Restrictions on data-driven political micro-targeting in Germany

1. Introduction

Despite the manifold opportunities presented by online campaign tools and multimedia channels nowadays, political parties in the US and Europe seem to have built a resurgent interest in an originally 'premodern' campaign tool to mobilise voters and ultimately generate votes: door-to-door canvassing.

This revitalisation heavily relates to campaigns´ growing possibilities for analysing voter data to gain knowledge about their constituents, identifying their most likely voters and serving up personalised messages through face-to-face conversations. A look at the extensive research literature about data-driven canvassing reveals a sharp contrast in the number of studies conducted in the US to research in European countries, and especially Germany. Therefore, the variation in the institutional frameworks, social and legislative conditions is almost limited to the one-country case and reveals the existence of a research gap regarding data-driven canvassing in parliamentary democracies with strict data protection laws.

This said, the Obama, Clinton and Trump campaigns made data-driven campaigning and micro-targeting known to a broader public. In Germany, data-driven efforts are a subject of controversial public discussion: critics fear manipulations of voters and violations of national privacy laws. On the opposite, supporters hope for a mobilisation of specific target-groups like swing voters or infrequently voting partisans who often abstain from voting.

Based on theoretical predictions derived from the extensively studied US case, we discuss data-driven political micro-targeting against the backdrop of canvassing in the German social, legal and electoral context from a political actor´s perspective. Three central research questions structure our work: (1) What importance do German parties attribute to the use of data for targeting voters in contemporary campaigns? (2) How do German parties use data for targeting voters through canvassing efforts? and (3) What are the restrictions on data-driven political micro-targeting in Germany?

In a first step, we discuss the theoretical, historical and legal principles of data-driven canvassing with a comprehensive review of literature on data-driven campaigning and data protection laws in Germany. Drawing on that discussion, we introduce a framework of constraints in strategic political communication and address our research questions with findings from an analysis of in-depth expert interviews with nine campaign coordinators of parties running in the state parliament of Rhineland-Palatinate in 2016 (Christian Democrats [CDU], Social Democrats [SPD], the Green Party [The Greens], the Left, the Liberals [FDP] and the populist right-wing party Alternative for Germany [AfD]).

We reveal several restrictions on the macro, meso and micro levels that hinder the implementation of sophisticated data-driven micro-targeting strategies in contemporary German campaigns. We argue that these strategies depend on system-level contextual factors, budgetary and legal restraints, specific campaign contexts, organisational party structures and even individual decisions and knowledge on behalf of a campaign’s leadership. Our study is a specific example of data-driven politics in parliamentary democracies with strict data protection laws, but it can also provide insights for comparative theory-building beyond the scholarly discussion of the US-practices. Furthermore, it has to be considered that the use of data in political campaigns is a topic of controversial public debates. We want to highlight that our main research interest is to analyse data-driven campaigning from a political actor´s perspective in the German setting by using a theoretical model of comparison and not to normatively evaluate these techniques. However, our concluding remarks contain main arguments of criticism of data-driven canvassing to point at further normative research.

2. Theoretical foundations of data-driven canvassing

With few exceptions (i.e., Anstead, 2017; Pons, 2013) the majority of the literature on the use of data and technology in political campaigning discusses US-presidential campaigns (i.e., Kreiss, 2016; Nielsen, 2012). Besides broader discussions about modes of professional campaigning in Europe (Tenscher, Mykkänen, & Moring, 2012) the literature on the role of data and technology for political parties in countries with strict national privacy laws has not been systematically reviewed yet. Therefore, we present a literature review of data-driven campaigning in Germany with a special focus on canvassing. In this process, we highlight and define crucial terms, give insights into the emergence of modern campaigns´ communication channels and the influence of data.

2.1 Data and micro-targeting in political campaigning

In the context of political communication, we understand micro-targeting as a strategic process, which is geared towards addressing persuadable or mobilisable voters with tailor-made messages while ignoring others. Castleman (2016) distinguishes two terms that are often used interchangeably: “modeling” and “micro-targeting”.

Modeling describes the practice of using algorithms and observed data to build statistical or machine learning models to mine users with similar attitudes and behaviours (clustering) or predict unobserved actions or preferences (predictive modeling). For political campaign communication, models are often built at the individual level using survey data or relevant voter information. On the one hand, compiled information by local and state election authorities (e.g., voter files, structural data sets) contain voters´ addresses, sociodemographic characteristics, voter turnout records or voting histories for specific districts or states. On the other hand, more individual (behavioural) information from a campaign´s own data or from commercial data vendors include donation activity, campaign contact histories, consumer records or digital trace data.

Micro-targeting is a commercial direct marketing practice and refers to the process of making strategic decisions at the individual level about which customer to target with what campaign message (see also the Editorial of the special issue). In sophisticated campaigns, algorithm-based models inform these decisions. Therefore, modeling and micro-targeting are clearly linked, but they are not inseparable as modeling can still be done on other types of data, and micro-targeting can be done in the absence of models. Therefore, different approaches of political targeting can be distinguished: geographical targeting is the oldest -- and most coarse -- approach. It is based on the analysis of precinct-level results from past elections to identify auspicious electoral constituencies. Another approach is targeting groups with shared demographic characteristics such as income, religion or occupation. Reliable data about a series of demographics is needed to build algorithmic models for distinguishing individuals for each targeted voter group. Sophisticated targeting-approaches are based on the analysis of individual attitudes, behaviour and values of the electorate. Building on the assumption that these characteristics are more coherent with the voting decision, they promise satisfactory targeting results if reliable data is used.

Election campaigns use modeling and micro-targeting to improve the efficiency of how campaigns with limited resources are run and communicate with voters. These techniques can help campaigns to minimise uncertainty by better understanding how different populations view issues, help to find voters who can be persuaded or mobilised with different campaign messages through different communication channels. Furthermore, the use of technology (e.g., digital platforms, apps) can improve these communication attempts and support the organisation and evaluation of a campaign.

2.2 Implementing data into German campaigning

Although the roots of data-driven campaigning in Germany can be dated back to the times when public opinion polls were on the rise in the early 1950s (Noelle-Neumann, 1955) tentative attempts with more sophisticated practices took place in the context of the federal election 2005 with grassroots campaigns from CDU (“teAM Zukunft”), SPD (“wirkaempfen.de”) and the Greens (“Mach Mit!”) (Jucknat & Römmele, 2008). These campaigns were carried out “completely isolated from other campaign activities” (Hennewig, 2013, p. 160) in distinct constituencies and lacked a “proper communication and entrenchment in the main campaign organisation” (Heinrich, 2013, p. 176). As a result, only widely scattered campaign aides and party sympathisers were reached and could not be used as strategic communication multipliers. As a result of the attempt to import campaign practices from the US-election cycle 2008, various improvements were made in the German federal election 2009: both CDU and SPD connected their organisation platforms (CDU: teAM2009.de; SPD: meineSPD.net) with databases to build up a central organisation structure and implement the grassroots activities into the main campaign.

However, the platforms´ level of interactivity, dynamic and participation clearly lagged comparable platforms like Barack Obama´s MyBO in 2008 (Lilleker & Jackson, 2011), resulting in a more or less ineffective mobilisation of volunteers throughout Germany. In addition, first attempts were made at targeting potential voters within demographic groups or mobilise sympathisers in geographical regions using so called 'mobilisation indices'. Nevertheless, voters were only marginally aware of target specific communication channels: only two percent of voters were conscious of email, social media or SMS contacts while a majority was aware of posters (75%) or TV ads (50%) (Schmitt-Beck & Wolsing, 2009, p. 51). In that regard, German campaign strategists started to understand the problems and complexity of implementing these new campaign practices and technologies from one national, legal and party context to another. Although German parties began to experiment with more sophisticated data analysis and technologies in the federal election 2013, they still struggled with staff, time, expertise, money and strict data protection regulations to establish a high-tech driven hunt for individual voters. As a result, the German targeting efforts were more of an exploration in the dark instead of a target-oriented campaigning strategy.

These developments go hand-in-hand with social, political and technological changes, which ultimately result in a political actor´s need for an enhancement of electoral communication strategies mostly described as “professionalization” or “modernization” (see Kamps, 2000). The evolution of data-driven campaigning can best be described along the adapted model of ideal campaign types by Magin, Podschuweit, Haßler, & Rußmann (2017) (see Table 1). It focuses on four ideal campaign types, which emerged within a certain time frame when the essential technological possibilities became available: 1) Partisan-centered campaigns address the core party members and partisans with face-to-face interactions, partisan press, newspaper ads, radio broadcasts and posters. Thus, campaigning was rather based on local and decentralised strategies. 2) With the establishment of nonpartisan media and limited-channel television, the range of mass media coverage expanded radically. Mass-centered campaigns emerged and provided the opportunity to address disperse masses by unidirectional messages while also being supplemented by partisan-centered campaigning

tools. As a result, campaigning was increasingly based on national strategies. 3) Target-group-centered campaigns developed while technological progress and historic events led to the increasing importance of individualism and modernism within civil society, echoing in an increasingly fluid electorate. Campaigns started to address voter segments with similar interests via multi-channel television ads and the internet. These campaigns are characterized by “top-down, centralized communication and supplement the previous campaigning tools with party and candidate websites, banner ads and direct mailing by e-mail” (Magin et al., 2017, p. 1702). 4) The establishment of new multimedia technologies and the developments in social structures resulted in an erosion of traditional milieus, reducing commitment of voters to political parties and a fragmentation of media use among the electorate. Thus, political parties needed to mobilise not only hardcore partisans but potential or wavering voters as well, refining the target-group orientation to individual-centered-campaigns with database technology, voter segmentation and micro-targeting as its key features. Data-driven canvassing and the manifold web 2.0 platforms provide ideal channels to address individual voters directly with tested tailor-made messages, all while bypassing the mass media threshold.

However, real campaigns will hardly ever meet these ideal types but rather use a mix of all campaigning tools depending on the strategic direction campaigns choose, who it wants to address with what message and under what restrictions.

2.3 Making use of data in German campaigning, drawing on the example of canvassing 2.0

We showed that improved capability to target individual voters offers campaigns an opportunity to concentrate their resources where they will be most effective. As Nickerson and Rodgers (2014, p. 71) put it, “[o]ne could argue that the growing impact of data analytics in campaigns has amplified the importance of traditional campaign work”. Thus, door-to-door canvassing is currently experiencing a renaissance among parties and election campaign strategists in diverse political systems. It was primarily during the 2004 presidential elections that data-driven canvassing emerged as one of the key instruments in the Howard Dean and George W. Bush campaigns (Kreiss, 2016) and sparked worldwide attention as decisive factor behind Barack Obama´s successful presidential bids in 2008 and 2012 (Nielsen, 2012; Kreiss, 2012; 2016).

However, canvassing is by no means an invention of modern election campaigns. On the contrary, it dates back to the rise of contested elections in the UK and was also used by the National Socialist Workers’ Party (NSDAP) during the rise of Nazi-Germany (Mühlberger, 2004). After World War II, canvassing was almost suspended until the 1960s, when German parties started using it in local elections and switched their canvassing strategies away from persuading eligible voters to identifying sympathisers or supporters and mobilising them. While television became increasingly popular since the early 1950s, parties shifted their resources from the ‘ground game’ to mass market advertising, with canvassing seen as “relics of the past” (Beck & Heidemann, 2014, p. 268) and an elaborate ritual bringing gratification to campaign aides and supporters, but making no difference to election results (Denver & Hands, 2013). However, with declining electoral turnout, an increasing diversification of electronic media and the corresponding fragmentation of media use among the electorate, the generic mass media´s one-way scattergun approach of campaigning became increasingly unsuccessful in reaching, let alone mobilising or persuading specific target-groups. Therefore, door-to-door campaigning attracts attention from campaign strategists. It allows for unique contacts that stand out from the media torrent, reaches a clearly defined universe of individual targets, and has measurable effects (see, among others, Green & Gerber, 2008; Michelson & Nickerson, 2011).

After Barack Obama´s re-election campaign in 2012 was lauded for its sophistication in ground organisation, data analysis and micro-targeting, German parties tried to find ways to translate door-to-door canvassing into the German electoral setting. First attempts with data-driven canvassing were made at the federal election 2013. The aim of these canvassing campaigns was to reach out to the voters, get votes from the undecided and draw those back to the party who defected in the last election (2009). The sentiment of going to the people instead of them coming to the party is a big shift in German campaign philosophy. German voters are used to only entering in contact with party members at a rally or stop by a party information booth in public areas. A representative survey by YouGov (2017) also reflects this sentiment and shows that a majority of Germans (65%) are still uneasy about opening their door to a political party.

3. A framework of constraints in strategic political communication

Against the backdrop that certain social, political, legal, technical and cultural contexts underlie strategic political communication, we adapted the analytical framework by Vowe and Wolling (2000) to explain constraints related to decision making in campaign communication in different contexts. A key statement condenses the underpinning theory into one sentence: “Every strategist risks the success of his venture if he neglects the conditions which determine his actions.” (p. 58)

Drawing on this, we argue that there are three constraint dimensions for the strategic use of certain campaigning strategies and tools: 1) external dimensions (Macro level), 2) internal-organisational dimensions (Meso level) and 3) individual dimensions (Micro level) (see Table 2). The model is suitable either to explain a particular case, for example a campaign in one country, or to compare campaign cases in different nations. In this paper, we present empirical findings of data-driven canvassing in Germany and therefore use this framework and its leading categories for interpreting data collection and presentation. In the following, we present each dimension along specific subdimensions:

1) Macro level of constraints

1a) Electoral and party system: Generally, presidential government systems offer more incentives for strategies of individual-centered-campaigns than parliamentary systems, which are characterised by party-focused styles of campaigning. This is especially the case for the German electoral system, where two votes are casted: one for a direct candidate (determines by plurality vote who will represent each constituency in the parliament) and one for a party list (determines the distribution of seats in the parliament). In Germany´s multi-party system, parties don’t enjoy much flexibility in addressing (non-affiliated) voters because they need to put distance between themselves and their opponents with distinct ideological profiles and issue agendas. This doesn’t suit the logic of individual-centered-campaigns very well. Furthermore, countries with high levels of party identification and high numbers of enrolled party members can potentially rely on stronger support for party-driven campaign operations. Thus, Germany´s tradition-rich ‘programmatic parties’ are relatively deeply rooted within society and are not so dependent on targeted practices. Lastly, German parties are mostly financed by membership fees and state funds which derive from the outcomes of the elections (statistics can be found at Deutscher Bundestag, 2017). Although donations by private persons and organisations are also allowed, they are much lower than in the US.

1b) Political culture: Germany can rely on comparatively high turnout cultures and constituents with political beliefs that are intact. However, recent studies observed a general process of partisan de-alignment also in Germany (Dalton, 2004), meaning fewer people have fixed attachments to political parties and fewer are now members of the same. This leads to a decline in voter turnout and an increased electoral volatility and makes elements of individual-centered-campaigns more likely to emerge. In addition, historical divisions between the former East and West still play a crucial role in modern German politics with old borders defining party loyalties and economic divides reflecting party associations. Deriving from political history, there is a great sensitivity among the German public about their political views, accompanied with a general distrust of intrusive political marketing techniques.

1c) Legal foundations: In contrast to US campaigns, where privacy laws are comparatively weak and fragmented, especially because of the dominant role of the First Amendment (Bennett, 2016), Germany is covered by very strong data protection rules. Although it is allowed to collect information on voters via administrative offices, these data remain on a basic level. There is particularly a lack of when it comes to more valuable information like individual participation in past elections and registrations in a party. Additionally, German data protection laws do not allow combining individual data with further information from other databases. The legal source important for data use in political campaigning is the Federal Data Protection Act (BDSG). The law determines fundamental principles for data collection and processing by public authorities and private institutions: public and private institutions shall avoid unnecessary data collection (data avoidance) and regard data economy (§3a BDSG). The collection of data is only allowed to fulfill a specific purpose (§28 (1) BDSG). Political parties can gather data if they are necessary for their organisational activities. However, this is limited to the personal data of party members or persons who have regular contact to the organisation (§28 (9) BDSG). Furthermore, data can only be collected when affected persons explicitly agree (§4a BDSG). People also have the right to have their data rectified, cancelled and blocked (§6 (1) BDSG). In contrast to the US, where political databases are updated regularly, the German law forbids a storage of personal data for the long-term. Data has to be deleted when the purpose is fulfilled (§35 (1) BDSG). In addition to that, political parties are not allowed to store data on racial or ethnic origin and political, religious or philosophical beliefs (§35 (2) BDSG). This leads to the problem that German campaigners cannot build larger databases for micro-targeting. Lastly, German data protection is integrated into a larger legal framework: the EU Charter of Fundamental Rights (Article 8).

2) Meso level of constraints

2a) Party culture: German parties differ highly in their organisational structure or culture. The SPD is an example for a highly democratic internal organisation structure which also extends to its processes (Langenbacher & Conradt, 2017, p. 140ff.). More than 12,000 local associations (Ortsvereine) make up the core of it. These are grouped together into sub-district, district, state and youth organisations. The subdivisions determine SPD´s policy and elect their party leader and the members of several committees, like the Executive Committee, which in turn elects SPD’s 13-member Steering Committee (Präsidium). This democratic structure is vertically and horizontally intertwined, making the SPD a highly democratic but also slow working apparatus. This can pose a problem in time of elections, in which fast and sometimes disagreeable or policy dissenting decisions must be taken. As the archetypal pragmatic and office-seeking ‘catch-all party’, the CDU is an example for a decentralized party organisation structure (Langenbacher & Conradt, 2017, p. 148ff.). The state, district, county and local organisations together with other auxiliary groupings are mostly independent of any national or central control. So, success in election campaigns often hinges on the questions of whether the subnational and auxiliary groups can be mobilised at the national level and if a national campaign strategy can be applied throughout all subdivisions.

2b) Personnel and financial resources: Sophisticated data-driven canvassing is money intensive and requires trained personnel and staff. As mentioned above, German parties have a rather small budget at their disposal. In today’s convergent media environment, German parties must make strategic decisions about the channels and tools they want to invest in without neglecting others. For a data-driven canvassing campaign, parties must invest in modern technologies, data, analytics, skilled staffers and a nationwide field organisation. Moreover, canvassing needs trained campaigners and volunteers. In comparison to the US, German parties can rely on established party structures with paying party members. While the SPD and CDU have close to 430,000 members each, the smaller parties have around 60,000 (The Greens) to 55,000 members (FDP, The Left), whereas the 2013 founded AfD has 26,000. With regards to canvassing, this means an advantage for the bigger parties.

2c) Technology and infrastructure: As Kreiss (2016) points out, the internet emerged as a hub for digital technologies, revolutionising the organisational “back-end coordination” (Hindman, 2005) of political campaigns. Especially individual-centered-campaigns need innovative and fast ways to connect with volunteers, supporters and political staffers. In addition, data processing, strategy development and message distribution across platforms and voters are highly dependent on modern infrastructures and technology. In that regard, online platforms, apps and especially databases serve as the backbone of individual-centered-campaigns. To have recourse to maintained databases and integrate them into the campaign makes for huge advantages in target-oriented communication efforts. Therefore, specialists and consultants with expertise in the use of digital tools have become ever more important for developing new strategies for resource allocation, evaluation and organisation of political campaigning (Kreiss, 2012). Although German campaigns show an increasing number of outside consultants in recent years, the employment of data analysis and technology experts is still subordinate.

3) Micro level of constraints

3a) Politicians & campaign managers: individual-centered campaigns need highly professionalised campaign organisations coordinated by central headquarters, subordinated offices and highly trained staff. Therefore, campaign managers bear responsibility for the strategies as well as for the tactics in the campaign. Constraints can occur when managers do not have the skills or expertise to implement data-driven methods or refuse to do so for other reasons, like ethical considerations or not enough experience in campaigning (Maarek, 2011). Politicians can be another constraining factor. As personifications of party ideologies, they are in the centre of the campaign organisation and have the power to decide over every aspect in a campaign. If politicians refuse to use data-driven tools, the whole campaign organisation is affected by that decision.

3b) Volunteers / Party members: canvassers are crucial in door-to-door campaigns. Constraints can derive from the fact that most volunteers have little access to the campaign headquarters and do not have the competence in postmodern political marketing. Missing professionalisation can then lead to a false implementation of the campaign plan or even to a breakdown of door-to-door activities in one region. Hence, in highly professionalised campaign assemblages the main tasks of the field coordinator is to stay in contact with volunteers and to organise activities (Maarek, 2011).

3c) Voters: Even if data-driven canvassing is working from the planning to the organisation of volunteers, there is a last constraint dimension, namely the voter. Constraints reach from temporal aspects, like the moment of the visit, to attitudinal factors, such as the political mindset of a targeted person.

4. Research questions

Against the backdrop of the comprehensive discussion about the theoretical, historical and legal principles of data-driven canvassing and the presented framework of constraints in strategic political communication, this article addresses three research questions:

1) What importance do German parties attribute to the use of data for targeting voters in contemporary campaigns?

Studies about data and technologies in German campaigning are scarce, so we firstly wanted to know what the involved German campaign strategists think about the role of data for their campaigns. We expected a high interest in these strategies from all parties, but depending on the individual expertise and experience, the attributed importance may vary.

2) How do German parties use data for targeting voters through canvassing efforts?

We secondly wanted to gather more detailed insights about the state of data-driven targeting strategies in contemporary German campaigns. Here, the type of used data, its analysis and strategic implementation into the campaign communication is of special interest. In the light of the party system, party funding and party culture variations between German parties may occur.

3) What are the restrictions on data-driven political micro-targeting in Germany?

Drawing on the assumption that political campaigns are shaped by its conditions and contexts we expect restrictions on the macro, meso and micro level that hinder the implementation of sophisticated data-driven micro-targeting strategies in contemporary German campaigns.

Our study will gives important first insights into the state of data-driven campaigning in Germany as parliamentary democracy with strict data protection laws. Therefore, our case might be useful for generating comparative theory and hypotheses beyond the scholarly knowledge of these strategies in the US context.

5. Methodology

Research on political campaigning means difficult access to the field in most cases. During and before political campaigns, campaign officials and staffers try to hide their strategies from the contestants and public (Jungherr, 2017). Besides experimental approaches to quantify the effects of certain communication efforts (see, among others, Green & Gerber, 2008; Michelson & Nickerson, 2011), studies on political campaigning either use theoretical or qualitative approaches (Jungherr, 2016; Nielsen, 2012). However, the studies of Jungherr (2016) and Nielsen (2012) provide impressive proof that research about campaigning is possible and yield insights into processes, interactions, organisational constellations, communication channels and tactical decisions in the course of the election campaign.

Federal elections have great significance for national politics, like possible blocking powers for federal legislation or as “useful indicator of movements in public opinion […] and […] a regional poll under realistic conditions […] taken as a serious indication of what may happen throughout the entire country” (Drummond, 1967, p. 385). However, there are some differences between German first order (federal election) and second order elections (federal state elections):

1) In German second order elections voter turnout is lower, and the willingness to punish the established parties greater than in the case of general elections (Bräuninger & Debus, 2011). This leads to a greater likelihood of new and extreme parties being given the vote, which at the same time means that mainstream parties, and especially ruling parties, must make greater effort to inspire their regular voters to go to the polls – in other words, to mobilise them.

2) Apart from that, local in comparison to federal elections are characterised by the fact that parties even have more limited electoral budgets at their disposal (Lewandowsky, 2013). Thus, direct methods of addressing voters constitute an important alternative to the costly advertising campaigns. 3) On account of a campaign’s professionalisation, federal state elections are less professionalised in terms of innovative communication efforts, data analysis and external consultants.

4) Nevertheless, federal state election campaigns can have recourse to the established party organisation structures on the local level, representing an opportunity for canvassing campaigns.

The study of statewide campaigns is fruitful for several reasons: First, there is still less research on state elections. Hence, this study can contribute to closing a research gap in political communication research in Germany. Second, state elections can be highly instructive for canvassing research: due to the presented facts it can be assumed that all parties in statewide campaigns are conducting canvassing campaigns. Third, statewide elections can be understood as test runs for national campaigns. In the elections for the German Bundestag in September 2017 some further methods of data-driven campaigning were used such as mobile applications. Fourth, the Rhineland-Palatinate election of 2017 ended in a rather close result: The SPD won with 36.2 percent followed by the CDU with 31.8. In several constituencies there were very close majority vote results concerning the direct mandates (Landeswahlleiter RLP, 2017). For instance, in constituency 26 (SPD: 36.3%, CDU: 34.6%), constituency 45 (SPD: 38.1%, CDU: 36.3%) and constituency 9 (SPD: 34.7%, CDU: 32.1%). It is likely that micro-targeting attempts were assumed to be important in these battleground districts. Overall, research on professional statewide campaigning can offer insights into data-driven campaigning on a level where campaign managers are most willing to take part in expert interviews.

In order to gain access to the largely unexplored phenomenon of data-driven canvassing in Germany, our study is also qualitatively oriented and uses six in-depth expert interviews with nine campaign coordinators of the most promising parties running for state parliament of Rhineland-Palatinate in 2016 (one strategist from the CDU; two from the SPD; two from The Greens; one from The Left; two from the FDP; and one from the AfD). The use of this methodology can be justified with reference to the objective pursued: by those means, election campaign experts should allow for systematic and comprehensive acquisition of information concerning data-driven canvassing in the German social, legal, media and electoral context.

The interviews were conducted during the campaign period between December 2015 and February 2016. For this purpose, we developed a semi-standardised guideline based on the comprehensive literature about data-driven campaigning presented above. It contains thematic blocks about the campaign organisation in general, the communication strategies adopted by the parties in order to contact, mobilise or persuade voters, data use and canvassing. Following the suggestions of Bogner et al. (2002) our guideline provided an open discussion where the interviewed experts could add further information when necessary. All interviews were recorded, transcribed and coded after Mayring´s (2010) guidelines for qualitative content analysis (reduction through paraphrase and generalisation) and with the use of the qualitative data analysis software MAXQDA. Therefore, our coding scheme was derived by combining a deductive and inductive approach, yielding the most important contents and themes for our analysis (see Appendix 1).

6. Empirical findings

To answer the leading questions about the importance, use and restrictions for micro-targeting and canvassing in Germany, we will present our findings along the framework of constraints on the macro, meso and micro levels.

1) Macro level: systemic restrictions

1a) Electoral and party system; 1b) Political culture

All German campaign managers agreed that data-driven campaigning has arisen during an era of partisan de–alignment and declining vote share. Therefore, parties “must find newer methods to engage with the electorate in a more complex, multiparty system” (SPD). Contrary to the US or the UK electoral system, which incentivise parties to use their resources to target voters in battleground states, the German electoral system is characterised by proportional representation, long-established party strongholds and rare competitive constituencies. Therefore, German campaigns need other data sources to decide which constituencies should be targeted by canvassing efforts. The Green campaign managers mention an “urban-rural-gap” which gives their Party advantages in cities, particularly those with universities. The other experts say that party associations also reflect economic divides (The Left), and that some parties tend to fare better in one region or the other because of “historic ties” (CDU) to the area or “persistent socio-economic differences between them” (SPD). Another finding is that new and smaller parties do not have good chances because the political landscape is more oriented towards the bigger parties due to stable electorate milieus. The right-wing party manager additionally mentions social marginalisation of the party because of its controversial standpoints (AfD).

In the German parliamentary system, campaigns must also decide if the canvassing campaign should get the first or second vote. This affects if a candidate has to join the canvassers at the doorsteps (The Left, CDU) and how to talk to the voters at the door. Taken together, the German political culture, electoral and party system makes the organisation of data-driven canvassing campaigns especially challenging. Because German parties lack other data sources and more detailed information, they use geographical targeting and focus on “constituencies that had been party strongholds in the past, but saw lower voter turnout in former elections” (CDU, SPD, The Greens).

1c) Legal concerns

All campaign managers show a great caution with regards to data protection standards in Germany. Exemplarily, the SPD manager says “Data is a highly sensitive topic. There are literally millions of data providers on the market and it was clear to us that only somebody serious is eligible for us.” The interviewee also named a main reason for that, namely that the party must be trustworthy and transparent because there is great distrust among the public about persuasive political campaigning techniques. This also applies to targeting technologies which differ noticeably from US techniques which, according to the Left campaign manager, “could never migrate to Europe because the sensitivity of data concerning political affiliation is rooted in a German political culture”. Therefore, “the smallest target units are streets and constituencies not households” (SPD). Although, all parties attribute a high importance of data for their campaigns, only some have experimented with different data sources to find the ideal way for the strategic organisation (SPD, CDU). Because of that, the SPD manager stresses the importance of focusing on a few constituencies with high potentials and low wastage of advertising. To ‘circumvent’ the strict data laws, German parties started to collect data with apps (CDU) or clipboards (SPD) at the voters’ doorsteps, which are computerised in a database. In terms of law-abidance, this also needs explicit declarations of consent which is not possible on a large scale. Additionally, data must not be stored over longer periods. This consequently means that German data laws impedes micro-targeting based on the US model.

Nonetheless, high standards in European data protection laws do not ultimately prevent the implementation of micro-targeting. Bennett (2016) reminds that the term “political opinions” in the European General Data Protection Regulation is not clearly defined. Thus, these data could be allowed to be processed if political players can prove that the targeted person is in regular contact with the party. Besides, Bennett adds further external forces putting pressure on European legislation: a shrinking core voter base which leads to more market-oriented campaign strategies. These define voters as customers who need to be contacted with the use of targeting methods. Furthermore, technological (read marketing companies) and social forces could influence the future design of data protection laws in Europe.

2) Meso level: organisational conditions

2a) Party culture

Drawing on the interviews, party cultures within political parties played a significant role in shaping their use of data and canvassing efforts. Members of party divisions with a long history tend to be very self-confident and often refuse to follow the headquarters’ rules for canvassing (SPD). So, the campaign managers are only able to declare focal points for canvassing but “cannot instruct the local chairman to do canvassing in a particular area.” (SPD). “Nevertheless I don’t want such a structure like in the US, as the German thing with traditional local committees is fantastic.” (SPD). Contrary to US campaigns, where campaign assemblages (Nielsen, 2012) get together before elections, German parties are working throughout the year. This means local party committees are additionally planning and carrying out campaign measures (SPD, FDP, The Left) which is useful for the campaign headquarters as they are “dependent on the local structures” (SPD) to save resources and to carry out the exhaustive campaign. According to the interviews, German party culture is characterised by top-down as well as bottom-up communication which can be used to produce a better understanding of new campaign strategies (CDU, SPD, FDP, The Left). Especially the Green Party tries to empower its members to become a part of the opinion-forming process and the planning of the campaign. Furthermore, the Green’s party ideology limits the use of data in campaigns because it’s a “question of faith” whether the party should purchase information on voters. Still, they commission the Deutsche Post for direct marketing efforts.

2b) Personnel and financial resources

Generally, the campaign experts stated that the campaign funding is based on their local savings without any financial help of the national party. The logic behind this system assumes an institutional separation between the two. Unlike PR methods and the use of social media, data canvassing is a highly resource-intensive instrument. As we have mentioned above, German parties have limited budgets at their disposal and have to “make a smart mixture” (SPD) of possible communication channels limited by a tense budget situation. Although there are attempts to send out paid staff, financial restrictions are a key constraint: “We cannot afford to pay everybody at any place for canvassing.” (SPD) Missing money also affects feedback from voluntary canvassers, as they do not have the obligation to report to campaign headquarters (SPD). In addition to that, investments to external consultants (FDP, SPD), data vendors (FDP, The Greens) and the purchase of data (SPD) are further positions in the budget. Particularly the data service by German Post Direct is “really expensive” (FDP).

Especially smaller parties struggle with financial constraints, which could consolidate power in the larger and better financed parties and make it more difficult for smaller parties to be nationally competitive. Therefore, concerning the organisation of campaigns, managers of the AfD and The Left “listen to their gut feeling”.

2c) Technology and infrastructure

Contrary to US campaigns, digital tools are not fully established in German federal and state elections. However, the interviews with the CDU, SPD and The Greens revealed attempts to professionalise campaign infrastructures in Rhineland-Palatinate: the SPD tried to mobilise volunteers by using a digital platform where party members can connect and stay up-to-date. This site can also be used to coordinate canvassing operations. However, SPD’s campaign managers did not see great benefits because the platform lacks usability, more detailed information about constituencies or streets and possibilities to interact. The SPD managers even put more emphasis on analogue channels of internal communication: “We are using stuff which suits our members best... and that is paper. I think we had most information in a final document, brought it to the trainings and distributed it in our offices” (SPD). This finding is also documented in the Green Party interview. The Greens recommend the use of notes and maps to their volunteers: “The people should use lists to document where they did their visit. Maps can be used to coordinate ‘Team A’ and ‘Team B’.” (The Greens) The CDU could have used an app for the organisation and data gathering during canvassing activities but personal opinions of leading personnel and the main campaign strategy opposed this idea. Overall, ‘offline technologies’ like clipboards (SPD) and excessive phone calls (The Left) prevailed in the campaigns’ infrastructure.

The use of databases to store voter information varies from party to party. Overall, legal foundations were often mentioned in the interviews and an equivalent of the US vote files are completely missing due to legal restrictions. Therefore, German parties rely on data from in-house distribution lists (CDU), past election results in single constituencies (CDU, SPD) and freely available data on turnout and demographic factors (SPD). In practice, the SPD, CDU and Greens build a so-called ‘mobilisation index’ revealing constituencies with a high probability of ringing doorbells of former Social Democrat voters, meaning former partisans who abstained from voting in latest elections. Although these data can be helpful to identify auspicious areas, because of strict data handling rules the campaigns do not have access to household information. For this reason, the SPD bought data from an external data vendor which was not really satisfying: “The company only has data for districts in a particular city. When we asked them how it is in other districts, they calculated something with their data. [...] In the end there are just some minor differences which do not have a statistical correlation to the election results - and that’s it.” (SPD) The manager of the Liberals added that external data by the German Post is not that significant and therefore not reliable for their campaign. Nevertheless, the German Post service called ‘Post Direct’ is used to exclude partisans of other parties for direct mail: “The German Post knows exactly if membership magazines of the AfD or the Green Party are distributed to households. Then you know that a FPD invitation is not suitable in these households.” (FDP) In addition, the Green Party uses the service of the German Post but only for direct mailings. Past election outcomes cannot be seen as reliable data for three further reasons: first, there are different lists at the municipalities and the state election commissioner. Additionally, some local authorities outsource their system of registration. Second, political gerrymandering changes the data of past elections which can then not be compared with older data. Third, data from municipalities is processed differently than the party’s dataset which complicates analysis (SPD). Contrary to the US, German parties do not have full access to useful data sets - a fact that can mislead campaign managers and is well-known in political communication literature: “[...] the quality of the outcome is no better than the quality of the input: Garbage in - garbage out.” (Burton & Shea, 2010, p. 91)

3) Micro level: individual constraints

3a) Politicians & campaign managers

One crucial limitation on the micro level are individuals. If a frontrunner feels uncomfortable with the idea of walking from door-to-door or a campaign manager focuses on other channels, a canvassing campaign is not a good idea (CDU). Furthermore, in some canvassing campaigns frontrunners are expected to be part of the effort, provided there are no other appointments (CDU, The Left). The experts refer to their advantage in getting in contact with voters. However, frontrunners take part in major events as well as media meetings and other statewide campaign activities, which limits the time budget for canvassing. Hence, the interviews revealed that local politicians are more important for canvassing. Well-known candidates in their home regions and towns play a significant role as opinion leaders (FDP, The Left, CDU). “There are some single cities or towns where we do have some very good outcomes. This is because of the persons on the ground. There are always persons who are very prominent and who are active for many years.” (The Greens) The CDU expert also stated that if the general secretary of the local party walks from door-to-door, people would automatically come out and talk to her.

Another result of our interviews are the various levels of knowledge about canvassing or data technologies, like predictive modelling. Two of our interviewees even reacted with the question “What is predictive modelling?” (The Greens, The Left) and after explaining, they responded: “I barely passed my statistic course. [...] You do not need such skills when you are working in a party. No matter what anyone tells you, it is not true.” Therefore, we could find a lack of knowledge about new technological and statistical instruments.

3b) Volunteers & party members

Members are described as “the centrepiece” (SPD), “the backbone of the campaign” (The Green Party) and as “very engaged” (AfD). However, party members and sympathisers are not professional and paid agents using elaborate techniques in their campaign activities, although there are attempts to integrate low-paid persons (SPD). Therefore, the major challenge is to mobilise and manage party activists to engage in the campaign. People must be convinced(The Greens, SPD) of the campaign before they get involved: “I think the great age of party soldiers is over except for a few cases. You have to grip people and they have to be keen on campaigning.” (The Left) The first crucial limitation of voluntary action is time: “The biggest problem is the reconciliation of family life, job and political engagement.” (SPD) There is also a risk that an excessive use of volunteers in campaigns could lead to a weariness for upcoming elections (SPD). The second limitation are the skills of canvassers (The Left, CDU): “You need openness, sympathy, eloquence and warmth if you want to talk to people at their doorsteps. These skills have to be transmitted.” (CDU) Furthermore, the SPD expert said that it is hard to run a modern data-driven canvassing campaign when your canvassers are mostly retirees (SPD), referring to the average age of the German party members. Therefore, parties (CDU, SPD, The Greens) developed systems to train their activists in how canvassing works. However, all campaign managers gave examples that there are limits to what could be achieved, as political parties are still voluntary organisations.

3c) Voters

Even if a door-to-door campaign is planned strategically and shouldered by many volunteers, success is only obtained if targeted voters are open to visits. One limitation concerns the timing of the visit. While most of the party experts recommended visits from 4:30pm on during the week, the Left Party campaign manager is against Sunday visits. Also, special working hours of shift workers and especially commuters (SPD) or religious habits (CDU) have to be kept in mind. The SPD manager emphasised the meaning of being honest concerning the reason of the visit and of showing the canvasser’s motivation to walk for the party (SPD). Another issue are negative reactions at the door (AfD, The Left, SPD). Particularly, the AfD manager named social desirability as a huge problem because the party is widely seen as a right-wing party: “People are afraid to profess themselves to the AfD. [...] Many want to keep distance to the party in public.” (AfD) Also, the Left Party expert mentioned the problem of getting negative reactions, especially in rural areas. Age and sex are further dimensions with impact for canvassing campaigns. Young voters in university cities must be addressed differently than older people from rural regions (The Greens) and the AfD interview revealed difficulties with female voters who are more likely to reject the party. Again, parties (CDU, SPD, The Greens) developed special trainings to teach canvassers these basic sources of failure and how to react. But even if canvassers are well prepared and contacts are seen as successful, there is no guarantee for a desired voting behaviour: “Voters are still humans and not computers which react to a formula you type in.” (CDU) In summary: “It is not easy to mobilise people.” (SPD)

7. Discussion and conclusion

We showed that German local campaigns make different attempts at targeting voters with direct contact at their doorsteps based on data analysis (see Table 3).

Although all German parties show wide interest in the use of data and canvassing for voter targeting, only the two main parties (Volksparteien - CDU, SPD) see them as important to their actual campaigns. As a result, they invest time, money, know-how and staff. However, our findings show that German 'data-driven' canvassing cannot be compared with the highly sophisticated US campaigns which use reliable data, experimental findings and data modeling to identify individual voter targets like swing voters or infrequently voting partisans – “many of whom pay little attention to news and who frequently are not interested in (or even decidedly disenchanted with) electoral politics” (Nielsen, 2012, p. 18). In contrast, German parties use geographical targeting based on the analysis of precinct-level results from past elections to identify auspicious electoral constituencies for their canvassing efforts. According to Nickerson and Rogers (2014, p. 51f.) these techniques were used “as recently as a decade or two ago […] and appear extremely rudimentary by current standards”.

Based on findings from in-depth expert interviews with campaign coordinators of parties running in the state parliament of Rhineland-Palatinate in 2016 we found several restrictions for the use of data-driven canvassing on the external, internal-organisational and individual dimensions. On the macro level, the use of individual-centered campaigning techniques like micro-targeting and canvassing are rooted in the electoral process, the political culture, the fragmentation of political parties, the rules for campaign financing and the legal foundations - especially the information privacy laws. On the meso level the use of these instruments can be restricted by the organisational culture of a party, the technological infrastructure, as well as personnel and financial resources. The capabilities of smaller parties lag far behind those of the Volksparteien, so that data strategies are distributed very unevenly among German political parties. On the micro level individual selection processes, knowledge, autonomy and time of local candidates, campaign managers, canvassers and the attitudes of voters decide over the implementation of data-driven canvassing into a political campaign.

Although this study is limited because it focuses on a single case state legislature election, it provides important insights for further discussion and generating hypotheses on a comparative and normative level. Our results show that certain factors on the macro, meso and micro levels hamper the strategic and sophisticated use of data-driven canvassing. Moreover, the relationship between data-driven targeting and electoral, legal or individual context factors seem important for its usage, raising the follow-up question: which contexts are most fruitful for these strategies? Furthermore, parties are eagerly interested in the implementation of individual-based campaigning techniques without thinking about implications for personal privacy, civil liberties and democratic values. Concerning personal privacy, legal regulations can be re-considered and adjusted by governments if personal data is not protected against data collectors. More fundamental changes could arise in the democratic discourse: since data-driven campaigning aims at “useful” voters, less “important” voters are excluded from the campaigns (Hillygus & Shields, 2008). Consequently, political topics of these voter groups could be neglected in the political discourse. In a worst case scenario this could lead to a defect in democratic discourse when specific minority issues are not even addressed by politicians for strategic reasons. Additionally, targeted voters are confronted with selected information on certain topics (Nielsen, 2012) which can be assessed as manipulative communication. Taking research on instant influences into advance, repeated and partisan information stimuli could have an impact on the attitude and behaviour of voters (Papakyriakopoulos, Shahrezaye, Thieltges, Medina Serrano & Hegelich, 2017).

Bennett’s (2016) claim that developments in other nations, particularly the US, combined with external forces from the technology sector and the economy could lead to an increasing demand of data-driven methods in Europe has to be taken into account. Consequently, political discourse will increasingly be characterised by the question of whether or not, and if so, in which form micro-targeting should be allowed in Germany and other European countries.

Appendix 1: Coding frame

1.    General campaign organisation
1.1.    Involved political entities (importance, implementation, functions)
1.1.1.     Voter
1.1.2.     Campaign headquarters
1.1.3.     State executive committee
1.1.4.     Country office
1.1.5.     State parliamentary group
1.1.6.     Party base
1.1.7.     External consultants
1.1.8.     Agencies
1.1.9.     Federal executive board
1.1.10.    Federal parliamentary group
1.1.11.    Hierarchy and relationships
2.    External influencing factors
2.1.1.     Structural
2.1.2.     Situational
2.1.3.     Intra-party
3.    Programmatic dimension
3.1.    Main campaign issues
3.2.    Other issues
3.3.    Issue evaluation
4.    Campaign direction
4.1.    Issue centered
4.2.    Person centered
4.3.    Target group centered
5.    Instrumental dimension
5.1.    Negative campaigning
5.2.    Focus of campaign communication (indirect/direct)
5.3.    Campaign budget
5.4.    Importance of campaign instruments (importance, usage)
5.4.1.     Social media
5.4.2.     Door-to-door canvassing
5.4.3.     Direct mail
5.4.4.     Leaflets and brochures
5.4.5.     Billboards
5.4.6.     Party information booth
5.4.7.     Email
5.4.8.     Print ads
5.4.9.     TV ads
5.4.10.    Radio ads
5.4.11.    Homepages
5.4.12.    Telemarketing
5.5.    Importance of survey methods (relevance, usage)
5.5.1.     Geomarketing/Potential analysis
5.5.2.     Data analysis
5.5.3.     Opinion surveys
5.5.4.     Databases
5.5.5.     Predictive modelling
5.5.6.     Opposition research
6.    Door-to-door canvassing
6.1.    Preparation
6.1.1.     Period
6.1.2.     Canvassers/Volunteers
6.1.3.     Trainings
6.2.    Organisation
6.2.1.     Carrying out door visits
6.2.1.1.    Communication guidelines
6.2.1.2.    Reactions
6.2.1.3.    Length
6.2.1.4.    Follow up communication
6.2.1.5.    Tools (apps, clipboard)
6.2.2.     Persons responsible for carrying out visits
6.2.3.     Data analysis for organisation
6.2.3.1.    Used data
6.2.3.2.    Data methods
6.2.3.3.    Gut feeling
6.2.3.4.    Walk sheets
6.3.    Aims
6.3.1.     Party image
6.3.2.     Currying sympathy
6.3.3.     Information conveying
6.3.4.     Persuasion
6.3.5.     Mobilisation
6.3.6.     Data gathering
6.3.7.     Turnout
6.4.    Target groups and communication strategies
6.4.1.     Partisans
6.4.2.     Sympathisers
6.4.3.     Undecided
6.4.4.     Other party supporters
6.5.    Success factors
6.5.1.     Assessment of success chances
6.5.2.     Strengths
6.5.3.     Weaknesses
6.6.    Evaluation
6.6.1.    Type of evaluation
6.6.2.     Point of time
6.6.3.     Qualitative evaluation
6.6.4.     Quantitative evaluation