The Center Holds: Obama and His Enemies

by Jonathan Alter

  The selection process for the Cave was especially rigorous. Dan Wagner and Andrew Claster began by searching Obama donor lists for those whose occupation included the word data or analytics. His finalists had to complete a four-hour online exam consisting of seven or eight fiendishly difficult analytical problems. In the final interview, Wagner informed applicants that starting in mid-2012 the job would be seven days a week until at least midnight every night. “The presidency is on the line and I don’t care about your personal life,” he told them. “We’re not selling popsicles here.” Analytics ended up with a motley crew of mostly under-thirty data scientists and financial analysts, plus a biophysicist, a former child prodigy, and three professional poker players.

  The prodigy, David Shor, had gone to a local college in Miami at age thirteen. When his friends in the Cave asked him why he didn’t go to MIT, he said that when you’re thirteen, you go to college where your mother tells you. Now he was twenty and in charge of assembling the Golden Report, a summary of the sixty-two thousand simulations that Analytics ran every night to create a sophisticated picture of the state of the race. The Golden Report was sent at 2 p.m. every day to a tiny handful of senior aides, including David Plouffe, who often shared the highlights with the president.

  THE BUILDING BLOCKS for both the Golden Report and Field were the Cave’s “ID calls,” the four thousand to nine thousand phone calls a night placed to voters in battleground states. The calls, which eventually numbered nearly one million, sampled ten times as many voters a night as a standard pollster surveyed in a week. While these short, four-question surveys provided less depth than regular polls, they offered more breadth, giving the campaign high command a 360-degree view of the state of the race. Most important, the ID calls allowed the Cave to build and update the models that Field and other departments used every day.

  For decades, campaigns had assembled giant lists of voters, then tried to turn out those who they guessed were likely to vote for their candidate. Now the game was much more about predictive modeling and extrapolation. The modeling produced a “support score” that ranked every registered voter in the United States on a scale of 0 to 100 (100 being a certain Obama supporter) to make canvassing and other voter contact more efficient. Why direct a volunteer to ring the doorbell of a 33 who was most likely a Republican? Instead of wandering aimlessly around the neighborhood dropping off leaflets, canvassers would have maps that mysteriously prioritized visits to residents. Volunteers didn’t know specifically that they were going to see those with the highest scores on different models, though they were vaguely aware that there was more logic than in the past behind their canvassing. The Cave dwellers figured that any new voters or voters with infrequent voting history above a 65 (having a 65 percent chance of supporting Obama) were worth contacting and those ranked 80 or above were must-see.

  Support scores—accessible only to Chicago and certain paid organizers in the field—sounded arid and impersonal. In fact they reflected what was relevant about an individual in an election—not just age, ethnicity, and the other usual demographic categories, but whether the person voted in school board Austan Goolsbee early elections, had a spouse from a different party, wanted to volunteer, hated phone calls from politicians, owned a home, would pay for an Obama bumper sticker, and nearly a hundred other variables. The data were obtained mostly from publicly available sources like voter rolls.

  Analytics also data-mined certain consumer preferences, but this was a small part of the model. The most critical data, obtained from the DNC, were door-to-door voter contact information going back to 1992, much of it from long-forgotten local elections. The voter file assembled over the years was the best resource available, especially when constantly updated with information from the field. It was much more useful for the Cave to learn that a thirty-five-year-old persuadable voter in Zanesville, Ohio, had volunteered in 2000 for the Democratic candidate for state senate than if she drove a Volvo, ate Brie, and listened to NPR.

  Even so, no one in Chicago would ever say exactly what data went into the models. That would disclose the analysts’ secret sauce, not to mention opening them up to charges that they were snooping on Americans. Whatever they had was enough for the Cave to confidently extrapolate from the data to build support scores for the entire electorate, more than 180 million Americans, though the information on those outside battleground states was much less complete. The Cave calculated not just their likelihood of voting for Obama but also the chances they might donate or volunteer and—of critical importance—the odds that they would vote early. Building this support model took much of 2011.

  Support scores were only the beginning, of course. The Cave also issued occasional “persuasion scores” that predicted on a scale of 0 to 10 the effectiveness of a volunteer in getting a voter to change his or her mind. Not surprisingly, those who scored as more persuadable received more voter contact. For the first time, canvassers were enlisted to test messages at the door and go through the awkward conversation necessary to gather data on why someone was not supporting the president. Wagner later said that the modeling on persuasion was perhaps the Cave’s biggest contribution of the whole year.

  When the Cave tested persuasion messages, most of the responses made intuitive sense and corresponded to public polls. But other findings took the campaign beyond mobilization of the base to picking off likely specially otherwise conservative women who responded positively to Obama’s positions on women’s health. If the campaign’s “youth track” was aimed at mobilization, the “women’s track” was more about persuasion. Chicago was targeting a very small number of people in ten states who might change their minds—only a dozen or so, on average, per precinct.

  FROM THE START, there was trouble in digital paradise—a culture clash between the engineers from tech companies and the more politically seasoned product managers and data analysts. Harper Reed’s code writers, though lacking in campaign experience, were often paid $100,000 a year, twice as much as some of their colleagues in other sections of the campaign. Reed said Tech could afford the higher salaries because it held down head count by hiring fewer people than rival departments. But that didn’t go down well in other departments. The pay gap was exacerbated by the Tech team’s habit of routinely leaving the office at the ungodly hour of 6:30 p.m., five, six, even seven hours before Digital, Analytics, and other sections went home. This schedule was explained by the fact that the Lamar Alexandervy were older (meaning a few were in their mid-thirties) and, unlike most Chicago staffers, often had families.

  A little humility would have gone a long way toward helping Tech blend in, but it wasn’t forthcoming. “Instead of ‘Listen and learn,’ they [Tech people] came in with a ‘Burn the place down’ attitude—real arrogant,” said one senior campaign official. “It was, ‘Fuck the vendors—we’ll build everything in-house.’ ” But the vendors, firms like NGP VAN that specialized in voter contact, knew politics, and Reed’s department did not. Tech team members used their fluency in tech jargon to their advantage, but they were often illiterate in basic political language, with everything from SEIU (Service Employees International Union) to GOTV going over their heads. And they often took their mandate for “disruption” too far. Some Tech staffers even dismissed email as old-fashioned and uncool, without understanding how indispensable it would be in saving the campaign.

  All of this would have been minor if the products Tech developed were working. But they were late and often useless, like an online fundraising tool that took six weeks of precious time to build and raised a pathetic $20,000. “It’s not 2011 anymore!” Finance chair Julianna Smoot would complain in meetings, her frustration growing. The beta testing schedules the product designers were accustomed to in the private sector were much too slow for a political campaign.

  Tech’s great white whale, dubbed “Narwhal” (after a toothed whale), was to integrate the more than 13 million Obama supporters now on the Email List with many other databases. Instead of modeli
ng and extrapolating, the dream was to match 25 million Facebook “likes” of Obama with county voter registration rolls, census data, 2008 voter contact information, contributor lists, and fresh information from door-to-door canvassers for a unified data platform on millions of voters in battleground states. Narwhal was supposed to be built on the fruits of the 2008 Houdini Project, the Election Day voter-tracking system that assigned Obama volunteers at thousands of polling places to record which Obama voters showed up and why. Much of Houdini crashed on Election Day 2008, but the data survived and became a vital part of the 2012 database. This time Chicago would name its Election Day voter-tracking system Gordon, in honor of J. Gordon Whitehead, a college student who punched Harry Houdini in the stomach in 1926, killing him.

  In Tech’s defense, the complexity of the Narwhal project was daunting. It was easy to give an Obama volunteer the names and emails of supporters in the same zip code. And making the DNC’s massive voter file, called “Votebuilder,” available to local field organizers was doable, though it hadn’t been accomplished in 2008. By 2012 the campaign had “support scores”—usually based on rudimentary data such as party ID—on all 180 million American voters. But Chicago had originally envisioned something much more sophisticated. The hairiest challenge, the goal of Narwhal, was identifying which Obama backer served in Iraq and might be willing to send his buddies a video of the president talking about the VA, or which dieting donor might want to learn more about Michelle Obama’s anti-obesity initiative, or which Obama supporter had once signed a petition for women’s rights and might now be willing to send her whole mailbox something about Mitt Romney’s attack on Planned Parenthood.

  This was much harder than it sounded. Instead of waiting for Tech to build Narwhal, Analytics compensated by relying more heavily on software from a company called Vert the twentieth century, s h2ica that helped produce enough new data sets to keep the Cave dwellers happy. And Chris Wegrzyn and Gabriel Burt of Analytics developed a tool dubbed “Stork” that allowed key vendors to transfer their data into campaign databases. Suddenly the fruits of door-to-door canvassing in a 1998 city council race in Youngstown, Ohio, could be used to help target voters with similar support scores in Littleton, Colorado. OFA ended up with pieces of Narwhal instead of the whole whale.

  The best piece of all was Facebook, which was growing so fast that it might be able to accomplish some of what everyone had hoped for from Narwhal. The social networking site was ten times as big in 2012 as it had been in 2008, with more than half of the U.S. population now active users. In 2010 Facebook embedded new widgets in its system that gave users a bumper crop of fresh information. “It cookies ya,” Teddy Goff said, using the slang for tracking software as a verb. Goff, four years out of college, joked that he was “a little too old to understand these things.” Fortunately for Obama, the campaign was loaded with geeks who knew what Facebook could do before Facebook did. The Cave’s Rayid Ghani, OFA’s chief data scientist, led a team that began to customize a Facebook app into something called “targeted sharing.” The idea was that if an Obama supporter had, say, one thousand Facebook friends, the campaign could determine that nine hundred of them were already for Obama, focus on one hundred who were persuadable, and ideally zero in on six or so who lived in battleground states and were in regular enough contact to be considered real friends, not just Facebook friends. When those potential Obama voters were identified, their friend (the active Obama supporter) would be notified. He or she could then send them Obama’s position on issues and urge them to register and eventually to vote. Because the message came from a real friend, it would be much more credible and influential than if it came from a stranger representing the campaign. That was the theory, anyway.

  All of this was much easier to envision than to execute. Common names meant thousands of cases of mistaken identity. Uncommon names were often misspelled and thus orphaned. Some voters used home addresses, others work addresses. Some states had clean, well-maintained voter rolls; others had what Digital called “hygiene” problems—messy, outdated lists that were of little use in cross-referencing with Facebook. This had been one of the reasons why the full Narwhal had been so hard to build. To succeed, OFA needed to develop a confidence level on the part of its supporters that “John Smith” was your friend John Smith and not someone else. Projects like Narwhal and targeted sharing were so complicated and secretive that beta testing was tough, which meant that the geeks would have to work out the bugs after it went live, if it ever did. Some of the people designing Narwhal thought it wouldn’t be fully functional until 2016, by which time Barack Obama, if he won, would be getting ready to leave the White House.

  A modified OFA product, designed for paid field staff and their most committed volunteers, aimed to help them keep track of all the money they raised, calls they made, and doors they knocked on. Originally called “the Wire” and later rechristened “Dashboard,” it was also constantly behind schedule. Harper Reed promised it in November 2011, but—to the fury of Field—it wasn’t delivered until the summer of 2012. The same went for the “Quick Donate” button, a one-click to allow supporters to text money without filling out credit card information each time. It was promised to Digital and Finance for June 2011, then September, and finally delivered (by Digital’s own tech crew) in January 2012. Lamar Alexanderv

  The delays and glitches could be terrifying for Chicago. Just five months before the election, a server crashed when a mere ninety-two people went on a site at one time after a campaign tweet. The Tech crew didn’t fix the servers, and three weeks later the site was brought down by three hundred hits. Someone on the Floor heard Teddy Goff yell, “What if Lady Gaga tweeted!?!” Goff wasn’t sure Tech would be ready for crunch time in the fall.

  Could Narwhal be reeled in? Was targeted sharing a Facebook fantasy? Would the servers crash and Gordon meet Houdini’s fate on Election Day? No one in Chicago knew.

  WHATEVER THE TECH frustrations, the broad targeting of soccer moms and NASCAR dads was all in the past for OFA. Now it was about using “propensity models” and a hundred other analytical tools to mobilize and persuade voters. Whether a voter was an 85 on the support scale or a 6 on the persuasion scale was more important than if she was a young African American woman in Pittsburgh or he was an old Jewish man in Cleveland or vice versa. Obama had come to national prominence in 2004 by saying in his convention speech, “We pray to an awesome God in blue states and we don’t like federal agents sniffing around our libraries in red states.” While his 2012 campaign couldn’t afford to target red states, it made a point of targeting outnumbered Democrats in bright-red precincts and otherwise moving beyond the crude stereotypes that had dominated voter contact in the past.

  There was a hierarchy to what the campaign called “supporter mobilization.” Friend-to-friend contact was best, which was why the campaign put so much energy into its Facebook strategy. Then came face-to-face contact with voters, aided by durable Votebuilder software that allowed canvassers to efficiently skip most houses and only knock on doors where supporters or potential supporters (ranked on a 1 to 5 scale by volunteers) lived. Door knocking was more effective than phone banking, which in turn was more persuasive (and brought higher call volume) when volunteers came into an office with automatic dialers and other callers rather than making the calls from home.

  The least effective forms of voter contact were robo calls (except when they featured the Obamas or Bill Clinton reminding people to vote) and paid solicitors. The latter, avoided by Chicago, featured low-wage workers with no connection to the campaign reading scripts in a monotone, a turnoff on the phone or at the door, especially if the solicitor said he or she was from a super PAC–funded group nobody had heard of. The point was, the Koch brothers couldn’t just buy a first-class field organization from afar; it had to be painstakingly built at the local level. That explained why OFA was so obsessed with building a million-plus corps of volunteers.

  THE ANALYTICS THAT worked so well for Fi
eld also helped Obama’s TV advertising. Larry Grisolano and Jim Margolis, the message mavens and Paid Media experts just below Axelrod, worked with the Cave to build a system they called “the Optimizer” that allowed them to target ads with a precision never before seen in politics.

  The traditional approach was to advertise heavily during local news, the theory being that people who watch news are more likely to be voters. In recent elections, ad b">Citizens United v. Federal Election Commission, ,Pauyers grew moRepublicans, e

  9

  Not So Great Communicator?

  In the movie Cool Hand Luke, the warden famously says to the inmate played by Paul Newman, of his recent predecessors.:off:0000000OK “What we’ve got here is a failure to communicate.” Obama had confessed to his own failure in this area as early as 2009, but he never tried a new communications strategy or otherwise gave it more than episodic attention after 2010.

  When Obama was good, he was outstanding, as he was when he spoke about the victims of gun violence in Tucson, in Newtown, and in his 2013 State of the Union Address. But much of the rest of the time his speeches failed to break through, partly because, from the start, the expectations had been too high and partly because of the fragmentation of the old media culture. As recently as the late 1990s (when the web, Fox, and MSNBC were brand new), the Clinton White House only had to worry about a few newspapers, newsmagazines, television networks, and talk radio. Now every president was confronted with a Balkanized, polarized, and ill-defined media where almost anyone could drop poison into the national bloodstream and make “news,” whatever that meant anymore.