Political parties and social media campaigning

Abstract

Do new media level the playing field during election campaigns (‘equalization’) or do they mirror existing inequalities between parties (normalization)? Empirical studies come to contradictory findings. Part of the answer is in the timing: first social media level the playing field, afterwards bigger parties see the benefit and invest in it. Yet, this raises a new question: given that social media are cheap and easy to use, how can investing in them tip the balance? Based on a critical assessment of the literature and in-depth interviews, we advance a new theoretical framework to address both contradictions: the motivation-resource-based diffusion model. We link this model to the broader party and campaigning literature and formulate expectations, in terms of party size and ideology, about which parties use social media professionally. Afterwards, we conduct a crisp-set qualitative comparative analysis (QCA) of the Dutch parties (2010 and 2012 elections) to assess these expectations. We find that populism, postmaterialism, and party size matter but in different ways in the different phases of diffusion.

Introduction

Do social media level the playing field for political parties (‘equalization’) or do they reproduce existing offline differences between them (‘normalization’)? Social media have provided additional campaign instruments to win votes. Social media activity can spillover, increase attention of traditional mass media, and influence journalists´ evaluation of the current public mood, e.g. in the aftermath of television debates (Kreiss 2016). Additionally, social media are often used to mobilize volunteers and increase the amount of donations and coordinate on-the-ground campaign efforts, as the 2008 and 2012 Obama campaigns nicely illustrate (Chadwick 2013).¹ As such, the question whether or not new technologies, like social media, balance the campaigning opportunities that parties have is a pivotal one for understanding their impact on the overall political power relations.² It is thus not surprising that cyber campaigning has attracted a lot of scholarly attention (Margolis et al. 1999; Small 2008; Vergeer and Hermans 2013; Larsson and Moe 2014; Dolezal 2015; Gibson and McAllister 2015). The majority of these studies have indeed been driven by the question that opened this introduction; ‘[e]qualization versus normalization is a key debate in the cyber-campaigning literature’ (Small 2008, p. 52).

Yet despite this academic attention, several gaps remain, particularly regarding the way parties use social media. Early campaigning research found that the Internet (Web 1.0) mostly benefited major parties, who had the resources to create professional websites with lots of bells and whistles (cf. Vaccari 2008), but social media such as Twitter and Facebook (Web 2.0) may induce equalization rather than normalization, because they are said to be cheap and easy to use (cf. Dolezal 2015, p. 103). Empirically, some studies find equalization (e.g. Gibson and McAllister 2015, p. 529); yet others find normalization (e.g. Vergeer and Hermans 2013, p. 399). This divergence might partly be related to timing, as suggested by Gibson and McAllister (2015, p. 530), who find equalization is followed by normalization. Smaller parties benefit from new media at first, but later on major parties ‘[a]re prepared to invest in it’ and catch up (Gibson and McAllister 2015, p. 529).

Gibson and McAllister lay bare an intriguing theoretical paradox: if social media are really cheap and easy to use, how then can investing money in them allow major parties to catch up? This paradox flags the broader issue that we hardly know why and how social media would induce normalization or equalization. The ‘normalization versus equalization’ debate seems to rest on premises stemming from earlier Web 1.0 research, which overlooks the important role of the ‘attributes of the innovation’ (Gulati and Williams 2013, p. 579). If social media are indeed different from older online tools (i.e. they are cheaper and easier to use), our theories need to be updated.

We engage with the aforementioned empirical and theoretical puzzles by formulating a new theoretical framework, the motivation-resource-based diffusion model, which synthesizes elements from the existing literature, teases out the mechanisms, and explicitly incorporates a temporal dimension. We argue that to understand the parties’ use of social media we need to take into account that parties behave differently depending on how widely used social media are in society. Additionally, the model stresses the need to understand the impact of parties’ size, ideology, and other characteristics on the use of social media in terms of the resources and motivations of those parties. By formulating and testing this framework, we address and try to answer the following research questions: (1) What role do resources and motivation play in (withholding parties from) using social media professionally? (2) When do which parties make professional use of social media? And derived from those two questions (3) Does the introduction of social media lead to inter-party equalization, normalization, or both?

Our empirical focus is on the Facebook and Twitter use of Dutch parliamentary parties (2010–2012). Contrary to the most recent election (2017), there are a lot of data available for these two elections, which makes it possible to study them more elaborately. Given the rapid developments in the field of online services, such a relatively short time frame is fairly typical in the field. For instance, the two most prominent studies, Gibson and McAllister (2015) and Larsson and Moe (2014), use an interval of 3 and 2 years, respectively. In fact, most studies only analyse a single election.³ Regarding the platforms under study, we focus on these Twitter and Facebook given that they were the most prominent social media platforms at that time. However, the theoretical framework presented in this study is formulated in general terms. The underlying mechanisms are also applicable to more recent social media platforms (e.g. Instagram) and new uses of existing technologies that emerged since 2012 (e.g. personalization via micro-targeting). To illustrate this, we apply our framework to micro-targeting in the concluding section of this paper. The resulting expectations can be tested on, for instance, the 2017 elections once sufficient in-depth data are available. Regarding our case, the Netherlands is a particularly well-suited case to address our research questions because it has a diverse range of parties. This allows us to study and compare multiple parties of different size and ideology.

Our multi-method approach combines in-depth qualitative interviews with a systematic qualitative comparative analysis (QCA). Together these two methods provide in-depth insight into (1) why and which resources and motivations might be at work and (2) the importance of the different factors (i.e. party size and ideology) from which these resources and motivations originate. The multi-method approach thus allows us to address the theoretical and empirical puzzle outlined above in tandem. First, in the theoretical part we formulate expectations about which party characteristics matter regarding the professional use of social media. These expectations are generated from both the existing literature and 11 semi-structured interviews with party officials about why and how resources and motivations play a role in parties’ use of social media. Second, we conduct a QCA to assess the existence of more general patterns in terms of parties’ use of social media along the lines of party size and ideology; factors stressed in the equalization–normalization literature. This analysis thus simultaneously tests the expectations derived from our model and from the broader party and campaigning literature.

Theoretical background

The equalization–normalization debate

The impact of new technologies on the power balance between parties during election campaigns has been puzzling scholars for almost 20 years (see Gibson and McAllister 2015).⁴ Some scholars find equalization (e.g. Gibson and McAllister 2011), others normalization (e.g. Vergeer and Hermans 2013), and recent studies argue that equalization is followed by normalization (Gibson and McAllister 2015; Larsson and Moe 2014). Below we detail each of the three strands in the literature. We first outline their explanations for the phenomenon, followed by a reflection on these explanations. This reflection will be the stepping stone towards our framework.

Equalization or normalization

Studies adopting the equalization perspective generally ‘revive’ the arguments in the Web 1.0 literature (Vergeer et al. 2013, p. 482) and stress that social media are user-friendly, decentralized, and interactive. They argue that new media are (1) cheaper; (2) require less expertise; and (3) allow smaller parties to bypass traditional media (see Vergeer and Hermans 2013; Gibson and McAllister 2015). Furthermore, it has been posited that (4) the interactive nature of new media benefits (small) postmaterialist parties (Vergeer and Hermans 2013).

This notion of equalization is contrasted with that of normalization, and regarding normalization the Web 1.0 and Web 2.0 literatures provide three similar explanations (e.g. Gibson and McAllister 2015, p. 543; Larsson and Moe 2014, p. 11; Small 2008, p. 53): online technologies simply replicate old power inequalities because (1) larger parties have strategic departments; (2) such parties also have professional politicians; and (3) ‘[m]ajor parties have the resources and motivation’ (Small 2008, p. 52, see also Gibson and McAllister 2015, p. 543; Larsson and Moe 2014, p. 11).

At the core of both perspectives are the notions of resources and motivation. The equalization perspective stresses that social media require fewer resources, whereas the normalization perspective stresses the resources larger parties can tap from. Similarly, the equalization logic refers to smaller parties having a stronger motivation to adopt social media, but the same claim is made (although hardly substantiated) by the normalization studies, and the equalization argument does not explain why larger parties lack this motivation. These seemingly contradictions are partly due to the relatively thin mechanisms presented in this debate. Further specification and conditioning is thus needed.

Equalization and normalization

In a recent study, Gibson and McAllister (2015, p. 533) partly break down this contradiction between equalization and normalization. Their study provides an important first step in further specifying how and when dominant or smaller parties use social media. The authors suggest that cyber campaign innovations follow a two-step diffusion process: equalization at first, followed by normalization. They argue that especially smaller (in their case postmaterialist) parties have incentives to experiment with innovative tools, which gives them an initial advantage if these turn out to be successful. When that happens, ‘larger parties [see] the added value of online campaigning and [are] prepared to invest in it’ resulting in a normalization of the inter-party relations later on (Gibson and McAllister 2015, p. 529).

This insight is an important step forward, because the logic presented by Gibson and McAllister has the potential to account for inconsistencies in earlier findings and has proved successful in other work (Larsson and Moe 2014). At the same time, the core argument on why and how parties use social media has not been specified in more detail, which is crucial for understanding which parties adopt social media and when they do so.⁵ The role of resources and motivations remains elusive, which means that new questions are triggered and others remain unanswered: Are only postmaterialist smaller parties likely to be ahead early on and, if so, why is this the case? Are some parties more likely to lag behind? Why are major parties able to catch up and why are they hesitant early on? In sum, resources and motivation seem important, but little is known about the mechanisms by which these two building blocks result in two-step diffusion.

Formulating a ‘motivation-resource-based diffusion model’

In this study, we build on Gibson and McAllister’s diffusion logic and set out to formulate more clearly the roles motivation and resources play in fostering (or withholding) parties from using social media in the two diffusion phases. From that starting point, we consider the implications regarding which parties can be expected to use social media professionally. Specifically, the model offers further substantiation of why factors considered important in the literature—party size and ideology (Small 2008; Schweitzer 2011)⁶—matter and also includes new expectations that contradict the existing literature.

To fill the theoretical gaps, we started by conducting and analysing 11 semi-structured interviews with party officials (see Appendix 1) who we explicitly asked (indirect) questions about the parties’ motivations and resources. We thus use these interviews to obtain a saturated overview of which resources and types of motivations might matter.⁷ At this stage, we are not interested in testing how many or which parties mentioned certain types of resources or motivations, but in finding out which mechanisms might be at work and theorizing what the implications of these are for the impact of factors identified in the literature. Unless stated otherwise, the substantiation of the mechanisms below is based on the interviews. The numbers in the text refer to the list of interviewees in Appendix 1.

Resources

While the literature tends to stress that setting up, posting, and maintaining a profile on Facebook or Twitter are all cheap, our interviews suggest several ways in which money matters. They also highlight the role of non-financial resources.

Most clearly, our interviews indicate that money can have a direct impact. Parties use it to boost their social media team’s professionalism and it enhances their capacity to engage and mobilize. Specifically, money was said to buy access to more advanced software needed to experiment with what kind of social media use yielded more engagement and reached more users.⁸ Money also buys a party better quality posts that include high-quality photographs and infographics.⁹ Furthermore, money can have an indirect impact. The party officials stressed the importance of personnel capacity and thereby time.¹⁰ Indeed, social media are fast media and may trigger quick snowball effects (i.e. ‘going viral’; Klinger and Svensson 2014). Extra time was said to allow one to ‘monitor the other parties’, ‘train candidates’, ‘be available for questions from politicians’ and ‘constantly experiment with what works best’ (Interview #4).¹¹ And if all these things fail, money allows a party to fly in high-level social media consultants to strengthen its ranks.

While the above suggests that financial resources are more important than previously thought, the interviews also reveal that resources are not just about money. There are ways parties can compensate for their lack of financial resources. One social media manager highlighted the importance of tech-savvy volunteers and politicians. Late in the election campaigns, this paid off: ‘the last few months of the campaigns we had six interns [students]. At that time we had the personnel capacity’ (Interview #5). Another party, which did not have this personnel capacity, compensated by having the active support of their ‘kindred spirits’—the online team (7.0 fte) of a large broadcasting service whose audience overlaps with the party’s electorate. They provided temporary access to advanced software and shared insights in how to use social media well from which the social media manager ‘learned an awful lot’ (Interview #3).

Motivation

While resources might matter, by itself this tells us very little about why parties have the ‘motivation’ to allocate these resources to social media (Small 2008, p. 52). Our interviews reveal two important things. First, they suggest that one should differentiate between ‘extrinsic’ and ‘intrinsic’ motivation. Second, they indicate that sometimes parties have motivations not to invest in social media.

Our interviews suggested two main extrinsic motivations: to reach out directly to voters and to get the attention of journalists. The reach-out-to-voters motivation was mentioned across the interviews in terms of spreading information, engaging with voters, and mobilizing them. As one social media manager said: ‘We go where our voters are. Once a network starts to take off, we’ll go there. But we will not invest time and energy in it when our voters are not there’ (Interview #8).¹² Social media were, however, not just seen as a means to connect directly to voters. Since the early days, especially Twitter is used by Dutch journalists to look for quotes as the platform’s one-liner nature and lenient copyright procedures make it well suited for them. The parties and politicians saw this as a new opportunity early on and mentioned it explicitly: ‘You can directly contact them [journalists] on Twitter and get into a debate with them’ (Interview #5). Also they were motivated to use Twitter particularly to monitor journalists, as a ‘weather forecast’ of the upcoming news to ‘have your defense ready’ (Interview #4; see also Kreiss 2016).

The intrinsic motivation argument featuring in the equalization literature (Vergeer and Hermans 2013) also surfaced during our interviews. Several parties mentioned that they considered themselves a party ‘that should be approachable’ (Interview #11), and this was said to serve as an incentive to use the interactive features of social media. More generally, it was mentioned that social media can be a ‘celebration of being modern’ (Interview #5). At the same time, the interviews brought up another form of intrinsic motivation, one not to use social media. Until now, such motivations have largely been overlooked in the existing literature, but clearly they should be included in a more general framework. The main ‘negative’ motivation is linked to the degree of party centralization and the autonomy of backbencher politicians more in particular. Some campaign managers and top politicians stressed a fear or dislike of many MPs using social media as it would allow backbencher politicians to build their own power base. It could thereby stimulate dissident behaviour or lead to leaks of information that was not yet approved by the party leadership: ‘We do not want (…) individual MPs all airing their own individual opinions and start criticizing each other on social media’ (Interview #2).

The motivation-resource-based diffusion model

The final step in building our model is to combine the two-stage ‘diffusion’ logic of Gibson and McAllister with our refined understanding of resources and motivation as discussed above. Focusing on the first phase of diffusion, the extrinsic motivations suggest that all parties can be expected to use social media to connect with journalists, but the parties whose voter are early adopters have an additional motivation to enter social media. Moreover, these parties can also be expected to be the ones to be intrinsically motivated. This motivation does not seem to originate in parties being small (as argued in the equalization perspective), but rather in so-called modern parties’ identity. In this first phase of the diffusion of social media use, such media might still be cheap, making resources less important in that phase.

In the second phase, social media become more mainstream among voters and their use professionalizes further. As high-quality use requires resources, these become more important in this second phase of diffusion. Resources, however, not only include money but also refer to knowledge and professional volunteers who bring in expertise. For these resources to be used on social media, there must be some motivation to do so. In this phase, more parties can be expected have extrinsic motivations, as only the older voters of the most conservative parties will be absent of social media in this phase. Hence, many parties will want to catch up, and the parties with the resources can do so. As social media are not that new anymore, the intrinsic motivation to appear modern will make less of a difference in this phase. However, the fear of social media undermining the party leadership may still be a strong motivation not to use them professionally, especially in the case of centralized parties.

Summarizing the gist of what we call the ‘motivation-resource-based diffusion model’¹³ leads to the following mechanisms:

• 1. During both stages, parties have the extrinsic motivation to use social media to reach conventional media.

• 2. In the first stage of diffusion, from here on: ‘early-adoption phase’, the intrinsic motivation grounded in parties’ identity is crucial in determining whether a party adopts social media at a professional level.

• 3. In the second stage of diffusion, from here on: ‘widespread-diffusion phase’, the extrinsic motivation pushes mainstream parties to use social media professionally to connect with voters and journalists.

• 4. In the widespread-diffusion phase, resources matter considerably more: parties that have financial or ‘free’ resources use social media professionally.

• 5. In the early-adoption and widespread-diffusion phase, the intrinsic motivation to keep control over the party will prevent centralized parties to professionally use social media.

Formulating expectations

Now we have articulated the more general causal mechanisms in terms of motivation and resources; it is possible to connect our framework to explanatory party characteristics found in the existing literature. For each, we will formulate expectations that we will test empirically.

Modern party family: postmaterialist parties

In the early-adoption phase, parties that want to showcase their modern identity will have the intrinsic motivation to use social media. Of all party families, this focus on technological innovation and progress fits the postmaterialist parties most (cf. Gibson and McAllister 2011, 2015). Moreover, these parties’ ideology includes valuing approachability (e.g. Inglehart and Abramson 1999). Therefore, the interactivity associated with social media comes naturally to them. In addition, they also have the extrinsic motivation to take to social media as their electorate is often younger and among the first to use new media. Following the same logic, they have in-house expertise of how to use social media. This ‘free resource’—tech-savvy volunteers and politicians—also allows them to stay ahead in the widespread-diffusion phase even if they do not have substantial financial resources. Indeed, postmaterialist parties are virtually always smaller and cash-strapped.

Expectation 1

Postmaterialist parties are ahead in the early-adoption phase and stay leading in the widespread-diffusion phase.

Resource-rich parties: party size

In line with Gibson and McAllister (2015), we do not expect bigger parties to be early adopters as (1) their voters are not on social media in the early-adoption phase and (2) they have no intrinsic motivation to do so (unless they are postmaterialist parties—see above). We follow Gibson and McAllister in expecting that bigger parties are less motivated to experiment because they have resources to catch up later on and they have fewer intrinsically motivated tech-savvy people in their party who are early adopters themselves. Hence, we do not expect them to invest their resources early on. However, in the widespread-diffusion phase, when their voters are on social media, these parties do have the extrinsic motivation to use social media. Indeed, at that moment their (potential) electorate can be reached directly via these media. As a result, they can be expected to invest their financial resources in social media.

Expectation 2

Non-postmaterialist bigger parties lag behind postmaterialist parties in the early-adoption phase, but catch up in the widespread-diffusion phase.

Centralized parties: populist parties

Some have highlighted that social media are unmediated and allow for (populist) parties demonstrating that the politician and the party are of ‘the people’ and not the political establishment (Tromble, Forthcoming, p. 9). This may especially hold for the party leader or other top candidates. Studies examining online forums have added that because people can be online anonymously, which allows the electorate of populist parties to safely voice opinions that diverge from the mainstream (Gibson and McAllister 2015, p. 531). Although the empirical evidence is limited, most of the existing literature on populist parties therefore expects that such parties benefit from social media. Yet some have found that if we go beyond the populist leader, such parties seem to lag behind (Dolezal 2015, p. 115). How to explain this?

To begin with, if anonymity is crucial for the populist electorate, it is unlikely they will connect with populist parties on not-so-anonymous social media. Overall, the extrinsic motivation to take to social media early is at least questionable. Additionally, populist parties are highly centralized (Mudde 2007): they have tightly controlled party organizations. This makes social media ‘dangerous’. As such, media allow politicians to build an unmediated relationship with the party base and potentially reach out to a large group of followers, and they actually pose a threat to centralized parties. Social media can empower backbenchers at the expense of the party leader. Moreover, tweets can lay bare internal conflicts within a populist party, conflicts that can spread like wildfire (Vergeer and Hermans 2013, p. 407). Together this reduces the motivation to make broad professional use of social media.¹⁴ Hence, populist parties can particularly be expected to discourage their politicians from using social media.

Based on our framework, we thus expect populist parties to be hesitant about using social media. In the empirical part of this study, we will therefore both test the expectation from our new framework (Expectation 3) and the traditional expectation (Expectation 3A) voiced in the existing literature.

Expectation 3

Populist parties are behind the postmaterialist parties in the early-adoption phase and stay behind the postmaterialist and major parties during the widespread-diffusion phase.

Expectation 3A

Populist parties are ahead in the early-adoption phase and stay in the lead in the widespread-diffusion phase.

Methods and data

Cases: parties in the Netherlands

To test our expectations, we will focus on the Netherlands. In the Dutch electoral system, parties present a list of candidates, and the ballot structure (flexible lists) is fairly typical in the European context (Andeweg and Irwin 2005). The campaigns are generally run by the Dutch parties, not the candidates. Candidates mostly rely on their list position to be elected and largely campaign for the party as a whole (Andeweg and Irwin 2005, pp. 92–97).

The highly proportional system results in a high number of parties in parliament (Andeweg and Irwin 2005), which allows for one of the most succinct tests of which party characteristics matter. In the two most recent parliamentary elections (9 June 2010 and 12 September 2012), a heterogeneous group of ten and eleven parties, respectively, received at least one of the 150 seats, and these parties provide variation on all key variables in our model. Hence, it presents an excellent case to disentangle how professional social media use does or does not link to certain (combinations of) party characteristics.¹⁵

Moreover, in the Netherlands, in 2010 social media were still in their early-adoption phase, and by 2012 the Netherlands already entered the widespread-diffusion phase (Jacobs and Spierings 2016). For instance, the number of candidates on Twitter rose considerably between these elections (from 34 to 76%), as did the number of specialized social media managers (Jacobs and Spierings 2016). So although the time gap between the two elections is only 2 years, we see a rapid development in social media use. This makes the 2010 and 2012 elections a useful testing ground for the two-step logic of our model. In the conclusion, we reflect upon the application of our model to more closed list systems and hyper-personalized systems.

Qualitative comparative analysis

We analyse the parties winning seats in two elections, which results in a total of 21 units of analysis—a large and diverse pool of cases compared to the existing literature, but in methodological terms it is still considered medium-N. Qualitative comparative analysis (QCA) is particularly well suited for systematic comparisons of such a medium number of cases (Berg-Schlosser et al. 2009, pp. 5–6). The method allows us to test whether the expectations that we generated from our framework based on the literature and in-depth interviews empirically hold. While QCA cannot probe into the underlying mechanisms and detailed timing of the events—that would require a more historical analysis—it does allow us to gain insight in the broader patters.

Just like there are different types of regression analysis, there are different types of QCA. Of the QCA family, we employ crisp-set QCA (csQCA), as it is particularly suitable for medium-N analysis relying on substantive case knowledge (Berg-Schlosser et al. 2009, pp. 5–6). In csQCA, characteristics are either present ‘1’ or absent ‘0’, which in Boolean formulas is also denoted in uppercase (= 1) and lowercase (= 0) variable names. The variables are thus dichotomous, but the dichotomization—or ‘calibration’ in QCA language—is not based on numerical cut-off points alone. In-depth case knowledge plays an important role too.¹⁶ Most of our variables are more or less dichotomous in nature. Still, some are not, such as professional use and party size. Therefore, we will explicitly discuss borderline cases in the operationalization sections (Rihoux and De Meur 2009, pp. 39–44) and test to what degree our results are sensitive to changes in the cut-off points (see “Results” section).

In this study, we follow the ‘best practices’ (i.e. ‘steps’) that Rihoux and De Meur (2009) have put forward in their QCA handbook. Specifically, Rihoux and De Meur outline six steps that each QCA should take. The first step deals with the operationalization and will therefore be covered in Methods and data section (Sections “Dichotomizing professional use of social media (dependent variable) (Step 1a)” and “Dichotomizing the explanatory variables (Step 1b)”). The other five steps are genuine analytical procedures and will be covered in Results section. Step 2 involves summarizing the data in such a way that cases with the same scores on the explanatory variables are grouped. In Step 3, it is assessed whether there are combinations of explanatory variables (‘configurations’) that yield contradictory values on the dependent variable, i.e. when similar parties have different scores on the dependent variable. Steps 2 and 3 are covered in the section “Configurations (Step 2 and 3)”. In Step 4, we analyse whether more parsimonious configurations can be formulated without losing explanatory power (Section “Minimization (Step 4)”). Step 5 builds on this, but includes configurations that were not observed empirically (‘logical remainders’) and makes the assumptions underlying their inclusion explicit (Section “Bringing in the logical remainders (Step 5)”). Finally, Step 6 in Rihoux and De Meur’s (2009) approach involves the interpretation, which we cover in Results section (mainly in Section “Bringing in the logical remainders (Step 5)”) and the conclusion.

Dichotomizing professional use of social media (dependent variable) (Step 1a)

Most recent research on social media use distinguishes two key dimensions: (1) social media presence and (2) a professional approach (see Williams and Gulati 2012; Koc-Michalska et al. 2014; Larsson and Moe 2014). The first dimension simply refers to whether the parties and their candidates are present on Twitter and Facebook. Regarding the second dimension, the literature provides four types of professional use (cf. Gibson and Ward 2000; Koc-Michalska et al. 2014; Schweitzer 2011):

• (1) Information Via social media, parties disseminate information about their identity and policies;

• (2) Engagement Via social media, parties create opportunities for users to get engaged with them (e.g. sharing video clips, pictures to change a user’s profile in line with the party, polls);

• (3) Mobilization Via social media, parties stimulate users to become active for their organization (e.g. donate, campaign, vote); and

• (4) Interaction Via social media, parties interact with users (e.g. so-called webcare).

We have classified our cases 0, 0.5, or 1 on each dimension as discussed below and as summarized in Table 1.

Table 1 Professional social media use

Presence

To assess ‘presence’ we rely on whether a party had a public Twitter and Facebook account in the respective elections (cf. first two columns in Table 1) and on how successful a party was in getting its candidates to open a Facebook or Twitter account (see Jacobs and Spierings 2016; Vergeer et al. 2013). There is no clear theoretical cut-off point to anchor the dichotomization, so following Rihoux and De Meur’s guidelines (2009, p. 42) we based it on the distribution of cases. For each election, we distinguish three groups. In 2010, candidates of the Greens were clearly ahead of the pack (Vergeer et al. 2013), and in the pack five parties were clearly behind the remaining four (Jacobs and Spierings 2016; Vergeer et al. 2013). By 2012, two parties lagged behind on both Facebook and Twitter. On Twitter, six of the remaining parties were ahead: 80% or more of their candidates were present there (Jacobs and Spierings 2016, p. 80). On Facebook, four parties were ahead of the others.

Professional strategies

To assess whether parties applied the four professional strategies emphasized in the literature, we used case knowledge derived from mini case studies and ‘behind-closed-doors’ information derived from interviews with nine of the eleven parties. Two parties, 50Plus and PVV, really kept the door closed: we were unable to obtain a formal interview; hence, we made use of secondary sources. Full descriptive reports on each party can be obtained from the authors.

In 2010, only three parties used both Facebook and Twitter to send out particular information, two only used Twitter, and several paid little attention to social media. For instance, the social democrats (PvdA) were ‘not really active on social media’ according to their own social media manager (Interview #4). Contrastingly, by 2012 most parties did have a clear plan on sending information via social media and used both platforms. Not all did. For instance, the socialist party (SP) told us that they ‘started using social media systematically for the 2012 campaign. But (…) it got snowed under. (…) We did not really have a strategy’ (Interview #2) (hence: 0.5 in table).

In terms of engagement, the situation in 2010 was clear: only the Greens, progressive liberals, and Animal Party focused on engagement. As one progressive liberal MP described: ‘I then [tweeted] I am ready, ask your questions. I have an hour to react’ (Interview #1). In 2012, engagement was more widespread, but five parties still did not focus on it. For instance, PVV hardly engaged their electorate; their leader Geert Wilders did not even reply to a single Twitter question (Dietz 2013).

A similar pattern was found for mobilization, which was even less widespread in both 2010 and 2012. In 2010, only the Greens, and to some extent the Animal Party, incorporated mobilization in their strategy, as illustrated by the latter’s social media manager referring to a book about the Obama campaign: ‘[it] gave me tips about (…) using volunteers, creating social media teams who retweet you and spread the message’ (Interview #10). In 2012, mobilization was more common. In an interview with the magazine Recruitment Matters (Drees 2012), the CDA social media manager described the activities of his party: ‘we … analyzed who posted these messages and approached people who wrote a lot and/or positively about the party online and asked them to act as an ambassador for the party’.

Regarding the fourth dimension, interaction, a similar pattern was found. For instance, the ultra-orthodox Christian SGP hardly had an interactive strategy. Their social media manager told us that even at the time of the 2012 elections ‘our MPs used Twitter, but … We were very hesitant in reacting to people’ (Interview #7).

Final dichotomization

The last two columns (Table 1) provide the overall picture per election. To determine the cut-off point for the final dichotomization of professional social media use, we first calculated the relative score across all the sub-dimensions. Then, we followed Ragin’s (2007, p. 187) guideline to consider all scores of 0.8 or higher to signify the presence of professional use of social media (‘PU’; ‘pu’ indicates the absence of professional use).

Cases between 0.6 and 0.8 should be considered borderline cases (Ragin 2007, p. 187). We have three, which are coded ‘PU’ based on the overall case information. For each of those parties, at least 3 out of 4 elements of professionalization were found. Also each one of them is above the empirical gap (between 0.5 and 0.71) in the actual distribution (see Rihoux and De Meur 2009, p. 42). Sensitivity tests regarding these three categorizations are conducted, and where relevant this is mentioned in Results section.

All in all, three parties are considered to have professionally used social media in 2010 and seven in 2012. The final dichotomization of our variables is presented in Table 2.

Table 2 Dichotomous data table

Dichotomizing the explanatory variables (Step 1b)

In line with the literature, we consider parties to be postmaterialist when they prioritize postmaterialist values, such as the quality of life, freedom, and self-expression (Inglehart and Abrahamson 1999, p. 665). According to Lijphart (1999, pp. 86–87), the Netherlands counted two such parties: the progressive liberals (D66) and Greens (GL). The newer PvdD, which advocates animal welfare (Partij voor de Dieren 2005), is also considered postmaterialist. The three parties are designated ‘PMAT’; the rest ‘pmat’.

Regarding party size, the Netherlands was long dominated by parties representing the three former pillars, the Christian democrats, social democrats, and liberals (Andeweg and Irwin 2005, p. 47). Still, these are regularly called the ‘big three’ (e.g. Louwerse 2013). In the 2010 and 2012 general elections, the social democrats (PvdA) and conservative liberals (VVD) were also the electorally dominant parties. The Christian democrats (CDA) remained strong at the local level as they were the largest national party in the 2014 local elections. They also have by far the most members (DNPP 2016). Looking at vote shares across all eight nationwide elections since 2009, indeed all other parties are smaller (Jacobs and Spierings 2016, p. 49). As the three are thus clearly the largest in organizational terms and vote shares (on which funding is based), we consider CDA, PvdA, and VVD to be the bigger parties (‘BIG’; the others: ‘big’).¹⁷

Regarding populism, we follow the seminal definition by Mudde (2007, p. 23), who views populism as ‘an ideology that considers society to be ultimately separated into two homogeneous and antagonistic groups, ‘the pure people’ versus ‘the corrupt elite’, and which argues that politics should be an expression of the volonté générale (general will) of the people’. Based on Mudde’s definition, Rooduijn (2014, p. 732) identified two populist parties in the Netherlands: SP and PVV (‘POP’; vs. ‘pop’ for non-populist parties).

Our model expects different variables to lead to professional use in the two phases of diffusion. This is captured by coding all parties ‘WID’ for 2012, the widespread-diffusion phase. In 2010, it was still the early-adoption phase (‘wid’) (see “Cases: parties in the Netherlands” section).

Lastly, we included a control for Incumbency, as some studies have suggested that non-incumbent—or ‘challenger’—parties have more incentives to experiment with new media to bypass barriers in the offline world (cf. Williams and Gulati 2012). Our operationalization indicates which parties were in government or not (‘GOV’ vs ‘gov’). In 2010, CDA, PvdA, and CU were the incumbents; in 2012 VVD, CDA, and PVV were included or formally supported the (minority) government.

Results

This section covers the analysis of which combinations of independent variables (‘configurations’) are sufficient for explaining which parties used social media professionally and when. As is the rule in QCA, we analyse both which configuration leads to the presence of professional use and those that lead to the absence of it (Rihoux and De Meur 2009).

Configurations (Steps 2 and 3)

Table 3 (the so-called truth table) shows the empirically observed combinations of independent variables (‘configurations’) and is basically a collapsed version of Table 2. Each row provides a unique empirical configuration. For instance, the ninth configuration covers all postmaterialist (PMAT), smaller (big), non-populist (pop), opposition (gov) parties in the widespread-diffusion phase (WID) (PMAT*big*pop*gov*WID). Three cases fit this configuration (D66, GL, and PvdD in 2012), and all three used social media professionally (see last column; PU).

Table 3 Truth table

The truth table also shows a combination of party characteristics that is related to both a positive and negative outcome (#7)—a ‘contradictory configuration’. Of the three non-postmaterialist, smaller, non-populist parties in 2012 that were in opposition, two did not use social media professionally, yet one did: CU2012. This was one of our three borderline cases (see “Dichotomizing professional use of social media (dependent variable) (Step 1a)” section), but it clearly scores higher on professional use than the other two (see Table 1). This indicates either an error in the operationalization or that a variable is missing in the model. Reassessing the operationalization does not lead to a different classification. Therefore, we follow ‘Good Practices #5 and #6’ of Rihoux and De Meur (2009, p. 49) by (temporarily) taking this case from the analysis and studying it in detail as an outlier later (see Step 5c). Evidently, we also discuss to what extent our results below are sensitive to taking this case out.

The remaining 12 configurations cover 20 observable cases and can be visualized in a five-dimensional Venn diagram (Fig. 1). Each condition basically splits the space in two areas, and all 32 combinations of 5 variables are included as a cell. The 12 cells with observable cases are marked green (or light-grey) if that configuration is associated with the presence of professional use of social media (PU) and red (or dark-grey) if it is with the absence of professional use (pu). This visualization of Table 3 helps checking and explaining the simplification of the configurations (Steps 4 and 5)

Fig. 1

Venn diagram on the professional use of social media by political parties. Green/light-grey = presence professional use; red/dark-grey = absence of professional use; white = absence of empirical cases. (Color figure online)

Minimization (Step 4)

To examine which party characteristics are relevant under which circumstances, we first have to assess which factors are superfluous in determining professional use. For instance, smaller postmaterialist, non-populist parties in opposition (big*PMAT*pop*gov) show professional use in both the widespread-diffusion phase (WID) and the early-adoption phase (wid)—the two green/light-grey cells in the centre of the Venn diagram. Substantively, this means that being such a party is by itself sufficient for professional use, regardless of the diffusion phase. Conducting this exercise systematically reduces the four configurations sufficient for professional social media use (Table 1: configurations #4 #8 #9 #11) to two more parsimonious configuration [covered cases are between brackets]¹⁸:

• (1a) PMAT*big*pop*gov [PvdD2010, PvdD2012, D662010, D662012, GL2010, GL2012]

• (2a) pmat*BIG*pop*WID [PvdA2012, CDA2012, VVD2012]

All smaller, postmaterialist, non-populist parties in opposition (regardless of the diffusion phase) (1a) and all bigger non-postmaterialist, non-populist parties in the widespread-diffusion phase (regardless of whether they are in government) (2a) used social media professionally.

Regarding the configurations that led to not using social media professionally (the red/dark-grey cells in Fig. 1 and configuration #1 #2 #3 #5 #6 #7 #10 #12 in Table 3), the minimization yields three sufficient configurations¹⁹:

• (3a) pmat*pop*wid [VVD2010, SGP2010, CU2010, CDA2010, PvdA2010]

• (4a) pmat*big*gov [SP2010, PVV2010, SP2012, SGP2010, SGP2012, 50Plus2012]

• (5a) pmat*big*POP*WID [SP2012, PVV2012]

Thus, all non-postmaterialist, non-populist parties in the early-adoption phase (3a); all non-postmaterialist, smaller, opposition parties (4a); and all non-postmaterialist, smaller, populist parties in the widespread-diffusion phase (5a) did not use social media professionally.

Bringing in the logical remainders (Step 5)

So far, we assessed which empirically observed configurations are sufficient for the presence/absence of professional social media use. Figure 1 indicates that including empty, white cells (the ‘remainders’) can lead to even more parsimonious results. However, to do so the empirical and theoretical priors need to logically point towards one and only one of the two possible outcomes. Doing so implies making assumptions about not-observed cases. Step 5 in csQCA makes these assumptions explicit and leads to a proper assessment of whether these assumptions are logical and valid (Rihoux and Ragin 2009, pp. 59, 63–6).

Figure 2 shows the final Venn diagram including the assessments for the logical remainders. The logical remainders included for explaining presence/absence of professional social media use are shaded in both colours, but have no empirical case in them, only the letters ‘a’ through ‘h’. These letters refer to the arguments we provide for those cells below.

Fig. 2

Venn diagram including logical remainders. Legend: green = presence professional use (incl. logical remainders); red = absence of professional use (including logical remainders; horizontally striped = impossible configurations (can be included in minimization for both outcomes). (Color figure online)

Step 5a: professional use

• a. Let us start with a logical remainder that is straightforward to handle: the absence of cases that are both populist AND postmaterialist (PMAT*POP—indicated in Fig. 2 by ‘a’). Indeed, the definitions of postmaterialism and populism simply exclude each other (e.g. Mudde 2007; Inglehart and Abrahamson 1999), which means these cells, by definition, can never contain an empirical case.²⁰

• b. Furthermore, it seems safe to assume that postmaterialist parties in opposition would still have used social media had they been bigger parties (cells ‘b’). After all, the main difference is that in such a case they would actually have more resources to devote to social media.

• c. Our interviews gave no indication that being in opposition (gov) mattered for the postmaterialist parties (gov*PMAT) in terms of professional social media use. They explicitly linked it to their ‘approachable’, ‘modern’, ‘creative’ identity, not them being in opposition. This suggests that we can include these two cells.

• d. Combining argumentation b and c, this also means that hypothetical bigger, postmaterialist parties in government would probably have used social media professionally.

Including cells ‘a’ through ‘d’ further minimizes configurations 1a and 2a to two combinations of party characteristics that seem sufficient for professional social media use:

• (1b) PMAT → PU [PvdD2010, PvdD2012, D662010, D662012, GL2010, GL2012]

• (2b) BIG*pop*WID → PU [PvdA2012, CDA2012, VVD2012]

First, being a postmaterialist party by itself is actually sufficient for professional use in both diffusion phases (1b). Second, bigger non-populist parties catch up in the widespread-diffusion phase, regardless of whether they are postmaterialist or not and regardless of whether they are in government or not (2b). This preliminarily conclusion confirms Expectations 1 and 2 and undermines Expectations 3A.

Step 5b: no professional use

Regarding not using social media professionally, further minimization is possible by examining mirroring or neighbouring cells in the Venn diagram for which we do have empirical cases.

• e. We start with smaller, populist parties in government in the early-adoption phase. If a small populist governing party in the widespread-diffusion phase (PVV2012) does not use social media professionally, there is little reason to assume that a similar party in an earlier phase would. Similarly, if a small, non-populist, government party in the early-adoption phase does not use social media professionally (CU2010), there is little reason to expect that a similar, but populist party would (cell ‘e’) as none of the populist parties in our analysis used social media professionally.

• f. For cells ‘f’, covering bigger populist parties in the early-adoption phase, a similar reasoning can be used. Indeed, their non-populist counterparts (CDA2010, PvdA2010, VVD2010) did not use social media professionally. The fact that SP2010 and PVV2010 did not use social media professionally either further suggests that we can include cells ‘f’ in the minimization.

• g. Moving to the bigger, populist parties in the widespread-diffusion phase, again we use a similar logic. However, in this case the argument is a bit more tentative as their bigger non-populist counterparts did use social media professionally (CDA2012, PvdA2012, VVD2012). Yet, as discussed in Methods section, especially at the national level the PVV is a borderline case in terms of party size. Were we to treat it as a bigger party, this would turn cells ‘g’ into observed cases that are in line with the argumentation assessed here. Therefore, we consider it plausible to include the aforementioned remainders in the minimization for ‘pu’.

• h. Lastly, we assess cell ‘h’: small, non-populist parties in government in the widespread-diffusion phase. The empirically observed cases one can compare them to are the small, non-populist opposition parties in the widespread-diffusion phase (SGP2012, 50Plus2012). They show no professional use. As none of the empirical observations indicated that government participation leads to professional use, we can tentatively assume that in cell ‘h’ a negative outcome would be found. This assumption is further strengthened by the observation that the SGP actually supported the minority government in the senate in the widespread-diffusion phase (Trouw 2011).

Including cells ‘a’ and ‘e’ through ‘h’ in the minimization leads to three configurations for ‘pu’:

• (3b) pmat*wid → pu [VVD2010, SP2010, PVV2010, SGP2010, CU2010, CDA2010, PvdA2010]

• (4b) big*pmat → pu [PVV2010, PVV2012, SP2010, SP2012, SGP2010, SGP2012, CU2010, 50Plus2012]

• (5b) POP → pu [SP2010, SP2012, PVV2010, PVV2012]

Populist parties do not use social media professionally (5b), nor do non-postmaterialist parties in the early-adoption phase (3b). Smaller, non-postmaterialist parties remain behind too (4b—more on this below). These results refute Expectations 3A but support Expectations 2 and 3.

Step 5c: outlier

The positive CU2012 case was taken out earlier (Section Configurations (Step 2 and 3)). Keeping it in would imply that the cell with SGP2012 and 50Plus2012 could not be used in the minimization. Based on case knowledge, we shall assess why this party was able to use social media in a relatively professional way, and whether this undermines our interpretation of the results.

Our analyses indicate that CU2012 somehow had the resources to catch up, even though we expected that smaller, non-postmaterialist parties do not have these. The in-depth interview with the social media manager actually confirmed our assumption that the party did not have financial resources (cf. bigger parties) or free expertise from tech-savvy volunteers and politicians (cf. postmaterialist parties). However, the party was in the unique position that it could circumvent this lack of resources by making use of their ‘kindred spirits’ at the Evangelical broadcasting organization, who had a social media team of seven professionals and used advanced software (interview #3). Hence, the general pattern that smaller, non-postmaterialist parties do not use social media professionally in the widespread-diffusion phase (4b)—because of a lack of resources, despite their external motivation—can be overwritten if that party finds a way to tap into free resources externally. At the same time, our earlier pattern remains valid as this explanation for our outlier is fairly idiosyncratic.

Conclusion

This study examined why, when, and how parties use social media professionally. We first developed our new motivation-resource-based diffusion model based on the existing literature and in-depth interviews with Dutch social media managers (cf. RQ1). Our framework builds on the two-stage logic of Gibson and McAllister (2015) and a refined understanding of how resources and motivations matter in parties’ professional use of social media. Indeed, contrary to conventional wisdom claiming that social media are cheap and easy to use, our interviews suggest that financial resources can convey significant direct and indirect advantages, such as higher-quality posts, expertise, time, and personnel capacity. Some ‘free resources’ such as tech-savvy activists are available, but not all parties can tap into them. Regarding motivation, we suggest that extrinsic and intrinsic motivation may play a role, whereby some parties also have incentives not to use social media professionally.

From this new framework, we derived tangible expectations about when which type of parties would (or would not) professionally use social media (cf. RQ2). Given our medium-N sample of 21 cases and the configurational logic of our expectations, we used csQCA as a method to test these expectations. We found that smaller, postmaterialist parties were ahead in the early-diffusion phase. Bigger parties caught up when social media became more widespread as was also predicted by our framework (see also Gibson and McAllister 2015). However, our analyses also suggest that if one of these major parties had been populist, it would most likely not have caught up. Indeed, populist parties remained behind, which is in line with the idea that they have motivations not to use social media professionally, regardless of their resources. Lastly, smaller, non-postmaterialist parties might have the motivation in the widespread-diffusion phase to catch up too, but they usually lack the money or expertise to do so, unless they can tap some external resources, as our outlier case demonstrated.

Not only did the QCA allow us to test our model systematically, it also helped to contrast different explanations. And it made it possible to refine our understanding of which characteristics matter most because our analysis covered a highly diverse set of parties. For instance, in Gibson and McAllister’s (2015) seminal study the Greens profited from social media, but they were green and small and in opposition at the same time. This makes it impossible to assess which (combination) of the three characteristics explains the outcome. Our QCA strongly suggests that postmaterialist parties would use social media professionally regardless of their size or incumbency status.

Taken together, our multi-method study also sheds new light on the equalization–normalization debate (RQ3) by providing a systematic analysis of the broader patterns (csQCA) as well as providing more insight into the ‘why’ question (in-depth interviews). Our analyses corroborate Gibson and McAllister’s two-step model: in 2010, the postmaterialist parties dominated social media and improved their position vis-à-vis the other parties (equalization). In 2012, the major parties caught up (normalization). Interestingly, this still means that the postmaterialist parties were ahead of, for instance, the populist parties. Indeed even in 2012, the power balance within the group of smaller parties was altered. Equalization and normalization thus also depend on which party one compares to. Moreover, we showed that populist parties are more likely to lag behind. So far, these parties have been mostly absent from social media analyses.²¹ They seem to have a love–hate relationship with social media. Their party leaders make heavy use of social media, but the other politicians of populist parties lag behind. As our analysis was at the party level, this study could just explore this particular tension. Populists clearly deserve more attention, especially concerning their intra-party politics.

Every study necessarily has limitations, so has ours. Specifically, we wish to stress limitations relating to timing and insights in the mechanisms behind our patterns. First, our study examined two elections. Since 2012, new technologies have emerged and the diffusion process is likely to have continued. It may well be that by the time the last parties (late adopters) finally make professional use of Twitter and Facebook, the early adopters have already moved on to new campaign innovations. In that sense, cyber campaigning resembles a perpetual cat-and-mouse game. However, our theoretical framework is based on more general building blocks (intrinsic and extrinsic motivations and resources), and as such, it should be a good starting point to formulate expectations regarding such new technologies (e.g. Instagram, Snapchat, vlogs, personalization via micro-targeting). To illustrate: when we apply our framework to micro-targeting, intrinsically motivated and tech-savvy activists postmaterialists can still be expected to adopt micro-targeting early on. Moreover, the bigger parties still have an extrinsic motivation and have the resources to invest in it. Yet our framework also suggests that the time gap between the postmaterialists and the bigger parties is likely to be even shorter because micro-targeting makes use of existing platforms where the voters already are (extrinsic motivation). Furthermore, the position of populist parties is less clear. As micro-targeting requires more financial resources, it is less of a threat to the centralized party leadership. Hence, the motivations remain mixed for populists parties, though less so compared to classic social media use. Once sufficient data on, for instance, the Dutch 2017 elections are available, these expectations could be tested.

A second limitation refers to the type of political system. We only applied our model to the Dutch case, and while the model can be easily used in other systems with a similar flexible list ballot structure, one could argue that this is less so for more personalized systems or systems where politicians have an even less independent position from parties. As highlighted above, one of the advantages of our model is that it is based on more general building blocks allowing us to formulate expectations for different contexts. In less flexible ballot structure systems (i.e. closed list systems such as Spain), individual candidates have lower extrinsic motivations to use social media professionally, as they cannot influence their electoral fates once they are on the ballot. ‘Self-promotion via social media’ (Kruikemeier 2014, p. 132) makes less sense in such a system. This implies that the centralized, populist parties have fewer reasons to fear that their politicians build a personal power base by using social media. At the same time, the risk of making internal conflicts public remains. The impact of other motivations and resources can be expected to remain the same to a large extent as well. For populist parties, we thus expect similar outcomes in countries with less flexible ballot structures than the Netherlands. Another case is the majoritarian, hyper-personalized system (such as the UK), where parties field just one candidate in each districts. In such a system whether or not candidates have an extrinsic motivation to use social media depends on characteristics of the electoral district they compete in: the motivation is the highest in close races where every vote matters (versus safe seats). Moreover, once elected, the role of the incumbent politicians is different than in proportional electoral systems: they act as the spokesperson of their district. This in turn may increase their intrinsic motivation to use social media as these allow them to connect with their constituents even if they are not present in their district. As such, one can expect that politicians are in general more likely to use social media, particularly incumbent politicians and those in close-race districts. The role of resources seems to be relatively similar than in proportional systems. Consequently, populists are expect to use social media similarly to other non-postmaterialist parties with similar resources, and parties with more incumbents might adopt social media more (professionally). Again, these expectations are mere starting points, to be tested by systematic analyses covering different countries (and political systems).

A third limitation is that we were unable to test the mechanisms and detailed timeline directly. This would require a more historical analysis. Making use of new technologies often is a strategic decision by the party leadership. A historical analysis including interviews with these key figures can provide crucial information (e.g. when did the party decide to hire specialist personnel? Why did they do so?). Our own interviews can give us some insight in the mechanisms. Where we have such information, we tried to include it in the manuscript. However, it is clear that our interview data are too limited to address the underlying mechanisms systematically, and this is something that remains to be done by future studies.

In short, follow-up research should examine whether the dynamics we propose here and tested for the Dutch case also hold in other countries and later years. While resources and motivation are likely to play a role elsewhere, the exact size, electorate, and party financing may determine how they work in practice.

Appendix 1

See Table 4.

Table 4 List of interviewees