Two crates of beer and 40 pizzas: the adoption of innovative political behavioural targeting techniques

Introduction

As political campaigns compete, they try to outsmart each other by all sorts of actions: from dropping witty puns during a televised debate, to strategically knocking on doors and convincing voters. Technological innovation can help political parties improve the effectiveness of their campaigns. By using technology to collect, process, and analyse information about voters, campaigns can improve their knowledge about the electorate. Subsequently, technology can extend campaigns' capabilities of targeting specific groups with tailored messages resulting in more efficient campaigning. We call this phenomenon 'political behavioural targeting' (PBT).

Several scholars have researched political behavioural targeting in the US context (e.g., Kreiss, 2012, 2016; Nielsen, 2012; Hersh, 2015). However, the US differs in several obvious ways from most European countries. One can imagine that differences in electoral systems, privacy laws, and party financing influence campaigns’ ability to collect, process, and use personal voter data. Therefore, the findings from these studies do not necessarily apply to European countries. As there is little research in a European context, it remains unclear to what extent and how campaigns in a multiparty democracy, such as the Netherlands, use PBT-techniques. Also, it is unclear if and why there are differences between parties. In line with Colin Bennett (2016, p. 261), we wonder: "can political parties campaign in Europe as they do in North America?"

Such a question is relevant, as some scholars fear that the use of data and targeting techniques hinders public deliberation (Gorton, 2016), weakens the mandate of elected officials (Barocas, 2012), has negative effects on citizens’ privacy (Howard, 2006; Rubinstein, 2014; Tene, 2011), and enables campaigns to send tailored messages directly to citizens, thereby avoiding scrutiny from journalists (Jamieson, 2013). As a result, campaigns can potentially make opposite promises to different people, without anyone noticing.

This article sheds light on how Dutch political campaigns adopt and use PBT-techniques. Through interviews with campaign leaders, using a grounded theory approach, we answer the following overarching research question: What barriers and facilitators for the adoption and use of PBT-techniques do Dutch political parties perceive?

Theoretical framework

We will first summarise innovations in political campaigns over time, leading up to the advent of political behavioural targeting. Then, we identify the factors influencing the adoption of PBT on a campaign team level. Finally, we explore the factors that can shape the adoption of PBT on the level of national systems.

Innovations in political campaigns

Political campaigns have continuously been adapting to technological developments. Pippa Norris (2000) describes how the advent of television and the shift from partisan newspapers to national television news triggered a process of modernisation in the way political campaigns operated. Notable consequences of this shift were the adoption of a media-centred strategy in order to set the agenda, the rise of political marketing, the collection and use of data (such as opinion polls) to "shape, fine-tune and monitor campaign efforts" (Blumler, Kavanagh, and Nossiter, 1996; p. 53). Another shift came with the internet and the new possibilities for party-voter interaction that came along with the medium, which led campaigns to a new stage of the modernisation process: the postmodern campaign (Norris, 2000).

It would be an oversimplification to point to 'the internet' as a game-changer in political communication, because of the rapidly changing nature of the internet itself. As David Karpf (2012, p. 640) notes: "the internet of 2002 has important differences from the internet of 2005, or 2009, or 2012". Accordingly, much more than the advent of the internet itself, it is the advent of social media such as Facebook (2004), YouTube (2005) and Twitter (2006) which provided political campaigns with new ways of communication with the electorate (e.g., Gibson & McAllister, 2011; Conway, Kenski, & Wang, 2015; Vaccari, 2012). Together with companies such as Google, whose core business is actually not its well-known search engine but rather its advertisement business, social media not only facilitate new ways of communication, but also the tracking and collection of behavioural data of internet users (Zuiderveen Borgesius, 2016). This technique ("behavioural targeting") originates from the advertisement business. Ad agencies monitor people's online behaviour and combine this information with consumer data provided by data brokers, to target them individually with tailored ads (Turow, 2011, p.75). When applying this concept to the political realm, we can dub this phenomenon as political behavioural targeting (PBT).

Of course, PBT is not about selling products but about winning votes. And political campaigns have different means to do so than advertisement agencies have (e.g. canvassing efforts); which means that PBT happens offline as well as online. We distinguish traditional canvassing from PBT-canvassing if campaigns are able to process information about individual conversations (such as the voter's likelihood to vote for a party or her most important voting consideration), and subsequently use that information to gain strategic insights about the electorate and/or to target the voter at a later stage with a tailored message, while skipping the 'wrong' doors in a neighbourhood (Kreiss, 2016; Nielsen, 2012).

Arguably, the use of PBT can be seen as the latest step within the modernisation of political campaigns. However, as we have seen in earlier phases, not all parties in all countries adopt new techniques at the same pace and rate. Below, we identify the factors influencing the adoption of PBT. We organise these factors at two levels: (1) the individual campaign around a candidate/party and (2) the national system (i.e., the electoral system, regulatory framework, and culture). This translates into the model shown in Figure 1, which will be elaborated on in the next paragraphs.

Figure 1: Factors influencing the adoption of PBT

The campaign team level

In his extensive research of US political campaigns, Daniel Kreiss (2016) identified four factors concerned with technological innovation within political campaigns. There are resource factors, such as campaign budgets and the number of volunteers a campaign can employ; infrastructural factors, such as technological tools or skills within the organization; organizational factors, such as organisational culture and structure; structural electoral cycle factors, such as election results. Building upon Kreiss' factors, we add an additional four (one campaign team level factor and three system level factors) to examine the use of PBT. On a campaign team level, the factor is ethical and legal concerns, such as normative reservations towards PBT. On a system level, the factors are electoral context, regulatory framework, and culture (discussed below). These new factors were identified through a review of literature about innovation in data-driven political campaigning techniques (e.g., Anstead, 2017; Kreiss, 2016; Jungherr, 2016; Hersh, 2015; Nielsen, 2012), and literature about (hybridisation of) campaign evolvement (e.g., Lijphart, 2012; Plasser & Plasser, 2002; Karlsen, 2010; Norris, 2000).

Resource factors

The main elements within this factor that could influence the extent to which campaigns can use PBT-techniques are: the budget and the effort needed to carry out a PBT-operation. A large budget enables campaigns to hire skilled personnel, acquire data, or buy targeted ads. The same dynamic applies to the number of volunteers a campaign can mobilise: having a lot of them facilitates a campaign in collecting data by canvassing, and sending potential voters targeted messages (the use of volunteers, of course, is dependent on their skills). Having a small budget and few volunteers, consequently, can be a barrier for campaigns because it bars them from acquiring the same amount of capabilities or from carrying out an operation on a large scale. This is in line with normalization theory (Margolis & Resnick, 2000), according to which the possibilities of the internet will not upset traditional power structures, but will rather develop along traditional lines as in the 'offline world'.

We can also view PBT as a means of using a campaign's resources as efficient as possible, to ensure parties do not spend money and effort on voters who will vote for another party anyway, or on citizens who will not vote altogether. Then, parties with limited resources could be more inclined to use PBT to not waste precious money, time, and labour. This is in accordance with the idea of equalization, which views the internet as an empowering tool for smaller parties due to its low costs and its new ways of direct communication with the electorate (Margolis, Resnick, & Levy, 2003; Bimber & Davis, 2003; Stanyer, 2010). A meta-analysis found evidence for the existence of both normalisation and equalisation in election campaigns (Strandberg, 2008). The occurrence of either process can differ per country and is dependent on several contextual factors, which will be discussed later on.

Organisational factors

The elements in this factor are about how campaign leaders perceive campaigning. Do they rely on proven best practices from previous campaigns or is there a culture of innovation? John Padgett and Walter Powell (2012) describe the concept of network folding. Applied to the political realm, this entails the extent to which campaigns employ skilled personnel from non-political sectors and to integrate that expertise into their existing institutions. An example is the hiring of Google engineer Stephanie Hannon as chief technology officer by the Clinton campaign (Easton, 2015). The 'cognitive diversity' following from network folding can lead to creative ideas (De Vaan, Stark, & Vedres, 2015). Furthermore, the organisational structure can be expected to resemble the way the campaign perceives PBT. A campaign with an autonomous data department is probably more prone to rolling out a PBT-strategy than a campaign that sees 'data' as only one of the many tasks of a communication staffer. Also, a change in leadership can be a facilitator for innovation (Gibson & Römmele, 2001).

Infrastructural factors

Elements are the technological tools available to campaigns, which enable them to roll out a PBT-operation. For instance, such tools might assist volunteers in the field by enabling them to collect data. They can be developed in-house or outsourced; in fact, there are specialised third party consultancies, that offer off-the-shelf tools, which in turn allow campaigns to employ innovative technology even though the campaigners do not have any technical expertise.

Structural electoral factors1

The actions of rival campaigns fall under the umbrella of structural electoral factors. A successful PBT-campaign of a rival can facilitate innovation in other campaigns, especially if those other campaigns themselves look back at an unsuccessful election. This connects with the 'critical event' (Kreiss, 2016), such as losing an election that should have been won, or with the experience of an 'external shock', which can be an incentive for professionalisation (Gibson & Römmele, 2001).

A second element influencing campaigns' likelihood to use PBT-techniques, is issue ownership (Petrocik, 1996), and the subsequent statements of party candidates propagating standpoints of the party. A political campaign 'caught' using privacy-infringing PBT-techniques, while its candidates present themselves as privacy champions, is likely to come across as hypocritical. Being perceived as such should be avoided, considering the negative electoral consequences following political-ideological hypocrisy perpetrated by politicians (Bhatti, Hansen, & Olsen, 2013).

Ethical and legal concerns

Elements within this factor consist of ethical and legal restrictions on how campaigns operate. For example, a political party could believe that PBT is ethically wrong as it infringes on citizens' right to privacy, and citizens' autonomy to form their own opinions. As a result, the party 'self-regulates' and refrains from using campaigning techniques violating its ethical beliefs.

Another element is the legal uncertainty that occurs when a campaign does not know how to behave in accordance with data protection and election laws, because of a lack of internal expertise. Such confusion can result in differences in the actions taken by comparable actors (e.g. Raskolnikov, 2017). Legal uncertainty can lead to 'overcompliance', which can be seen as a barrier towards the adoption of PBT-techniques, or to 'undercompliance', which facilitates the adoption of PBT-techniques (Calfee & Craswell, 1984). For instance, Anstead (2017) notes how parties felt disadvantaged by targeting possibilities facilitated by the perceived undercompliance with UK campaign finance law during the 2015 general elections.

The system level

Aside from campaign level factors, we look at contextual factors as well. These factors may limit the extent to which (US-American) campaigning techniques can be adopted in other countries (Karlsen, 2010). Therefore, we add three new contextual factors to our model. We expect that the electoral system, the regulatory system, and the culture of a democracy influence the extent to which the campaign team level factors are applicable. Below, we explore how the adoption of PBT-techniques can be influenced by properties of different systems. We will later apply our model (see Figure 1) to one specific case.

Electoral system

The three dominant electoral systems are first-past-the-post (FPTP), proportional representation (PR), and two-round (TR) (Birch, 2001, 2003). How these systems function, can influence how campaigns are run. The FPTP-system, first, can lead to an overvaluation of some key districts. Such districts sometimes 'swing' to one party and sometimes to another party, whereas other districts go to the same party in each election. As an effect, campaigns in a FPTP-system are inclined to spend a disproportionate amount of money and labour in these key districts in the hope of swinging the election their way (e.g., Anstead, 2017; Lipsitz, 2004). The PR-system, second, does not favour a select group of voters in a few key districts (Plasser & Plasser, 2002). This is especially true when the PR-system consists of only one district, in which every vote counts equally. As a result, campaigns have to spread their means more equally over the country. The TR-system, third, makes for a relatively unpredictable campaign, since it often is unclear which candidates will make it to the second round. Furthermore, the TR-system makes it important for campaigns to collect the votes of the supporters of the losing candidates of the first round. Therefore, campaigns should not only focus on their own base but other candidates' bases as well (Blais & Indridason, 2003). This has consequences for PBT, since campaigns should not only correctly classify potential voters as their own, but the other voters as well in order to target them in the next round.

A different aspect of electoral systems that influences how a campaign is run, is the degree of fractionalisation in a democracy (Duverger, 1959; Lijphart, 2012; Wang, 2012). FPTP-systems favour relatively few candidates/parties. PR-systems, in contrast, enable a large number of parties to run in an election. The first round of a TR-system can consist of many different candidates. As a result, campaigns that operate in a PR or a TR-system are less likely to launch attack campaigns against competitors. This is because PR-systems generally require a coalition of parties working together after the elections (Plasser & Plasser, 2002). And in a TR-system, campaigns should not mistreat competing candidates too much because winning campaigns have to court the bases of losing candidates in order to win the second round. Furthermore, in a highly fractionalised democracy, parties represent different (minority) groups within the electorate. This results in a high risk of 'mistargeting', in which campaigns approach a member of group A with appeals for a member of group B. Mistargeting can lead to voters penalising the campaign for their mistake (Hersh, 2013). These contextual circumstances may call for different PBT-strategies.

Regulatory framework

We distinguish strictly regulated campaign environments, moderately regulated campaign environments and minimally regulated campaign environments (Plasser & Plasser, 2002). Strictly regulated environments are characterised by "severe restrictions on the contact and communication between candidates and their constituencies" (p. 137). Moderately regulated environments typically focus on regulating access to TV-advertising and campaign funding. Minimally regulated environments impose few regulatory restrictions on political campaigns. It may be infeasible to implement innovative PBT-techniques in strictly regulated environments. Legal uncertainty can play a role on a system level too (e.g. because of a gap in the law).

Culture

Differences in the adoption and use of innovative PBT-practices can also be influenced by the culture or tradition in a democracy. For example, turnout culture is important because campaigns operating in countries where turnout is high will focus more on convincing voters on getting out the vote than campaigns operating in a low-turnout culture. In a low-turnout culture, campaigns sometimes target specific groups of voters (e.g. the elderly, who are more likely to turn out) more than other groups (e.g. the poor), who are unlikely to turn out (Herrnson, 2001). The turnout culture can influence the data campaigns collect on someone (and how campaigns tailor their messages), because a campaign message meant to convince someone typically leans on more data than a message meant to mobilise a voter does. Furthermore, cultural norms can dictate the strategy of political campaigns. In Japan, for example, posting dark post attack ads, such as the 'super predator' ad Trump launched against Clinton (Green & Issenberg, 2016), is improbable because of the cultural convention of averting direct conflict (Plasser & Plasser, 2002).

System level context is likely to affect campaign level factors. A campaign operating in a multiparty PR-system needs to pour more resources into identifying potential supporters than a campaign in a FPTP-system. After all, identifying potential Republicans or Democrats is easier than identifying potential voters in a ten-party race. Moreover, other than in US campaigns, most European campaigns are unable to access voter registration files provided by an electoral register. In many countries citizens can just show up at the voting booth, which means that the whole act of ‘registering’ for voting, as it is the case in the US, does not exist. Since Hersh (2015) has found that voter lists are one of the most valuable pieces of data to US campaigns, this principal unavailability, or non-existence, of such data poses a challenge for the PBT-capabilities of campaigns. But this challenge should by no means imply a PBT-operation in Europe is impossible. We would argue that while the lack of access to voter lists makes it very difficult to achieve the same level of granularity when engaging in PBT as in the US, by using other commercially available or self-collected data, it can be possible to come reasonably close to the desired level (which may be more modest for European campaign leaders than for American ones). The extent to which there is an actual difference between the degree of granularity between US and European campaigns, however, is outside of the scope of this paper, as we focus on perceptions and strategies of campaign leaders.

System level context also affects infrastructure. For instance, should the groundwork be spread equally across the nation, or focused on a number of battleground states? Also, a campaign operating in a heavily regulated context is likely to encounter some legal barriers. For example, because of campaign financing regulations (may influence resource factors), and data protection regulations (infrastructural factors). The absence of regulations, conversely, can facilitate PBT. Cultural context, finally, can influence campaigns' ethical considerations regarding PBT. Campaigns operating in a culture that favours privacy, for example, can be expected to avoid (or use less-invasive) PBT-techniques than campaigns run in a culture in which privacy is less important. In sum, there are several factors, both on campaign and system level, which can form a barrier or facilitate the extent to which campaigns are able to use PBT-techniques and how they use them.

Extending existing research to a European context, we have developed and will apply an improved model (applicable in different electoral contexts) to analyse barriers and facilitators to innovative PBT-practices by political campaigns. As the context of the research case differs from the US, we expect to contribute to the framework and to shed light on how contextual factors influence innovation of political campaigns. Furthermore, in answering our research question, we provide insight into the way political campaigns in a multiparty democracy organise, communicate and innovate. Given these considerations, our key question is: What barriers and facilitators for the adoption and use of PBT-techniques do Dutch political parties perceive?

Method

This study focuses on campaigns in the Netherlands because of the national elections taking place in the research period (15 March 2017), the advanced technological infrastructure (Coy, 2015), and the interesting contextual factors. The Dutch electoral system is one of open list proportional representation (PR), in which all members of parliament come from one nationwide district (Lijphart, 2012). This means that in the Netherlands, every vote counts equally. Moreover, the system of PR (and the very low de-facto threshold) enables a relatively large number of political parties to run in an election. 28 parties participated in the 2017 national election ("Partijen nemen deel", 2017). Of these parties, 13 actually gained a seat in parliament ("Officiële uitslag", 2017).

The Dutch national elections have a relatively high turnout: around 80% in the previous two elections ("Officiële uitslag", 2017). But where the US presidential campaigns can spend hundreds of millions of dollars (Narayanswamy, Cameron, & Gold, 2017), the Dutch campaign with the biggest funds (VVD) has no more than 5 million dollars to spend. And even if the budgets were sufficiently large, the question is whether voter data would be usable for a political campaign. The Dutch data protection law categorises political preference as sensitive personal data. This means that campaigns are only allowed to process such information if the potential voter explicitly gives permission to do so.

Finally, as party membership in the Netherlands steadily decreases, political campaigns can rely less on their members to do labour-intensive tasks (such as canvassing). In 2016, the number of party members of all political parties combined, was at its lowest point since the second world war. Although this number has picked up slightly since, party membership is still quite low ("Membership Dutch parties still low", 2017).

After approval from the ethical committee of the University of Amsterdam, we carried out eight in-depth interviews with campaign leaders. We interviewed 11 campaign leaders, belonging to eight political parties in total (three interviews were double-interviews). In addition to this, we held two background interviews (with one local campaigner for the municipality of Amsterdam and one political consultant offering PBT-services). The eight elite-interviews on average lasted 53 minutes. Two were conducted by phone, the others face-to-face. We took a qualitative research approach for several reasons: the small group of people concerned with the coordination of political campaigns in the Netherlands, the lack of knowledge on this topic in the Netherlands, and because interviewing is a suitable method for understanding the mechanisms behind and perceptions of a phenomenon (Boeije, 2005). As we want to understand how campaigns see PBT, what they are actually doing, and how they perceive possible barriers and facilitators to the adoption and use of PBT-techniques, the interview is a suitable data-collection method. Using an interview guide (see appendix A), we held semi-structured interviews, allowing follow-up questions.

Interviewees

We selected the interviewees via purposive sampling. Campaign leaders qualified for an interview when they had a coordinating role in the campaign and were campaigning for a party that gained at least one seat in the 2012 national parliamentary elections. Eleven campaigns satisfied this second criterion (see Table 1). We contacted interviewees via email, explaining the objective of the study. The interviewees signed an informed consent document before the interview started. We also promised the campaign leaders anonymity, and confidentiality until after election day (15 March 2017). By doing so, we tried to provide the interviewees with a safe environment in which they felt like speaking freely, and without concern of somehow 'leaking' strategic information. Because no information would become public before election day, the risk that interviewees might provide biased information due to a strategic agenda was minimised. Another large advantage of interviewing the campaign leaders before the elections took place, is the prevention of hindsight bias by the interviewees. Unfortunately, we were unable to convince three parties to comply (VVD [right on the political spectrum], PVV [right-wing nationalist party], PvdD [Party for the animals; left-wing]). These three parties were unwilling to cooperate, either because they still found the risk of leaking their strategy too large, or they did not offer an explanation.

Analysis

Using a grounded theory approach, this study has passed four phases: the exploration phase, the specification phase, the reduction phase, and the integration phase (Wester, 1995). In the exploration phase, two background interviews took place (with a campaigner for the municipality of Amsterdam and with a political consultant offering PBT-services). These were coded using AtlasTI, 'tentatively labelling' relevant information (Glaser, 1978). Thereafter, the first interviews with campaign leaders took place. These were transcribed and open-coded. Furthermore, fellow researchers also coded these interviews and discussed the content (peer debriefing). In the next phase, new interviews took place and the data was subject to axial coding. The first dimensions were identified (e.g., what forms a barrier and what facilitates the use of PBT-techniques?). The reduction phase saw the emergence of the core category (innovation). In the integration phase, we completed the conceptual framework, finalised our analysis, and had the campaign leaders approve the quotes used (member checking). This means the campaign leaders agreed with the way they were quoted, and with the publishing of the names of the political parties. Member checking increased the willingness of campaign leaders to cooperate with the study.

Results

We first describe the field: to what extent do campaigns use PBT-techniques? Then we explain differences between parties by focusing on the five campaign level factors concerning the use of PBT (resource, infrastructural, organisational, structural electoral cycle factors, and ethical and legal concerns). Finally, we zoom out to the system level and discuss the influence of contextual factors (electoral system, regulatory framework and culture) on the adoption of PBT-techniques.

PBT in Dutch campaigns

As campaigns in the Netherlands can have recourse to relatively detailed public census data and detailed election results, all campaigns, to some extent, adopt a PBT-approach. Furthermore, Facebook is an important tool for all parties, but the parties differ in how they use Facebook's capabilities. Some parties occasionally post content targeted to broad age groups, while other campaigns frequently post content tailored to more specific groups. Two campaigns stand out, as they have developed their own PBT-tools, which they can use to continuously refine their knowledge of the electorate. We will now use our model to explain the differences between campaigns.

Resource factors

All campaigns cite financial costs as a barrier. Table 2 shows that budgets are modest, and differ between parties.

Source: parties' annual financial reports, on file with authors.

These small budgets form a barrier for the cooperation with expert political consultants (such as Blue State Digital) to enhance their PBT-operations. Parties refer to the financial costs as the main reason not to hire consultants. Campaign leader 1 of the Social Democratic Party (PvdA) explains why he does not work with Blue State Digital (BSD):

Their system is very expensive, that's a factor. And you need the people to carry out the work for you. In an ideal world, such a cooperation would be really cool though."

Liberal democrat party D66 agrees: "because it costs a lot of money and we don't have that kind of money. And if we spend it on a consultant, we can't spend it on the campaign itself."

The same barrier appears when campaigns speak about other technological means, such as canvassing apps, allowing campaigns to directly process information from canvassers. Christian Democrats CDA, for instance, would like such an app. Campaign leader 8: "yes, but that would demand a financial investment that we can't afford." Green Party GroenLinks has a contrasting perspective: "I believe it usually costs around €100,000 to build an app such as our own. (..) We, however, paid our programmers two crates of beer and 40 pizzas." Several facilitators help GroenLinks and also socialist party SP to overcome this barrier of financial costs. First: the personal network of the campaign leader. This facilitator is especially prominent for GroenLinks, where campaign leader 7 employs his own network to optimise the BSD-systems, but also to help him with setting up other parts of the campaign:

We had to adjust it [the BSD system] somewhat for the Netherlands. The people with whom I did so, Swedish folks ... they are simply a little network of people of around my own age, and some people who are a bit older and have already set up a similar campaign in their own countries. A guy who set up the grassroots organization for Trudeau, for example, he's a couple of years older than I am, but I Skype with him to talk about how I should handle certain things.

Campaign leader 7's personal network plays (or at least played) an important role in cheaply setting up technological tools and creating content:

Through the network, I'm aware of the crowdfunding streams for a normal campaign. (..) I'm meeting a friend tomorrow, who has experience with mail flows. (..) I have a network of volunteering writers, poets, freelance journalists who write for us for free. (..) So partly, I just have a good personal network.

However, as PvdA notes, having lots of data is of no use if you don't have the capacity to use it. GroenLinks tries to overcome this barrier by organising their campaigns, to some extent, in a citizen-initiated manner (Gibson, 2015). A citizen-initiated campaign (CIC), devolves "power over core tasks to the grassroots" (p.183). As campaign leader 7 puts it:

Grassroots is about creating an infrastructure to enable as many sympathizers as possible to volunteer as canvassers on a large scale. So voter contact on a large scale, but also – and that's Bernie's [Sanders] lesson – to have places in which a few people make stuff by themselves without us having any control over it. (..) Embracing people's creativity without managing it.

SP has less need of a citizen initiated campaign, because of their relatively large number of active party members. "The big difference [with GroenLinks] is the fact that we already have the volunteers. Many other parties lack the numbers. We have thousands of party members who gladly canvass for us two weekdays and on Saturday as well." [Campaign leader 11, SP].

Infrastructural factors

Having a good infrastructure allows campaigns to actually collect data and send tailored messages. What kind of PBT-infrastructure can parties rely on and how does it facilitate their use of PBT?

All campaigns use the PBT-infrastructure Facebook offers, although some more than others. Nearly all campaigns use its lookalike audiences function to find new potential voters. Campaign leader 8: "we search for profiles of people who look like the ones who’ve already liked our Facebook page, and then serve them with advertisements." Campaigns also look at people who like pages that are close to the values of the political parties. Christian party ChristenUnie, for instance, tries to target voters who like the page of evangelical broadcaster EO. So does the Calvinist Political Party SGP, which tries to find out people’s interests on Facebook: "For example… farming, or Israel, off the top of my head; you try to approach people along the lines of their interest, or the region in which they reside." [Campaign leader 9]

Some campaigns also employ ‘dark posts’, a Facebook function that enables campaigns to opaquely target specific audiences, while its messages are not visible to untargeted Facebook users. Campaign leader 1 exemplifies:

We’ve managed to get something done related to gas extraction in Groningen. It doesn’t make sense to share that on the national Facebook page, because it was only important news locally. So we put out a dark post, only for Groningen residents. Sometimes we can specify it even more.”

Using Facebook for PBT-purposes, campaigns do not actually gather or own data themselves. There are a few campaigns that do gather their own data, by using canvassing apps. Campaign leader 7:

We use the election results per voting location and use that information to establish the GroenLinks mindedness of a neighbourhood. Then we can prioritise which addresses to visit and which to ignore. When we visit addresses, our volunteers use the app to answer the following questions: 1. Is anyone home? 2. Does she want to talk? 3. Is she going to vote? 4. Is she planning to vote for GroenLinks? 5. What is the most important theme to her? 6. How GroenLinks minded was she? If she considers to vote for GroenLinks, two questions follow: 1. Do you want to stay informed of our campaign by e-mail? 2. Can I have your phone number, so we can ask you to do canvassing talks?”

The GroenLinks app facilitates large scale collection of information about people's political preferences, thereby informing strategic decisions. Also, the personal data can facilitate accurate PBT on an individual level. The secondary objective of the app is to provide an infrastructure for volunteers to campaign on their own terms, whenever they feel like doing so:

Our app, built by hackers, enables others to campaign for us. (..) Someone in [small town] Lutjebroek can install our app and go ahead and work for our campaign. No campaign leader needed. [Campaign leader 7]

Some campaigns monitor the visitors of their own websites. Campaign leader 1: “What are people searching for on our website, how do they get to our website, how much time do they spend, (..) which button should you colour red? How does that work?” At the time of the interview, CDA was not yet tracking their website visitors, but: “we’ve just migrated to a new website, on which we want to start collecting more data on our visitors. I’m curious what kinds of people are visiting the website. And what kinds of people don’t, and therefore have to be reached through different channels.”

SP has built a system which combines previous election results, census data and their own membership Constituent Relationship Management (CRM) data. Plotted on a Google Map, they can identify interesting areas for them to canvass. This system facilitates efficient use of means:

We would do nothing more happily than knocking on every single door in every city, but unfortunately, we do not yet have that kind of manpower. So we do an analysis: What kinds of neighbourhoods are especially interesting for us? We have built our own system to help us make that decision [Campaign leader 11].

Organisational factors

Circumstances within the campaign’s organisation itself can form a barrier for the uptake of PBT-techniques. Less innovative parties, for instance, do not have a dedicated data, tech, or digital department. As a campaign leader notes: "The department responsible for that [tech/data/digital] is our Communication department. So that's four or five people. And sometimes someone of the department picks it up, but there's not one specific person who's responsible." This contrasts with GroenLinks, which has a Digital and Grassroots department and with SP's Digital department.

The "state of mind" within a campaign can also be seen as a barrier: "In the sense that internally, people are still very much inclined to think offline. The culture within the campaign is quite offline." [anonymous campaign leader]2

New leadership and younger staffers can play a facilitating role in political organisations. Campaign leader 11 argues that, because he is young, their new party chair brings a more tech-savvy vision than his predecessor. According to the campaign leader, younger staffers are more likely to implement tech and data in their work procedures.

A final organisational barrier is the primary goal a political party pursues. Campaign leader 10:

Maybe the strange thing about SGP is that we do not care that much about seat maximization. For us, it's about the impact of our principles. And sure, we would rather have four seats than three, but if we have to settle for three seats: that's fine too. And that's, in my opinion, a reason why we have a feeling like: do we really need data?

Structural electoral cycle factors

These circumstances are largely beyond the control of the campaigns, but they can influence the uptake of PBT-techniques. Campaign leaders see the PBT-actions of other political campaigns as a motivational factor. As campaign leader 11 notes about the development of their app: "I've looked a little bit at how GroenLinks have their app and canvassing system." Or as campaign leader 2 concludes: "If every party does it, you don't win very much by it. But if you're the only party that does nothing..."

Ethical and legal concerns

Especially D66 and the seniors’ party 50PLUS take a principled stance against the collection of data and the use of PBT. Where D66 presents itself as a privacy champion and therefore will never gather and use information about (groups of) voters, 50PLUS campaign leader 6 warns about the risk of irresponsible use of the data gathered by the "almost stalking of people", which he calls "morally irresponsible".

Furthermore, a lack of internal legal expertise appears to contribute to a feeling of legal uncertainty, which affects the likelihood of adopting PBT-techniques: "Legislation has grown so very comprehensive and complex. It's almost impossible to cope for us as a small organisation." [Campaign leader 5]

While ethical and legal concerns can form a barrier, a left- or right-wing orientation does not seem to be instrumental therein. After all, we have seen left-wing parties GroenLinks and SP develop relatively advanced PBT-tools. And we have seen right-of-centre party CDA express clear interest in advancing their own PBT capabilities. At the same time, left-of-centre 50PLUS and right-of-centre D66 both oppose the use of PBT.

System level

Electoral system

Although the Dutch one district PR-system should make for a rather equal distribution of campaign efforts, campaigns still divide the country into smaller areas of interest called 'key areas'. These areas differ per party, but do receive a relatively large part of campaign attention. Campaign leader 1 describes these as areas: "where we know the turnout is low, but the number of PvdA-voters is high". All campaigns use data provided by the Electoral Council, showing the election results per party, per voting location to establish key areas. Campaign leader 3 explains:

Using that [the election results], you see: Okay, we do well in this neighbourhood or this street. And then you combine that information with the CBS3 data, to find out what kind of neighbourhood it is, what kind of people live there, what are their backgrounds, how much do they earn, what does the family composition look like, et cetera.

Facilitated by these public data, campaigns enrich their knowledge of specific areas. A next step would be to use those data to make personalised appeals to (subgroups of) people living in those specific key areas.

Regulatory framework

Although the Netherlands would qualify as a minimally regulated environment (Plasser & Plasser, 2002; Esser & Strömbäck, 2012), campaigns all experience regulatory pressure and legal uncertainty on a system level. They cite an abundance of regulations, forming a barrier to their ability to innovate.

The technological developments have been taking place so very quickly. And, in that timeframe, to adjust all your procedures and everything. And also to meet the privacy regulations, I think many parties face a huge challenge in that respect." [Campaign leader 4]

Campaigns sometimes face a dilemma, having to decide between innovative techniques and privacy regulations. Campaign leader 11:

Regulations sometimes are unclear, which leads us to decide to go for the safe option because you do not know where the red line is. And you never want to abuse someone's personal data. So yes, regulations sometimes cause us to hit the brake and that's a good thing.

Culture

There is a recurring worry about the perceived low level of political knowledge of the average Dutch voter. PBT-techniques can facilitate campaigns' efforts to convince or educate such low-information voters, for example by "having a conversation with someone, especially if you share some characteristics," [Campaign leader 7] or by interesting "people for things that are relevant to them and to make them aware of the political dimension of those things." [Campaign leader 8] Campaign leader 3, in contrast, concludes that the electorate's low level of political knowledge (together with the perceived volatility of the electorate, the decrease in political trust, and their focus on persons instead of parties) forms an insurmountable barrier, making PBT-techniques irrelevant.

Discussion and conclusion

In the 2017 elections, used here as a case study, all campaigns use PBT through Facebook, but some parties are more advanced than others, and have even developed their own PBT-tools. We have established what the main barriers and facilitators for PBT are, using five factors on a campaign level and three factors on a system level. Not only does this study shed light on the conditions under which these barriers and facilitators manifest themselves, it also gives insight into their different workings across parties. Our study provides information about the data collected by parties and the PBT-techniques used to attract voters. We demonstrate how personal networks and cognitive diversity within a campaign can level barriers. We show how PBT is not only perceived as useful for campaigns in a FPTP-system, but in a PR-system as well. And we show how regulatory pressure is perceived as an obstacle and as a welcome 'normative red line'.

A triangulated research approach can improve our understanding of the campaign leaders' constructs. Observation of their (use of) PBT-tools and how these tools help campaigns make strategic decisions, can give more insight into the workings of these techniques. Another approach would be to interview canvassers and identify 'field-level' barriers and facilitators. Furthermore, ideally, we would have spoken to all parties holding a seat in parliament. Unfortunately, three parties did not cooperate. Two of those became the largest (VVD) and second largest (PVV; in a very close field) party. Since we did have access to eight of 11 parties, we are confident about our findings and we do not expect to identify additional factors influencing the adoption of PBT from interviews with the remaining parties.

Compared to related recent studies by Anstead (2017), Hersh (2015), Kreiss (2016), and Nielsen (2012), this study makes a number of contributions. In general, we focus our exploratory research on a PR-system instead of a FPTP-system, and we develop a model that takes system level contextual factors into account. Specifically, unlike Anstead (2017) we have found evidence for equalisation (which occurs when smaller parties take advantage of the internet's low costs and direct communication possibilities, and, in doing so, use the internet as a tool of empowerment [e.g. Margolis, Resnick, & Levy, 2003]). This evidence is especially clear in the case of GroenLinks, which was, at the time of the campaign, one of the smallest parties in parliament (now the fifth party). Furthermore, we provide an insightful point of view into Anstead's question of whether "parties develop data-driven capabilities more rapidly in electoral systems with a tendency towards disproportionate outcomes" (2017; p. 23). In comparison with Hersh (2015), we focus less on how differences in data-availability lead to different strategic decisions, and more on how differences in the perception of campaign level and system level factors lead to a variation in the occurrence in PBT-innovation. With regard to Kreiss (2016), we have extended his model and applied it to a multiparty democracy. In comparison with Nielsen (2012), we focus solely on the perception of campaign leaders and not on canvassers. Furthermore, we focus on PBT on online as well as offline platforms.

Our attention for the system level factors has enabled us to identify perceived influence of the PR-system on the adoption of PBT. Contrary to theoretical expectations (Plasser & Plasser, 2002), campaigns in a one-district PR-system do identify key-areas that are more heavily campaigned than other districts. These key-areas differ from 'battleground states' in FPTP-systems in the sense that the key-area does not sometimes swing one way and sometimes the other, but rather that potential voters in key-areas are supportive of a certain party, but not very likely to show up at the polls. Campaigns use PBT-techniques to convince these potential voters of the personal relevance of politics and to motivate them to cast their vote. Areas with firm turnout numbers and clear support for a certain party, in contrast, are perceived as less decisive and less of a priority. This leads to a hierarchy of areas, which differs per party. Also, as a PR-system typically leads to a relatively large number of parties partaking in an election, PBT can be seen as an asset for a campaign to organise in a more efficient manner. Moreover, according to the campaign leaders, PBT-techniques offered by Facebook do allow smaller parties a degree of visibility that they are unable to achieve through traditional media.

On a campaign level, in the coming years, we expect more citizen-initiated campaigning (Gibson, 2015) by campaigns low in labour-resources. This requires a solid infrastructure, which opens the door for third party intermediaries offering off-the-shelf infrastructure. In this regard, it would be interesting to track the development of PvdA, which has suffered its biggest loss in history. This critical event could lead to the prototyping (Kreiss, 2016) of GroenLinks' innovative campaign by PvdA. As the party's chairman has resigned, the door is open to a more cognitive diverse party structure (Du Pre, 2017; De Vaan et al., 2015). Of course, these developments might apply less to parties that are officially more cognisant of campaign ethics (e.g. D66). This is why ethics and legal aspects are important factors to take into consideration. It would be interesting to see how these campaigns act as PBT-capabilities of rival parties improve. Their self-imposed barrier can limit their future chances, but can also attract voters growing more aware of the value of privacy. In the former case, this could lead to an overhaul of their privacy principles, or perhaps to a legislative push towards the restriction of PBT (similar to Hersh, 2015). In the latter case, campaigns can be expected to develop innovative non privacy-invasive campaigning techniques. Either way, our model would provide tools to study the process.

So 'can political parties campaign in Europe as they do in North America' (Bennett, 2016)? We would say ‘mostly yes’. We agree with Bennett (2015) that there are important differences between the US and Europe, and indeed, they influence how PBT is used. But based on our findings, we are hesitant to conclude that those differences (severely) constrain the export of PBT-practices to European multiparty systems. We have shown that relatively small campaign budgets do not need to bar parties from engaging in PBT-practices (or even from cooperating with BSD, an 'expensive' American political consultancy). The same is true of the electoral system: campaign leaders generally perceive PBT-techniques as useful in a PR-system. What remains is the relatively strict Dutch data protection law, labelling political preference as 'sensitive personal data', which can only be processed with explicit consent from the potential voter. 'Explicit consent', however, sounds harsher on paper than it is in practice and is easily achieved (e.g., Beales & Muris, 2008; Calo, 2012; Joergensen, 2014). Of course, because of data regulations and/or their non-existence, European campaigns are unable to consult voting lists showing whether an individual showed up at the polls in the last elections. In most European countries, the electoral register is inaccessible to political parties. One might argue that, from a campaign's perspective, US voter data are superior to European voter data. We would argue that European data are different, but they do not bar European campaigns in the use of PBT-techniques. Dutch campaigns, for instance, can (and do) rely on election results on voting booth level (which comprises a couple of streets). They can (and do) combine these results with detailed, accurate, and a multitude of data about the neighbourhoods surrounding those voting booths. And then there is Facebook, facilitating easy targeting of its users with personalised messages. As potential challenges for democracy come with PBT, such as ignoring 'less valuable' citizens (e.g. reliable non-voters), more research into the workings and effects of PBT is needed.