly of academic concern. Now, though, I think we’re starting to see Democrats translate those insights into practical improvements, namely the use of persuasion modeling whose value may be most immediately manifest in the way the Obama campaign has deployed volunteers this year:
Earlier this year, Obama put his volunteers’ ability to do that to the test. The campaign administered an experiment in several states in which phone-bank volunteers were given a script with a few talking points and broad instructions to open up a conversation with a potential voter. Before and after these interactions, a professional call center surveyed the targeted voters to identify which candidate they supported, and campaign analysts set to work developing a statistical portrait of those who moved in Obama’s direction after talking with a volunteer.
The result of that analysis is the campaign’s so-called persuasion model, which generates a score predicting, from zero to 10, the likelihood that a voter can be pushed in Obama’s direction. (The score also integrates a voter’s likelihood of casting a ballot altogether, so that field organizers focus the attention on those with the best chances of turning out.) A zero designates a voter likely to be repelled by the interaction, and actually pushed toward Romney or a third-party candidate; a one projects a minimal possibility of persuasion; a nine someone who can be easily pushed.
Campaign strategists have traditionally been so fearful of triggering a backlash that they rarely entrust volunteers with persuasion efforts. When placed at a phone or given a clipboard to knock on doors, volunteers usually are given tasks that do not require them to discuss sensitive or complex topics—their role has typically just been asking voters who they support, and reminding those who declare their support to turn out.
“Persuasion calls are a more difficult thing for a volunteer to do because it's a lot easier to hang up on someone than slam a door in their face,” says Wisconsin Democratic Party chairman Mike Tate. “You're not just asking someone who they're going to vote for or reminding them to vote—you're going to people who are undecided, who don't want to hear from you, and are often sick of politics.”
It’s a big change in how campaigns think about what they can accomplish with field. When Obama’s staff brag about all those campaign offices, they’re not just thinking of them as registration and mobilization hubs but centers for targeted persuasion. Unlike most campaigns, they aren’t necessarily counting on television, direct mail and free media to do the work of changing minds.
Posted Tuesday, Oct. 2, 2012, at 9:24 AM
A political campaign no longer has to be liked on Facebook to hunt for votes there.
Photo by ROBYN BECK/AFP/GettyImages
This, from today's Wall Street Journal, is the thing that could finally make Facebook a useful platform for winning votes:
To amp up the effectiveness of its ads, Facebook in recent months has begun allowing marketers to target ads at users based on the email address and phone number they list on their profiles, or based on their surfing habits on other sites….In September, Facebook began allowing marketers with their own lists of email addresses and phone numbers to target ads at specific groups of Facebook users of at least 20 at a time. Facebook matches up that outside data with information users have entered into their profile.
Facebook’s tough grip on member data made it impossible to match up users’ Facebook profiles to their voter-file identities, unless supporters identified their friends manually. Ads had to be targeted through users’ “likes,” which naturally made it a poor platform for the two dominant modes of electoral communication: persuasion and mobilization. Is anyone who clicks to “like” Barack Obama going to possibly be persuadable? And is someone who goes to the trouble to “like” Mitt Romney going to be the type of voter who requires an extra nudge to the polls?
Since no campaign wants to individually engage voters in those terms without knowing who they are and how they are likely to vote, campaigns stuck to ancillary activity on Facebook. They produced content they hoped would be shareable, or engineered stunts that could go viral. They experimented with ad buys, but typically with the goal of converting supporters into volunteers or donors, where poor targeting carries little risk. A fun example of this is suggested by Joe Trippi in a Politico story about small-dollar online fundraising: sending ads to promote a meet-George-Clooney fundraising contest to those who liked Ocean’s 11. If one of the ads is served to a Cheadle-loving Republican, what’s the downside?
But the ability to link Facebook users to individual data points on their voter-file records ought to dramatically expand the type of electioneering that can be intelligently conducted on social-media platforms. A recent get-out-the-vote experiment by UCSD's James Fowler involving 61 million Facebook users demonstrated the network’s unique potential to combine online social pressure with offline voter identification.
Posted Monday, Oct. 1, 2012, at 7:04 AM
Election Official Myrna Carter registers a voter for the Wisconsin recall election at the Bradford Town Hall in Bradford Township, Wisconsin. Does telling citizens their right to vote is under attack make them more likley to exercise it?
Photo by Scott Olson/Getty Images
A few weeks ago, I wrote about the work being done by Pennsylvania Democrats to identify Obama supporters likely to face problems voting under the state's new ID laws and the GOTV-style program launched by party operatives to secure documents for them.
At the same time, some Democratic strategists have proposed that while new laws may demobilize some of the party's base by introducing hurdles (or simply confusion) around the voting process, there was potential upside. A they're trying to disenfranchise you message, especially targeted at young voters and minorities, could motivate unreliable voters to cast a ballot this fall.
“It’s going to be an amazing get-out-the-vote effort for us because there will be stories about people being denied the vote,” John Anzalone, one of Obama’s pollsters, said of voter ID laws at a panel discussion hosted by National Journal during the Democratic Convention in Charlotte. “This has the ability to really piss a lot of people off.”
This spring, a group called the League of Young Voters put this theory to the test by running a randomized-control experiment in Wisconsin before the gubernatorial recall in June, and I joined their canvassers on the streets of Milwaukee. The results of the League's test are now back, shocking a lot of tacticians on the left who had been plotting to use these voter-ID themes in their messaging.
It has opened up a broad debate about the psychology of voting: does telling citizens their right to vote is under attack make them more likley to exercise it? I write about that debate, and the research that triggered it, here.