This spring, the Cato Institute identified 600 Americans who read more than 20 books per year and made arrangements to send them each one more. The libertarian think tank split these readers into three groups. One group received a free copy of Ayn Rand’s Atlas Shrugged, one got longtime Cato executive David Boaz’s The Libertarian Mind, and one a book that Cato scholars considered a useful placebo to free-market doctrine: the Bible. After three months, six months, and 12 months, members of all three groups would be surveyed to see if the unsolicited books they had received could explain differential response rates to one question: Do you consider yourself a libertarian?
The Cato researcher behind the project explained to other members of a below-the-radar Republican group known as the Center for Strategic Initiatives, or CSI, that the 600 books were just part of a pilot test. If the design appeared to work properly, the experiment would be replicated on a larger scale: 12,000 books this time. “Political books have never been tested,” says David Kirby, now a vice president and senior fellow at Cato. “Think tanks think that books persuade people. Do they?”
Very few other members of the CSI circle had ever used books as tools for changing minds. A range of political consultants and vendors, they tended to trade in more ephemeral modes of communication: television ads and robocalls, direct mail, digital ads, and door knocks. But they were there for the same reason that Kirby had been willing to entertain the perfidy of using Cato resources to question whether reading Ayn Rand actually led people to libertarianism—a willingness to take everything they thought they knew about what works in politics and hold it up to empirical investigation.
The mere existence of the CSI field experiments seminar represents the right’s most constructive engagement with the continued traumas of its loss in the 2012 Presidential race—not just the fact that it had lost, but that it didn’t know at the time it was losing, and even afterward was at a further loss to understand how or why it had done so. While many Republicans responded by conceding catch-phrase-ready deficiencies—a need to do more and better with Big Data or the Ground Game—others were willing to acknowledge that the underlying problem was the lack of a culture within the GOP to encourage innovation. For a small but significant share of the party’s electioneering class, however, any true reckoning with 2012 invites a deeper epistemological crisis about how to run smarter campaigns in the 21st century.
“We should not assume anything. Absolutely every aspect of the campaign, from the best way to knock on doors to the best way to broadcast television, should be tested,” says Blaise Hazelwood, a Republican voter-contact specialist who founded CSI. “This is the way I did it on this campaign that won, so this is the way we should do it on all campaigns,” Hazelwood says, mocking the prevailing sentiment of entrenched political consultants. “The test for them is whether they win or lose on election day. That cannot be a valid test.”
The conservative establishment has a long tradition of organized gatherings. First there were Paul Weyrich’s weekly “coalitions” lunches, held on Wednesdays when Congress was in session. Anti-tax activist Grover Norquist later claimed the breakfast slot that day, his sessions focused more on economic policy than social issues. Donors and journalists in New York started meeting one Monday per month for an hour and a half; after relocating to South Carolina, Monday Meeting founder Mallory Factor took the concept south with him, launching the Charleston Meeting. All these were devoted to ideological cohesion and legislative strategy, the matching of like-minded donors and politicians. They tended to reaffirm certainties, rather than challenge them.
The CSI circle has yet to fall into a reliable schedule, and its gatherings—which now take place roughly every six weeks or so at Cato’s Washington headquarters—already mark a very different mode of collaboration. There is not a politician in sight, or many brand-name operatives; few attendees appear to be over the age of 40. This sphere of political operatives and party hacks angling to remake Republican campaigns includes strategists and tacticians for many of the party’s top presidential candidates, along with staffers from the Republican National Committee and consultants attached to various elements of the Koch political network.
“I sense it’s one of the few places where the warring factions of the conservative side of the aisle play together in the same sandbox,” says Columbia University political scientist Don Green, who has advised the group since its launch. “You have people who are close to the Tea Party and people who are antagonistic towards the Tea Party, and they’re all trying to learn from the same research method.”
About 50 people responded to Hazelwood’s invitation to gather at Cato in early June and, arrayed before her that Thursday afternoon, at long tables in lecture-hall formation, the schisms of the Republican Party in the early days of the presidential campaign were unmistakable. One of the people she had invited to present research that day came from Deep Root Analytics, which former Florida Governor Jeb Bush's presidential campaign has hired to help target its television ads. Another presenter represented FLS Connect, a phone vendor that has worked for Wisconsin Governor Scott Walker and has close links to his campaign-in-waiting. A pair of analysts from 0ptimus consulting, now advising Florida Senator Marco Rubio, sat in the back of the room, not far from where the founders of Targeted Victory, the firm former Texas Governor Rick Perry has hired to handle digital advertising and media analytics, found seats after arriving a few minutes after the scheduled 3:30 start time. Hazelwood herself is a longtime advisor to Louisiana Governor Bobby Jindal, through her firm Grassroots Consulting, and has since joined his allied super-PAC as a consultant. She began by reminding those in attendance of CSI’s one rule: What they were about to hear was off the record. “Ask questions like you usually do,” she said. “Attack our speakers, like sometimes happens.”
Hazelwood has been asking and answering some of the toughest questions in Republican politics for years. Shortly after the 2012 elections, she faced one from Sally Bradshaw: What did Republicans need to do to improve their electoral mechanics? A longtime adviser to Bush, Bradshaw was one of five party eminences who had been tapped by Republican National Committee Chairman Reince Priebus to oversee the Growth and Opportunity Project, the postmortem report its authors were repeatedly instructed not to describe publicly as an “autopsy.”
Twelve years earlier, Hazelwood had been in a similar position when, as the RNC’s political director, she had organized what was known as the 72-Hour Task Force to fine-tune field operations that Karl Rove and other top advisers to President George W. Bush blamed for his unexpectedly narrow victory. Starting in 2001, Hazelwood deployed dozens of controlled tests in low-profile elections, usually with the goal of convincing party officials and activists that it was worth investing in the infrastructure to support volunteer-based voter contact. (Many of those tests were modeled on the types of field experiments conducted by social scientists, although not conducted to academic standards.)
Many Republicans credited the findings of the 72-Hour Task Force with helping Bush in his re-election, but after his victory, the culture that had incubated them withered away. When one of Green’s protégés approached an official on the Indiana Republican Party’s 2006 coordinated campaign asking if he could integrate experiments into the party’s efforts, there was only one question that mattered: Was the political scientist “in the family”? That described a very specific qualification—having worked on one of the two Bush-Cheney campaigns—and it was not on David Nickerson’s CV. Soon, though, Nickerson had a Ph.D. from Yale, then an associate professorship at Notre Dame, and then, in 2012, a title that had never before existed in the history of presidential politics: director of experiments at Obama for America.
Nickerson’s trajectory confirmed the extent to which Obama-era Democratic operatives have scientized the project of electioneering. The party’s ascendant young operatives—notably Robby Mook, Hillary Clinton’s campaign manager—nearly all had formative experience working on voter contact (rather than strategizing communications or managing political relationships) in a period of technical transformation. They were conditioned to see every aspect of citizen behavior as potentially predictable through statistical modeling, and everything a campaign did to shape it as a potentially testable hypothesis. Meanwhile, the fractured remnants of the Bush-era Republican Party did not indulge such curiosity within its ranks, and when it did emerge, did little to channel it productively. “After 2004 the drive just left. It was no longer there—we no longer did the testing and we ended up where we ended up in 2012. We just kind of stopped,” Hazelwood says. “It was terribly frustrating after 2012 to see what happened, and know we needed to get back to testing, and having that housed in one place to help all conservatives.”
Hazelwood made the case to Bradshaw that the party should take the initiative in spurring such a cultural shift, albeit with a different objective than the one that had inspired the 72-Hour Task Force tests. “Back then we framed it completely differently. It was, ‘We’re getting back to grassroots and showing why grassroots works,’” Hazelwood explains. “Now it’s about tests, and getting to test return on investment and how you measure efficiency and effectiveness.” Hazelwood helped Bradshaw rewrite the task force report's section on “campaign mechanics” to offer its imprimatur for what had become her pet project. Republicans should “identify a team of strategists and funders to build a data analytics institute that can capture and distill best practices for communication to and targeting of specific voters,” as the report put it. “Using the GOP’s data, the data analytics institute would work to develop a specific set of tests for 2013 and 2014—tests on voter registration, persuasion, Get-Out-The-Vote, and voter mobilization—that will then be adopted into future programs to ensure that our voter contact and targeting dollars are spent on proven performance.”
The report cautiously avoided suggesting that the RNC incubate or host such an institute itself. The national party was a subject of more distrust among conservative activists than it had been during the early Bush years, and after Citizens United even more political spending was being driven by groups that were legally forbidden from coordinating with the party or its candidates. Hazelwood found a model for such a “data analytics institute” in the organization that had paved Nickerson’s way into Democratic campaigns after he was turned away by Indiana’s Republicans. In 2007, similarly driven to self-examination by their inability to dislodge an incumbent president they assumed was beatable, a group of lefty operatives launched the Analyst Institute, to serve as a “clearinghouse for evidence-based best practices in progressive voter contact.” To do so, they formalized a series of “geek lunches,” held at the AFL-CIO, at which individual operatives and researchers presented the scattered experiments they had been running on their own, into a stand-alone entity. Since then, the Analyst Institute’s work has transformed the techniques of Democratic campaigners in the Obama era, and eventually contracted to embed its analysts in the president’s re-election effort in 2012. Hazelwood decided to structure her project similarly, as a for-profit consulting firm that could operate with little concern for ever turning a profit, a status that allows it to work with parties and campaigns as clients without being subject to disclosure laws itself. Hazelwood was transparent in her debt to the Analyst Institute, down to the unhelpfully abstract name.
She pitched her new effort to collaborators across the right, eager to take advantage of the oddball calendar of off-year elections to begin running tests. There was one in Massachusetts, where the American Crossroads super-PAC agreed to randomly assign its robocalls on behalf of Gabriel Gomez, a first-time candidate challenging Representative Ed Markey to replace John Kerry in the Senate. In Minneapolis’s mayoral campaign that same year, Hazelwood worked with the pro-business Minnesota Jobs Coalition to test what type of messaging works best to boost a conservative candidate in a officially non-partisan race with a large fragmented field.
In parallel, Kirby was also digging more deeply into experiments. He had spent 2012 working at FreedomWorks, the well-funded conservative grassroots organization that had invested heavily in field activity designed to help elect a Republican president and a more conservative Congress. Kirby emerged from the experience resentful of the the confidence political professionals brought to their decisions about tactical approaches. “We were disappointed in 2012 with how much money we spent and how little we had to show for it,” Kirby says. “I felt integrity-bound to ask whether we could do any better.”
Kirby was drawn to the experimental method as a tool for resolving that curiosity. He signed up for a brief summer course on randomized trials taught by Green, the Columbia political scientist recognized as his discipline’s most prominent evangelist for field experiments as the only tool capable of truly disentangling cause and effect in campaigns. (Green’s work recently gained broader attention when he was forced to retract a paper he had published in the journal Science, about the effectiveness of sending gay canvassers to advocate for same-sex marriage, amid accusations that his co-author Michael LaCour had falsified survey data.) Green was inspired by Kirby’s enthusiasm, and—after years of finding growing interest in his methods from from labor unions, environmental advocates, and ACORN —the opportunity to see if conservative targets responded differently to political communication than liberal ones. “Almost all the research to date has been on the left or center-left. Very little of it has been on the center-right or right,” Green says. “There isn’t a base of knowledge about the message, the messenger, and the audience they care about.”
Kirby was interested in starting with lawn signs. “It is the most elemental political tactic,” he says. “At FreedomWorks we spent probably at least a million dollars on that—hundreds and thousands of yard signs.” Yard signs were often held up as an example of wasteful and pointless campaign spending (“yard signs don’t vote” is a frequent refrain of the bien pensant political class), although there was little empirical evidence to sustain that skepticism. The limited research literature could be attributed partly to methodological complications—it is not easy to control for the spillover effects of, say, voters who drive past signs in a precinct other than the one in which they vote—but some stemmed from a lack of obvious incentive for political professionals to run an experiment. “In the case of lawn signs, what’s odd is if you’re dealing with a person who is deeply skeptical, they’re not interested enough to do a test,” says Green. “People working in low-salience races see them as their one and only affordable tactic and don’t want to take out a control group.”
Working with Green, Kirby ran an experiment in which Virginia precincts were isolated and then randomly assigned to have signs placed at a certain density in road medians, each of them photographed and geotagged so it would be possible to monitor whether they were removed by opponents or vandals. The signs had little impact on election results in those precincts, they believe—“maybe a glimmer of an effect” on both turnout and vote share, according to Green—but given how cheap signs are to print and place, the experimenters thought that they might have demonstrated that signs might not be a total waste after all.
Kirby became an enthusiastic booster of Hazelwood’s project. After he went to work at the Cato Institute in early 2014, he offered one of the think tank’s large conference rooms to host the right’s version of the “geek lunch.” The first CSI meeting, in late 2013, drew more than 100 people from across conservative politics. “It was the right moment,” says Kirby. “People were curious.” Green led a three-hour introduction to the experimental method, and encouraged attendees to think of it as a tool for assessing the effectiveness of their current tactics and auditioning new ones. Many of the experiments that attendees concocted in response—to test direct mail and phone calls in relation to turnout—addressed research questions that some lefty groups felt they had settled through replication as much as a decade earlier. “It wasn’t breaking any new ground,” Kirby says, “but at least it was getting the muscles flexing.”
Within the CSI circle, there is resignation that one cultural problem Republicans face modernizing their campaigns remains beyond their immediate control: Academic social scientists capable of designing, administering, and analyzing field experiments are overwhelmingly on the Democratic side. For his part, Green—with his politically catholic view toward experimentation and promiscuous search for collaborators who will let him test scientific theories in the real world—remained something of a unicorn. “The left enjoys a lot of sympathetic academics and legions of ideologically aligned grad students that we on the right just won't have the same level of access to,” says Brian Stobie, a former official at the Charles Koch Institute who attempted to recruit academic collaborators to test the Koch network’s election-year operations before leaving to start his own firm. (One political scientist approached for this article did not want to acknowledge on the record any openness of graduate students to partner with conservative organizations on research projects, for fear of causing the graduate students reputational damage.)
The emergence of the CSI circle has also unmasked a generational divide more fundamental than the one between practitioners of online and offline politicking that defines most press coverage of the Republicans’ technology predicament. There is a Web-savvy new guard that chafes against the old-line consultants who continue to profit from spending on television, phone, and mail. But these disputes—about the percentage of a campaign budget devoted to Web and mobile advertising, or whether a digital consultant has a proverbial “seat at the table” on strategic decisions—are lost on the CSI crowd, who have a scientific bent. Many there are eager to take on the digital consultants and the grandiose claims some of the digital people make, unchallenged, about their potential impact on elections. Those in the experimental set pronounce themselves agnostic about tactics, interested only in those whose worth have been proven, empirically. What can be measured to deliver votes—rather than just clicks or eyeballs—and at what cost?
Last May, strategist Dave Carney demonstrated for the CSI seminar what such skepticism could look like when put into practice. He and University of Texas political scientist Daron Shaw had come to present findings from a complicated series of experiments that Carney had invited Shaw to conduct within the gubernatorial campaign of Texas Attorney General Greg Abbott. (The two had worked together on Perry’s 2006 re-election, when Shaw and Green served on a team of in-house experimenters that Carney named his “eggheads,” a collaboration that has produced more peer-reviewed papers than any other campaign in history.) Shaw could have merely presented his findings in the form of charts and graphs, but that would not have fully satisfied Carney’s preference for dramatic colloquy. He invited the consultants and vendors whose services he had audited to be part of the discussion—an equivalent of beauty-pageant contestants being forced to stand on stage while judges render a verdict on their presence and poise.
Shaw had randomized the delivery of $1.6 million worth of campaign communications—across broadcast and cable television, a variety of digital formats, radio, and direct mail—with effects measured at both individual and neighborhood units. Three entire media markets, including El Paso and Amarillo, were assigned to a control group that received no Abbott campaign communication over the course of that month. Unlike many studies, Carney had begun by asking Abbott’s consultants to devise plans at a budget that would “allow them to do what they needed to in order to produce an impact.” “One of the most common criticisms one hears from practitioners about political science experiments is that the treatment is ‘not what we’d do in a real campaign,’” Shaw explained. “The obvious solution is to give the professionals the resources and discretion to design and implement their preferred outreach within the context of the experiment.”
Comparing polls conducted before and after the experiment, Shaw explained to the CSI seminar that three weeks of campaigning increased Abbott’s popularity by nearly eight points among the voters he was trying to persuade, with targeted broadcast television buys doing so with the greatest force. At the same time, voters in zip codes served Facebook and online pre-roll video ads were less likely to support Abbott afterward, and turned out at the same rate as those who lived in areas that never received the ads. (Shaw did find that, in some combinations, Internet advertising did have a modestly positive impact on turnout, according to voter-registration records updated after the primary.) Digital consultants “might argue that the relatively modest effects associated with online outreach in our tests still constitute a more cost-effective investment than broadcast television,” Shaw acknowledged in an article, which, as part of his deal with Abbott’s campaign, he has submitted to political-science journals for publication.
His presentation of “The Abbott Campaign Experiments: Persuasion and Turnout in Texas” wowed the CSI seminar, as much for its ambition and audacity as any specific finding. But to the disappointment of many attendees, few other consultants expressed an interest in mimicking Carney’s adventurousness. “Testing, whether done on the right or left, tends to show that the majority of what you think ‘works’ doesn’t,” says Stobie, who after the 2012 election launched 0ptimus Consulting to conduct experiments with the promise of helping clients to constrain their spending. “To the majority of operatives and organizations, paying for the opportunity to find out much of what you are doing is ineffective is not an appealing investment.”
A few weeks ago, Hazelwood met Priebus for lunch at the Capitol Hill Club and handed over the results of an experimental study he had commissioned from her firm. Hazelwood’s reputation had impressed Priebus long before the two first met, in 2009, back when he was a lawyer who oversaw field offices near his Wisconsin home and had “done every yard-sign job in the party you can imagine.” Sharing a ticket with Bush in 2004 during a failed state Senate run, Priebus saw how, by “dividing states into turf, and measuring outcomes and metrics,” as he puts it, Hazelwood’s internal reforms had modernized the party’s approach to field organizing. “I’ve known her as the author of that concept,” Priebus says. “I really do understand the ground game and what we need to do.”
The 72-Hour Task Force had been established to convince local party activists like Priebus that the ground game was not merely something one did or didn’t do—that some tactical approaches to it were superior to others. Hazelwood had isolated “precinct organizing” (assigning a full-time field staffer for a month), “GOP flushing” (indiscriminately turning out Republican voters in the last two days before an election), and “volunteer calling” (compared to paid call centers) as three distinct activities, and subjected each to separate tests. As a result of the impact—two to three percentage points increase in turnout for each, they found—the Task Force prescribed that Republican campaigns assign one person to be individually responsible for every 40 to 100 voters. “Back to People Power,” an internal RNC presentation titled the recommendation at the time .
After Bush’s re-election, the party let much of that field infrastructure wither, along with the ethos behind and intellectual justification for it. During the years since, experiments have revealed a cognitive logic for understanding what mobilizes voters. Basic interactions immediately became more potent when they were personalized in a way that introduced a sense of individual responsibility around voting. Campaigns now frequently remind citizens of their past history casting a ballot, and recruit them to sign pledges to do so again. (One of Green’s most famous tests, conducted with his former Yale colleague Alan Gerber, sent voters a copy of their past registration records with a threat to publicize a set after an election, so others could see whether or not they voted. It proved the most effective get-out-the-vote technique ever measured.) The behavioral psychology behind such techniques gave new depth to campaigns’ understanding of “people power.”
Hazelwood developed a hypothesis that people power could learn something from stalking. Campaigns have long thought of voter contact as part of a narrative progression: a flight of six direct-mail pieces over which a candidate’s argument unfurls, or a “layered” approach in which phone calls and door-knocks are delivered in a deliberately choreographed sequence. But campaigns rarely share these plans with voters, often preferring to let the interactions speak for themselves and not draw undue attention to their volume. Instead, Hazelwood thought it was time to mash up the various tactics that the 72-Hour Task Force had segregated into distinct programs. Just as the pledges and the vote-history techniques had, walking voters through the process as it was happening could produce a personal engagement with the election.
Hazelwood was anxious that the party not let a busy midterm election year pass without using every opportunity to hone tactics it could deploy in earnest in 2016. After San Diego announced a special mayoral election, scheduled for February 2014, Hazelwood convinced a friend managing Republican Kevin Faulconer’s ultimately succesful campaign to let CSI get involved. Hazelwood took command of some of the party’s field programs there, randomly assigning some areas to receive specialized attention from local precinct captains. When she was able to show that her tactics had increased turnout by 8.5 percent among the party’s so-called “low-propensity voters,” the party’s chairman took notice.
That finding intersected neatly with what Priebus had identified as his party’s strategic challenge, and one of the areas the Growth and Opportunity Project had blamed for the party’s inability to win the presidency or the Senate in 2012. “We’re a really good midterm party, but for some reason presidential-year-only voters have been trouble for us,” Priebus says. “If we put people out in the field—and we have a limited amount of people and limited amount of time—what’s the combination of things we can do that have the best effect of turning out people who vote only in presidential years?”
Over the summer, Priebus signed off on Hazelwood’s proposal to have CSI help answer that question. He was insistent that Hazelwood’s experiments not interfere with the party’s plans to spend $105 million across key races in mitdterm battlegrounds. In three states, however, Hazelwood’s firm was permitted to layer atop the party’s extant organization an additional precinct program that would target the type of low-propensity voters that Republican strategists did not expect to necessarily vote in 2014, but knew they needed to mobilize in 2016. After working with Hazelwood on how to structure her study, Green directed her to enlist Oklahoma State University professor Brandon Lenoir, whom he had guided on another lawn-sign experiment, for help implementing it. (There is such entrenched skepticism about yard signs that Green has waited to publish until he has five experiments, each conducted in distinct geographical and political contexts, that can be assembled into a single paper with more comprehensive findings than Kirby’s alone.)
In nine counties across Colorado, Iowa, and Arkansas, Hazelwood’s team worked with the state parties and the Republican Senate candidate to hire dedicated local organizers it called “precinct captains.” Each was given a list of voters who had cast a ballot in one or fewer of the past four major elections, with directions to visit each voter in mid-October for an introductory conversation. “I was designated your Precinct Captain for the Republican Party, for the duration of the election,” the canvassing script instructed. Then, over the next few weeks, the precinct captain would make contact four more times, in a specific order: again at the door, then by phone, followed by a postcard, and a then final door visit. If the voter was home each time, it meant a recurring interaction with an increasingly familiar neighborhood figure.
If the voter didn’t answer the door or phone, however, the precinct captain was instructed to make the missed encounter just as personally memorable. Each door hanger left behind should have a handwritten Post-it note attached to it, with a message that the “personalized note instructions” recommended should say: “Hi Mary, sorry I missed you today! … I will be back in the neighborhood again next week and try to catch you then.” The postcard would pick up the theme: “I have stopped by your house a couple times and tried to call to make sure you have all the information you need to vote.” For a voter who wasn’t home after two door visits and a phone call, the next note would turn even more ominous: “I will keep checking the list to see if you have voted and if not will drop by your house again to follow up.”
Over the course of three weeks, Hazelwood’s “stalker test” deployed some of the most potent psychological tricks known to nudge a citizen into the act of voting—along with some new ones—all as part of a borderline-creepy courtship that would be difficult to ignore. After the election, in which Republicans won the Senate races in all three of those states, Green and Lenoir found that it had a considerable impact, increasing turnout among targets by between two and three percentage points. It had not been easy, though, as stalking demanded attention, resources, and planning that campaigns cannot always muster for their field programs. (The report did not calculate the cost of such contact, and it is certainly conceivable that at a much lower price yard signs could prove more cost-effective—if less targetable—than the elaborate program that required the same field staffer to be available for every interaction with an assigned voter.)
When Hazelwood presented the analysis to Priebus at the Capitol Hill Club, she put the paper that Green and Lenoir had written into a thick, glossy brochure with far more elaborate graphics than are typically packaged with academic articles. Priebus says the findings confirm what he had assumed about the value of localized field programs that focus on sustained personal contact, but now “I can take that information and go sell it to the state parties, and go to into a boardroom to get people to cut a check.”
The finding that most excited both Hazelwood and Priebus, however, grew out of an accident. In Colorado, canvassers were mistakenly assigned to go to some of the party’s reliable voters, who had cast ballots in all four of the most recent elections. After being hounded by a precinct captain, they voted at a lower rate than those who received no contact at all. “We have a proven fact now that we should stop calling these people,” Hazelwood says, with comic impatience. “Stop wasting time!”