Technical innovation in campaigns has a long history. That’s perhaps the most surprising takeaway from Sasha Issenberg’s “The Victory Lab,” an exploration of the development of analytical and data-driven methods in political campaigns over the years. His history of the methods and their applications begins as far back as the 1920s, in an implicit challenge to the news coverage that frequently implies each year’s campaigns have developed something entirely new.
Tracing those roots is interesting for historical reasons, but it also sheds some light on the nature of the campaign analytics field. Despite the hype they attracted, Obama’s 2008 data innovations addressed old problems in politics: identifying supporters, persuading people who weren’t already supporters, and turning out enough of both to win an election. Other campaigns had been tackling the same problems for years, from Harold Gosnell’s pioneering 1920s field experiments to the 2004 Bush campaign’s ahead-of-its-time efforts. Data-driven methods are not, in other words, a fad or a distraction: they’re a natural application of new technological tools to the same problems campaigns have always confronted.
Issenberg’s work is best read as a history of those tools and campaigns’ increasing reliance on them, rather than an attempt to develop or push a theoretical perspective on why that might be so. He has rich detail, vivid writing and deep reporting1 - the RNC’s “stalker” (Issenberg, postscript) precinct captains are a good example and a nice touch - but not a theoretical scaffold to hang it on. Perhaps appropriately for a book about the scientific method, writing The Victory Lab this way raises more questions than it answers. Among them:
- How do these new techniques impact the political system beyond campaigns? Which groups do they empower in the actual making of policy? Are politicians with a scientifically enhanced ability to raise small-dollar money less inclined to go along with the wishes of big donors?
- Why is this happening now? With its roots reaching to the 1920s, analytics has taken quite a while to break through in campaigns. What changed recently?
- Where does campaign analytics go from here? What’s on tap for future elections?
Issenberg doesn’t say, but he does provide a good historical starting point for considering and answering those questions.
“Microtargeting” vs “analytics”
The Victory Lab’s history of Republican “MicroTargeting” (131), as it was originally written, is a particularly good example. It began in earnest with the Bush reelection campaign in 2004, as an attempt to add some science to the ways campaigns historically figured out who their target audiences were and how to talk to them. Traditionally, campaigns had relied on their strategists to identify the most important voter groups and what messages would move them, what Issenberg refers to as the “guru” (147) style of campaigning. (The “security moms” that most of the political world obsessed over during the 2004 election are a classic example.) The new approach substituted data for human gurus’ judgement, with “[a]lgorithms [finding] the variables that were pulling people together in ways that informed their likelihood of backing Bush’s reelection, whatever they may be.” (130)
The algorithmic approach, after the Bush campaign adopted it, didn’t just shift control of message segmentation away from human strategists, it shifted the nature of the messaging that was delivered and the campaign contact that delivered it. Messages could be more finely tuned to segments constructed to be receptive to them; the way that the Republican Party was able to send “social conservative messaging” to those who had “shopped at a Christian bookstore” (132) or otherwise appeared to be religious conservatives in Pennsylvania’s 2003 elections foreshadowed what was to come nationally a year later. Voter contact also became more targeted, with the RNC and the Bush campaign being able to find “three people out of ten on a block” (134) who were likely supporters.
Automated segmentation may have been the flashiest innovation, but it wouldn’t have worked in isolation. The new methods relied on infrastructural data work, like the consumer profiles built up by Acxiom and the RNC’s painstakingly constructed national voter file. (Finding favorable segments of voters isn’t very useful without being able to get lists of the voters in those segments, after all.) More importantly, the new methods’ effectiveness relied on “the political structure [the Republican Party] had built around them” (161) more than on any technical changes. The Democrats, after losing the 2004 election, had work to do at all levels, from targeting techniques to the data they depended on to business processes that could take advantage of them, to catch up with the Republicans.2
They did that work surprisingly quickly, all in all, with microtargeting being superseded only four years later by newer methods, which have collectively come to be called “analytics.” (e.g., 171) These newer techniques, building on what had come before, carried the same dependence on voter and consumer files, but turned them to different ends. Rather than trying to derive small, targetable segments of voters, the new techniques focused on both broadly applicable individual-level “percentage probability” scores (270) and on randomized experiments. The combination of the two allowed a campaign to both learn which messages or contacts were most effective (via experiments) and derive finely calibrated target universes to apply them to. These newer techniques were most effectively developed on the 2008 Obama campaign, but they had deep roots in past practice. To give one example, the same Ken Strasma who built the 2008 campaign’s voter models tried in 2004 to “design similar models [to an even earlier pilot in Iowa] for voters nationwide,” (159) but was rebuffed by campaign leadership. The experimental side of the new analytics had even deeper roots, with Alan Gerber and Don Green’s famous New Haven experiment kicking off the modern age of campaign experiments in 2000. (80) Both methodologies have since become standard on Democratic campaigns, with even small campaigns that don’t use them recognizing them as best practices.
Indeed, from a campaign strategy perspective, the difference between microtargeting and analytics is larger than it may seem. The newer techniques, which provide individual scores for things like candidate support, are less applicable to message development. (When each person is a segment, segments have become too small to use in calibrating messaging.) It’s certainly possible to build models of issue opinion for targeting purposes, but doing so requires running a survey and having a sense in advance of which issues to ask about on the survey. Microtargeting techniques, which do unsupervised clustering of voters, have a greater chance of turning up clusters of voters the campaign can tell itself a clear story about.
And this difference is practically relevant: with the institutionalization of analytics as a separate department or team in campaigns, the more closely its tools drive it to associate with the field program, the less associated it often becomes with message development. I’ve certainly had the experience of seeing messages developed primarily by campaign strategists and traditional pollsters, with an analytics team involved (if at all) only at the end to run a final experiment.
The differences between the new analytics tools and traditional polling are also instructive. David Simas, for example, found the 2012 Obama campaigns’ analytics surveys useful for things his team’s more traditional polls had difficulty with: precisely identifying changes in specific, quantifiable things like candidate support. “The sheer scope of the analytics research” let Simas identify “movement too small for traditional polls to perceive.” (347) As with the older microtargeting approach, though, neither was strictly better: there were advantages and disadvantages to both. The analytics polls’ disadvantages centered on the very brevity and narrowness of focus that made them “significantly cheaper to conduct” (347): once again, they weren’t much use for message development.
Issenberg focuses mostly on the voter-contact applications of analytics in campaigns. He has good reason to: those applications have historically driven most innovation in analytics, and it’s certainly true that winning votes is the ultimate point of electoral campaigns. The book isn’t called the “Victory” Lab for no reason. But there are applications and effects that aren’t about vote-getting, and Issenberg gives them short shrift.3
The biggest such effects are almost certainly on fundraising. Campaigns from Howard Dean in 2004 to Barack Obama in 2008 to the present day have found that the new data-driven dispensation is fantastic for raising money, and especially so for raising small-dollar money. The Obama campaigns’ multi-billion dollar fundraising was mostly online donations, and quite a bit of the rest was model-targeted direct mail. It’s quite likely that in context, these effects are much bigger than analytics’ effect on the chances of actually winning the election. Voter contact has become more efficient, sometimes significantly so, but campaign fundraising was completely transformed.
The fundraising improvements’ effect on the broader political system are difficult to isolate, and are certainly less visible than Obama’s vote totals in Ohio, but there’s reason to think they’re rather large. With the ability to raise prodigious quantities of online money and targeted small-dollar money, plenty of candidates are viable in primaries who wouldn’t have been before. Bernie Sanders is a topical example, but Barack Obama would also likely have lost the 2008 primary without his digital operation and its fundraising.
More speculatively, campaigns which view the world through a set of data-colored glasses may end up suffering from measurability bias. The idea that, as the saying goes, “not everything that counts can be counted” is one an increasingly data-driven campaign culture would do well to keep in mind. Issenberg doesn’t evince much concern with the consequences of failing to do so, preferring rosy predictions that data would “recognize the sovereignty of the people who participated in politics” (325) and titling one chapter “The Sovereignty of Data” (323). It’s nevertheless a real concern, with several real-life examples to choose from.4 Fundraising provides a good one: organizations like the DCCC, which is notorious for the tone of its emails, can optimize their fundraising programs so completely for revenue that they lose sight of the effects of doing so. In the DCCC’s case, there’s widespread speculation, though admittedly a lack of hard evidence, that such emails turn people off from donating and engaging in the long term.
Why now? What next?
Issenberg’s reporting and historical work discuss campaign techniques in great detail, from microtargeting to randomized experiments, and he frequently raises questions about the effects and implications of those techniques (as with microtargeting vs analytics, above). He avoids, however, providing a unified theory of them and their effects, rendering it difficult to make predictions for campaigning in the future.
For example, there’s the question of why these analytical techniques are only coming to the fore in campaigns recently. Many campaigns over the years have made improvements on previous practices in data management and polling (Issenberg’s thorough discussion of Hal Malchow’s career provides many examples) but until 2008 or arguably 2004, the scale of the changes was similar and relatively small each election cycle. Witness, for example, Malchow’s repeated and unsuccessful attempts to interest campaigns in CHAID (63).
There was a breakthrough in 2008, with antecedents in 2004, but it’s not immediately clear why it didn’t happen earlier. The computing power to support it has been available for decades, and data brokers have been compiling consumer information for almost as long. Very few of the statistical methods being applied in political analytics are particularly recent - usually they made their first appearances in the academic literature decades ago.
So why now? Issenberg doesn’t say. It could be the falling cost of computing power, the greater availability of personal computers to democratize access to the data, the spread of expertise from academia and industry into politics, or a number of other things.5 In all likelihood, the answer is all of the above, in some socially mediated combination, but Issenberg doesn’t provide a framework for figuring it out. (The media hype that for a surprisingly long time has accompanied each election cycle’s advances in data is certainly also unhelpful. Dan Rather’s characteristically gee-whiz claim in 1972 that “chances are[,] your name is on the computer” in Nixon headquarters bears a family resemblance to the 2008 and 2012 hype about the Obama campaigns.6)
While the lack of a heavy-handed theoretical perspective makes The Victory Lab a more enjoyable read, it also leaves the book silent on where campaigning goes from here. The most recently written chapter, Issenberg’s postscript, was finished in early 2016, before it was possible to have a fully formed perspective on the uses of data in that year’s election. Indeed, it’s still too early: there’s a lot that isn’t known about the tactics that year of all three sides (Trump, Clinton and Russia), and the ways they used technology and analytics for advantage. At the very least, though, the existence of a sophisticated and analytically optimized Russian influence campaign should falsify some of Issenberg’s techno-optimism. Data may have “measured the ability of people to change, and be changed by, politics” (348), but it won’t do so only for those with good intentions.
Issenberg, Sasha. The Victory Lab: The Secret Science of Winning Campaigns. New York: Broadway Books, 2013. Kindle edition.
Silver, Nate. “The Real Story of 2016.” FiveThirtyEight. https://fivethirtyeight.com/features/the-real-story-of-2016/ (published Jan 19, 2017; accessed March 4, 2018).
(Citations with only a page number are to Issenberg 2013.)
From personal experience: Issenberg was remarkably persistent in trying to get information about the Obama campaign from my coworkers on the digital analytics team in 2012. The creativity and (appropriately) good targeting of his emails became a running joke. ↩
As Hal Malchow put it, “[t]he best way to get anyone to do anything on the Democratic side - and I’m sure it’s the reverse on the Republican side - is to tell people that the Republicans are doing it.” (160) ↩
He certainly talks about them - the DNC’s ability to fundraise from its “roster of more than 150 million new targets” (166) is a good example - but they’re a side note to the discussion of voter contact. ↩
The 2016 polls are not among them. The media’s overconfidence in Hillary Clinton’s general election chances was never justified by the polling data, which showed her narrowly leading a tight race at most points before Election Day. Nate Silver’s lengthy retrospective analysis makes the argument in much more detail (Silver 2017). ↩
The Help America Vote Act of 2002, which required the states to clean up their voter rolls, is a strong dark horse contender. ↩
Why the media hypes campaign technology this way is an interesting question in itself, but not one that’s within Issenberg’s scope. ↩