After the original and re-run Kenyan presidential elections, the Austrian legislative elections, the snap election in the UK, and the US presidential election in November 2016, fake news on Facebook, computational propaganda on Twitter, and other interventions have revealed modern democracy's vulnerabilities. Erosion of trust in public institutions, information overload, narrowing margins of victory, and preference for our friends over traditional media outlets as gatekeepers have all incentivized political campaigns to leverage digital tactics to win over hearts and minds. The most valuable asset underlying these methods is data, and political campaigns are becoming disruptors of their own as start up-like tech supplants old-fashioned approaches to politics. Campaigns now accumulate and curate data because it serves as a window into understanding voters. As politicking changes, the tradeoffs made by campaigns are raising fundamental questions about democratic participation.
To understand how political campaigns are evolving, Tactical Tech's Data and Politics team attended CampaignTech Europe, an inside look at leading political campaigns in the US and Europe and the strategies they employ. A glimpse at the data and practices touted by political campaigns and strategists suggests a new political modus operandi.
American campaigns collect and analyze more data to inform political strategy than anywhere else. The abundance of personal information accumulated and exchanged by companies, relaxed data protection, and publicly accessible voter logs have facilitated novel digital campaigning methods in the United States.
Jim Walsh founded American-based DS Political that ran the UK Labour Party's first video campaign. It has jumped political, legal, and geographic boundaries for nearly three years through its international engagements in the UK, Canada, Australia, Brazil, South Africa, Turkey, and Sweden. Its website claims to have the "best data and technology to reach every vote, on every device, everywhere." At CampaignTech Europe, Walsh remarked *:
There's lots of geo-fencing opportunities. We discovered in this last cycle, where you could actually geofence a polling location and serve ads to every single person who's waiting in line to vote that day. How fantastic! We actually called it..."The Last Word" because it was the actual last opportunity someone had to see an advertisement before they voted.
Geo-fencing is the process of building a digital fence around an enclosed area. In this case, people inside the fence can be targeted with ads, while those outside are excluded. US-based campaigns that employ technologies like these operate on a level of granularity and technological sophistication that campaigns in other parts of the world are emulating. In fact, European campaigns have long been adopting and integrating data-centric practices into their own tactics.
Paris-based Liegey Muller Pons has served political campaigns in six European countries. Its Co-Founder & CEO, Guillaume Liegey, worked as National Field Director for François Holland's presidential bid in 2012, when he mobilized 80,000 volunteers across France to visit 5 million households. He also canvassed for Barack Obama and Emmanuel Macron. In an open group discussion at the conference, he attested:
There are actually tons of data available in Europe. And that was my surprise, coming back from the US, where it's data paradise. And every European telling, 'We have no data'...well, none of this is true...there is a lot of data...We don't have individual-level data [in France], but there is a lot of aggregate data available at the neighborhood level — roughly 1,000 people. So in France, for example, we built a database on 67,000 neighborhoods, and we have election history, and all the census data, so that's dozens of variables that we then use to create a targeting model to tell our clients where they should concentrate that input.
Strategists maintain targeting models like these help campaigns identify demographics and characteristics of persuadable voters, enabling them to employ their limited resources more cost-effectively. The belief is that the more a campaign knows about you, the more it can address your concerns and reduce communications irrelevant to you while also informing your vote.
The data feeding these models come from many different sources but in the US virtually always include voter roll information. Many European nations with robust data protection do not furnish public voter rolls, which contain basic information on all registered voters in the US on a state-by-state basis (e.g., a voter's name, date of birth, gender, address, phone number, registered party, whether a vote was cast in a given election). Even without this information readily available, European campaigns have built digital interventions on data from private companies.
Walsh, who spoke fondly of geo-fencing, also noted, "If you don't have a voter file, it's okay. There are tons and tons of consumer data on the ad exchanges that we built." The company conducted an investigation into targetable consumer segments available online in Poland, a country without a public electoral role, and found over 100,000. Campaigns can incorporate these segments into targeting models and mine them to uncover the features of swing voters.
When the desired data is not available from voter rolls or formal markets, campaigns may just collect it. Liegey said he witnessed this practice in both Holland and Macron's campaigns and incorporated this tactic into his company, whose mission, in his own words, is to transform public opinion campaign and polling techniques by combining advanced technology and human[s], using...door-to-door interaction."
But you can also create your data, and that's what we did with Macron when he started online. [The] Holland [campaign] started from nothing...The first thing they wanted to do was a door-to-door campaign...to ask people their view of France, something they don't tell to pollsters...but something they would tell to normal people coming to their door and asking them open questions.
In total, Liegey reported that Macron's team completed 25,000 conversations lasting 14 minutes on average all about individual concerns, hopes, and stories of French voters.
Unlike in Europe, getting data like Macron's in the US doesn't always require a centralized, labor-intensive data collection campaign. It's available from services like DS Political and other firms selling data. As Walsh stated, "In the USA, you have registrations that are publicly available information. You can take those voter rolls and match those to our online profiles and serve those people ads individually online."
Bryan Whitaker, Chief Innovation Officer of Target Smart, a communications company that specializes in consumer and political data, underscored the value of data in politicking. His experience using data in elections includes several Democratic campaigns in the United States and Justin Trudeau's bid for the Prime Ministership in Canada. As COO of the company NGP VAN, his expertise helped thousands of progressive campaigns — political and non-profit alike — leverage technology. While explaining the benefits of data science in campaigning as part of a panel discussion, Whitaker shed light on the specificity and volume of personal data available in the US:
You can actually see if a cat owner who also rides a motorcycle who's also female and works within the psychiatry field...is more than likely than not to be predisposed to supporting John Kerry for president, which is exactly what we saw in the Iowa Caucuses in 2004.
Whitaker pointed out that the 'female, cat-owning, Kerry-supporting motorcyclist-psychiatrist' stood out "because it was part of the thousands of data points that we had on individuals." Though he didn't describe what other sorts of data comprised the thousands of observations and predictions the campaign had amassed on voters, it is clear that even as early as 2004, campaigns were investing resources in data collection.
In the lead-up to the 2012 presidential elections in the US, during his tenure as Head of Technology for the Democratic National Convention, Whitaker recounted that President Obama's re-election team started winning the campaign with its national canvassing program. Volunteers were sent door-to-door asking people whether they support the re-election of the president and what their biggest concerns were for the next two years. The large-scale data gathering effort enabled the campaign to "figure out who would support [Obama's] re-election, what issues mattered to them. And then we could do the analysis on the backend of what are their demographics, what are their socioeconomic variables, what magazines do they subscribe to, what are their consumer purchasing behaviors."
Campaigns build models to detect patterns between purchasing behaviors, for instance, and voting patterns based on the people they spoke to. They then apply these patterns to the purchasing behaviors of people they haven't spoken to in order to infer their voting preferences. The resulting predictions dictate whom the campaign invests resources in reaching.
Chris Young, former National Field Director for the Republican National Committee (RNC), spoke on the panel with Whitaker about the RNC's data practices and lessons. Young noted that the national branch of the RNC accounted for 25 million door knocks in 2016, and that campaigns outside the scope of the national committee likely produced three times as much data. Before, that information was siloed off into individual campaigns and "when a campaign ended, that data died." Now, instead, all of that information is compiled and stored in a centralized GOP data center, so "it's a whole ecosystem benefitting both" national and local races.
This consolidated repository of information prompted the need for more sophisticated technological solutions given the demands imposed by sheer volume and maintenance. Young remarked:
The RNC starting working with Microsoft and moving [the data] into the Azure Cloud...so the RNC's voter files are 300TB, which, I was told this morning, if you started streaming that on Netflix 24/7, it would take you 75 years to watch that much Netflix program. So it's a lot of data, going all the way back to 1992. So if anyone's wondering who was voting for George H. Bush in '92, I can get that for you.
Looking ahead, Whitaker predicts "deep canvassing" will take root; he describes it as a practice in which information from in-person conversations, like those from door-to-door organizing, is used to predict "how people are feeling, what their sentiment is, and then actually ranking the order of...the issues they identify." Doing so would enable campaigns to tailor communications around stated concerns and their relative importance, person-by-person, instead of approaching voters "with an agenda." He expects that Smart TVs, Google Home, Amazon Alexa, and other in-home technologies will present campaigns with opportunities not yet realized. Chris Young's take on the future was different. He remarked that "we are probably heading into a new arms race" but on a digital field in which power is acquired through data.
Some scholars have suggested that the solution to these developments is more stringent data protection standards. Notably, however, strict data protection laws have not prevented European campaigns from adopting American-born data practices, as these strategists affirm. Even in the absence of voter roll data, European campaigns can — and do — target voters by leveraging less granular data or consumer segments built for online advertising or by collecting data en masse.
These challenges and the technologies behind them are global. The four experts mentioned here have political experience spanning North America, South America, Africa, Europe, Asia, and Australia.
Today's campaigns are disrupting the political status quo and introducing a host of questions in the process. In this digital arms race, how do we conceive of voter suppression in which individuals are reminded to vote selectively via targeted ads? What are the risks of political campaigns purchasing our own purchasing behaviors from private companies? Do we want to live in a world in which political communications are personalized and in which people are served political ads at polling stations? These ought to be questions for the democratic process.
* All block quotes in this piece are from the conference CampaignTech Europe. ↩
Varoon Bashyakarla is a data scientist and researcher at the Tactical Technology Collective. His past statistical undertakings led him to a variety of domains: public health, public safety, sports, finance, and cybersecurity. After working as a data scientist in Silicon Valley, he is now living in Berlin and exploring how personal information is used for political influence.
Thank you to Arikia Millikan for her editorial guidance in writing this piece.