Does Digital Campaigning Make a Difference for Individual Candidates in an Open List Proportional Representation System? The Case of the 2019 Election in Belgium

Abstract

This article looks at the equalization/normalization problem through the lens of campaign spending and investigates the effect of expenses for digital tools on the electoral result of individual candidates in an open list proportional representation system. A multilevel analysis of the campaign expenses of 1253 serious contenders in the Belgian 2019 concurrent federal, regional, and European elections shows that the investment in both owned Web 1.0 media and paid Web 2.0 media does not have an effect on electoral performance. Investing in traditional tools, by contrast, does have a significant positive effect. While most candidates use digital tools, they invest only a small part of their budget in these, which may explain the absence of a digital expense effect. These findings put the use of digital campaigning in perspective, showing that the effect of paid online tools should not be overestimated, and that the role of traditional campaigning is still dominant.

Keywords

digital campaigning political ads campaign expenditure electoral performance equalization/normalization

Introduction

It is often argued that political systems in Europe are prone to an increasing personalization. In the electoral field, this implies that voters tend to decide on the basis of their sympathies for political personalities, rather than the ideology or program of the parties (Dodeigne et al., 2023; Dodeigne and Pilet, 2021; Karvonen, 2010; Van Aelst et al., 2012; Wauters et al., 2018). In such a context, it is crucial for politicians to relate personally to the voter, or at least give this impression. Although digital campaigning has been around since the 1990s, it is especially the emergence of social media that allows candidates to direct seemingly personal and tailor-made messages to individual voters. Such targeting via paid digital media appears to be the perfect campaign tool in an era of highly personalized politics. If this is the case, we would expect these tools to be more effective than traditional media or older web-based digital tools. Yet, this has never been investigated.

According to the equalization model of digital campaigning, it is particularly the challengers and candidates with less resources who benefit from these new digital tools, which are relatively cheap and easily accessible (Corrado and Firestone, 1996; Gibson and Mcallister, 2015; Gibson and Ward, 2000; Norris, 2003; Xenos and Foot, 2005). This view is disputed by the proponents of the normalization model, who argue that politicians who are most powerful offline also hold that power online (Jacobs and Spierings, 2016; Lusoli, 2005; Margolis et al., 1999; Margolis and Resnick, 2000).

This equalization/normalization debate in the digital campaigning literature echoes an older discussion in studies on campaign expenditure effects. To what extent are expenses more beneficial to unknown challengers than to highly visible incumbents? As we will discuss further below, such an incumbency effect appears typical for majoritarian systems, but is much less robust in proportional systems. However, these studies almost always measure the effect of expenses in general, and do not distinguish between the campaign tools on which the money is spent. If digital campaign tools, and especially social media, involve an equalization dynamic, we would expect these tools to be particularly effective for challengers. In other words, the incumbency effect may materialize in proportional representation (PR) systems if we focus on digital campaign expenses.

While it is useful to look at the traditional issues in the campaign expense literature from a digital campaign perspective, the reverse is also true. We may gain new insights into the equalization/normalization problem by viewing it through the lens of research on campaign expense effects. As already argued by Gibson and Mcallister (2015), it only makes sense to talk of normalization or equalization to the extent that digital tools are effective and result in electoral gains. Yet, till date, this has hardly been investigated. Some studies look at the candidates’ adoption of a single online instrument, such as websites or one of the social media networks, and the quality of their content to assess the equalizing potential of online tools (e.g. Gibson and McAllister, 2006; Spierings and Jacobs, 2014; Sudulich and Wall, 2009; Suiter, 2015). But what we really need to assess the impact of digital tools, as compared with traditional tools, is a precise measure of their use. We argue that the expenditures for these tools constitute such a measure.

In this article, we leverage the availability of highly detailed data about the campaign expenses of candidates in the Belgian federal, regional, and European elections of 2019. These data allow us to make a clear distinction between expenses for traditional campaign tools, Web 1.0 tools and Web 2.0 tools. Traditional tools refer to offline campaign instruments such as flyers, posters, and offline ads. Web 1.0 tools specifically refer to the older online technologies which are best suited for top–down information sharing such as websites. In contrast, Web 2.0 tools refer to the more modern online technologies which allow for (micro) targeted and interactive campaigning such as (ads on) social network platforms (Copeland and Römmele, 2014; Gibson, 2020; Southern, 2015). We estimate the effect of these three types of expenses on the electoral success of 1253 high-quality candidates in terms of preferential votes. We use the term “high-quality candidates” to refer to candidates who have a realistic chance of winning a seat as opposed to the no-hoper candidates who do not. The analysis shows that the expenses for both owned Web 1.0 media tools and paid Web 2.0 media tools do not have an effect on the number of preferential votes. Investing in traditional tools, by contrast, does have a positive effect. We do not find a robust interaction effect between either of these expense variables and incumbency.

Theory and Hypotheses

This article relates both to the recent literature on digital campaigning and to the older literature on campaign expenses effects. As will be seen below, campaign expenses are generally found to have a positive effect on the electoral performance of candidates. But these studies do not usually distinguish between expenses for various types of media tools. Research on the efficiency of older and newer digital media tools in comparison to traditional tools is scarce and does not lead to unequivocal results.

Some studies focus only on the effect of digital campaigning on electoral performance, omitting traditional tools from the analysis. For instance, Gibson and McAllister, (2011, 2015) found a positive effect of digital campaigning on the electoral result, albeit in a majoritarian system. Focusing on preferential votes in the Danish PR system, Hansen and Kosiara-Pedersen (2014) demonstrated that the effect of digital campaigning disappears once the use of traditional tools and other relevant variables are controlled for. These results are not confirmed by studies on Poland (Koc-Michalska et al., 2014) and Estonia (Trumm, 2022), which do find an effect of digital campaigning on the candidate’s electoral performance. But contrary to the Danish study, these authors do not control for some relevant control variables such as media coverage, the overall campaign budget or, in the case of the Estonian study, the use of traditional tools.

Neither of the above studies makes a distinction between various forms of digital tools. In the marketing literature (e.g. Vieira et al., 2019), a distinction is generally made between the older Web 1.0 owned media applications, such as websites, and the more recent Web 2.0 paid media applications, such as targeted ads and sponsored post on social media. Applied to election campaigning, the Web 1.0 owned media mainly pertain to candidates’ websites. These are created and controlled by the candidate and are publicly accessible but cannot be targeted at specific groups. Paid digital media, by contrast, are particularly well suited to target specific groups of potential voters with tailored campaign messages that vary according to the target audience. Commonly used paid digital media tools in campaigns are sponsored posts on social media and online ads (Kruschinski and Bene, 2022).¹

A second issue concerns the measurement of digital campaigning. Koc-Michalska et al. (2014) apply a more qualitative measure tapping into the degree of sophistication of the digital campaign. Hansen and Kosiara-Pedersen (2014) calculate an online score on the basis of a number of qualitative indicators, such as having a website, updating it daily, having a Facebook site. The latter authors (2014: 213) argue that “it is difficult, if not impossible, to link all campaign activities with individual voters and their voting choices and thus accurately estimate the effects of campaigning due to the plethora of actual campaigning conducted (. . .).” While this is undoubtedly true, we argue that we can get closer to capturing the actual (digital) campaign by measuring the expenses of candidates for digital tools, as declared in the official disclosure documents. In this way, we can not only identify the candidates who have invested in these tools, but also measure to what extent they have done so. An additional advantage of these data is that they allow us to make a clear distinction between the use of Web 1.0, Web 2.0, and traditional tools.

On the basis of earlier studies on the effect of campaign expenses in general, we can expect that the spending for each of the campaign tools distinguished in the analysis will have a positive impact on the electoral result in terms of preferential votes. But we expect the expenses for paid digital media (Web 2.0) to have a larger effect compared with both traditional tools and owned Web 1.0 media. After all, Web 2.0 technology allows for fine-grained campaigns including (micro) targeted political ads on social media. It has the potential to deliver political messages which are congruent with the ideas and even the personality of potential voters. This results in messages that are more effective when it comes to influencing voting intentions compared with more generic campaign communication (Römmele and Gibson, 2020; Zarouali et al., 2022). On the basis of these arguments, four testable hypotheses can be formulated:

H1: Expenses for owned Web 1.0 media have a positive effect on the candidate’s number of preferential votes.

H2: Expenses for paid Web 2.0 digital media have a positive effect on the candidate’s number of preferential votes.

H3: Expenses for traditional offline campaigning tools have a positive effect on the candidate’s number of preferential votes.

H4: The positive effect on the candidate’s number of preferential votes is stronger for investments in paid Web 2.0 digital media than for the investment in the older Web 1.0 tools or the traditional offline tools.

Most of the earlier research on campaign spending effects on electoral performance centers around Jacobson’s (1978) finding that spending efficacy is higher for challengers than for incumbents. This is so because for challengers, spending is the only means to obtain visibility and credibility, while incumbents are already well known and have less to gain from spending. This literature is mainly focused on majoritarian systems (Abramowitz, 1988; Green and Krasno, 1988, 1990; Jacobson, 1978, 1990; Johnston and Pattie, 2006; Palda and Palda, 1998; Pattie et al., 1995). In proportional systems, on the contrary, such an incumbency effect is less robust. Some studies, for example, on Brazil (Samuels, 2001) or Belgium (Maddens et al., 2006) point at equal returns for incumbents and challengers. Benoit and Marsh (2003, 2010) show support for a challenger advantage in Ireland, but once the possibly endogenous nature of campaign spending is accounted for, the difference in spending effectiveness is less outspoken (Duggan, 2020). In their study of the Swedish case, Folke and Rickne (2020) find that the positive effect of communicating more is limited to candidates not occupying the highest positions on the list, which are mostly challengers. The absence of a robust incumbency effect is probably because incumbents in multi-member constituencies are less well known, especially in large constituencies, and also face the competition of other incumbents, both on other lists as on their own list.

These studies on the incumbency effect normally take overall spending as explanatory variables and do not distinguish between the various media tools for which expenses are made. Most of these studies also predate the digital wave in campaigning. This digital revolution has given rise to a new scientific debate, centering on the equalizing versus normalizing effects of digital campaigning. The distinction between challengers and incumbents again plays a crucial role in this new debate, just as was the case in the older literature on campaign spending effects. As will be explained below, on the basis of this more recent literature the hypothesis can be formulated that the incumbency effect on spending efficacy will be particularly strong for Web 2.0 expenses.

Because digital tools are relatively cheap and easy to use, the proponents of the equalization theory argue that these can level out the electoral playing field. According to this reasoning, the new online possibilities mainly benefit the minor candidates, mostly challengers, with less financial resources and less access to the mainstream media (Corrado and Firestone, 1996; Gibson and Mcallister, 2015; Gibson and Ward, 2000; Norris, 2003; Xenos and Foot, 2005). The normalization theory rejects this thesis and states that the richer and stronger candidates’ offline dominance will be mirrored online (Jacobs and Spierings, 2016; Lusoli, 2005; Margolis et al., 1999; Margolis and Resnick, 2000).

According to the equalization model, digital campaigning is a cost-efficient way for the challengers to obtain name recognition and compete with the incumbents. If this is correct, we would expect challengers to be more inclined to invest in digital campaigning than incumbents. There is some empirical evidence pointing in that direction. In Belgium and Colombia, challengers appear more likely to spend on digital tools, and to invest a larger part of their budget on these tools than incumbents (Vanden Eynde et al., 2022; Vanden Eynde and Maddens, 2021). The equalization model is also supported by some more qualitative studies (e.g. Gibson and Mcallister, 2015; Norris, 2003; Southern and Lee, 2018). But other studies point in the direction of normalization (e.g. Margolis et al., 1999; Margolis and Resnick, 2000; Southern, 2015; Van Aelst et al., 2017).

In any case, both the equalization and the normalization theory assume that online campaigning, whether cheap or expensive, has a positive effect on the electoral result. Viewed from an equalization perspective, we would expect digital campaign expenses to be particularly effective for challengers. These tools offer newcomers the best value for their money as they are relatively cheap while giving them maximum return when it comes to enhancing their visibility and obtaining name recognition. This will be particularly the case for paid digital media, such as (micro) targeted ads or sponsored posts on social media. These tools can lead to messages which are congruent with the ideas and even the personality of potential voters. That makes them more effective when it comes to influencing voting intentions than more generic campaign communication (Römmele and Gibson, 2020; Zarouali et al., 2022). Web 2.0 applications take the personalization of campaign messages to the next level, making them the ideal campaign tools to familiarize the voters with the candidate in a relatively cheap way (Jacobs and Spierings, 2016: 21–24; Sandberg and Öhberg, 2017). This reasoning leads to the following hypothesis:

H5: The positive effect of paid Web 2.0 digital media spending on the candidate’s number of preferential votes is weaker for incumbents than for challengers.

At the same time, the literature on digital campaigning suggests that the applicability of this hypothesis is limited in time. Gibson (2020) explains the inconsistent findings in the normalization/equalization literature on the basis of a four-phase model (Table 1). The first and third phases are experimental phases and took place when new Web 1.0 and innovating Web 2.0 tools were introduced, respectively. These phases lead to equalization as minor candidates and parties with limited resources are the first to experiment with innovative campaign instruments. In the second and fourth phases, professionalization kicks in. This involves a normalization dynamic, because only the stronger and richer candidates and parties can afford to hire digital campaigning experts. The implication is that the advantage of Web 2.0 tools for challengers will be limited to the third phase. In the fourth phase, the competitive advantage of challengers will disappear as only incumbents can afford to pay for the professionalization of social media campaigning.

Table 1.

Gibson’s Four-Phase Model.

	Phase 1 experimentation	Phase 2 professionalization	Phase 3 experimentation	Phase 4 professionalization
Tools	Introduction of Web 1.0 tools	Digital campaign experts take over and use more sophisticated versions of the Web 1.0 tools	Emergence of interactive Web 2.0 tools such as social media platforms	Data-driven targeted online campaigning
Dynamic	Equalization	Normalization	Equalization	Normalization
Cost	Low	High	Low	High

Source: Based on Gibson (2020: 18–19).

Gibson (2020) shows that countries go through these phases at a different speed and that only a limited number of countries have reached the fourth phase. It can be argued that Belgium is a good example of a “third-phase” country, to which H5 applies. As will be seen below, the use of Web 2.0 tools is already well established in Belgian election campaigns, as about three-fourths of the candidates invest in them. At the same time, the absolute amount spent on these tools is still relatively small, which shows that candidates have not yet reached the level of expensive professionalization typical of Phase 4.

Elections and Campaign Expenditure in Belgium

Belgium applies a proportional system with flexible lists for European, regional, and federal legislative elections. Voters either cast a list vote or a preferential vote for one or more candidates on the same list. The allocation of seats to candidates is determined both on the basis of the preferential votes and the list votes, as explained in the Online Appendices under Section A.1.

The pre-electoral intra-party competition for a realistic position on the list is more important than the electoral competition for preferential votes. Even so, the latter competition is also important, as the number of preferential votes is considered a good indicator of a politician’s popularity, and also influences the position on the list at the next election (Andre et al., 2017). In this way, candidates are incentivized to invest time and money in a personal campaign. Furthermore, because voting is compulsory in Belgium, the chances of reaching out to somebody who will eventually vote increase significantly (Gibson, 2020: 56; Vaccari, 2013: 117).

Regional, federal, and European elections are held concurrently in Belgium, most recently on 26 May 2019. This analysis is limited to the federal parliament, the Flemish parliament, the Walloon parliament, and the 21 Belgian representatives in the European Parliament.² There is a substantial variation in district magnitude (M) both within and between these elections. The 150 members of the federal parliament are elected in 11 constituencies with an average M of 13.6, ranging from 4 to 24. The six constituencies for the election of the 124 members of the Flemish Parliament are larger, with an average M of 20.6 ranging from 6 to 33. The 75 representatives of the Walloon Parliament, by contrast, are elected in 11 relatively small constituencies, with an average M of 6.8, ranging from 4 to 13. Finally, the 21 Belgian MEP’s are elected in three region-wide constituencies, with M = 12 for the Flemish constituency, 8 for the Francophone, and 1 for the German-speaking. Parties normally nominate as many candidates as they are allowed to, that is, M effective candidates and about half that number successor candidates.

The Belgian campaign finance legislation imposes a spending cap on both the parties and the individual candidates during the period of 4 months preceding the election.³ A distinction is made between ordinary candidates and a limited number of top candidates. For ordinary candidates, a spending cap of €5000 applies, and €2500 for the candidates on the successor lists.⁴ The spending cap for the top candidates is calculated on the basis of the registered number of voters in the constituency at the previous election and ranges from €15,875 in the tiny constituency of Luxemburg to €54,937 in the constituency of Antwerp and €82,411 for the largest European parliament constituency. The number of top candidates for a party in a constituency equals the number of seats at the previous election plus one (Maddens et al., 2019: 74–76). As already mentioned, political parties are also subject to a strict spending cap. A party cannot spend more than €1 million on its general campaign for all elections taking place on the same day. This spending cap is relatively low and the amount has not changed since the year 2000. Consequently, most parties spend approximately €1 million on their general campaign (Maddens et al., 2019: 74–86). Furthermore, both parties and candidates are subject to strict disclosure rules. They have to submit a declaration regarding their campaign expenses and revenue, with highly detailed information regarding the expenses for different campaign tools. This declaration is public and can be consulted by the voters during a limited period after the election.

Finally, the law imposes strict limits on private donations and bans corporate donations. The result is that only a very marginal part of the candidates’ expenses (less than 1% on average) is financed with donations. This also means that in Belgium there is no simultaneity-dynamic, in the sense that candidates who are expected to win, draw more donations, and therefore wage a more expensive campaign (e.g. Green and Krasno, 1988). In that respect, it is also important to notice that the polls in Belgium generally ignore the fortune of individual candidates and focus on the region-wide results of the parties. As a result, there are no predictions as to how strong individual candidates will perform.

Data and Methodology

High-Quality Candidates

The normal practice in research about campaign expenditure effects is to limit the analysis to a subset of high-quality candidates. In PR systems, parties always nominate a number of marginal candidates who “fill” the list without a realistic chance of winning a seat or even gaining a significant number of preferential votes. The smaller the party, the higher the number of these symbolic candidates on the list. As these candidates combine a very low spending level with a poor electoral result, including them in the analysis artificially inflates the spending effect (Maddens et al., 2006: 163; Samuels, 2001). In identifying the high-quality candidates, we apply the strategy proposed by Cheibub and Sin (2020). All incumbents are included in the analysis. Non-incumbents are included if they participated at least once in the two most recent prior federal, regional, or European elections⁵ and obtained a number of preferential votes larger than 10% of the constituency’s Hare quota in any of these elections. To avoid missing new high potentials, we also included elected candidates who do not meet these criteria. In this way, the total dataset of 4548 candidates running in the 2019 election is reduced to a subset of 1253 high-quality candidates.

Dependent Variable

The electoral result of a candidate is the number of preferential votes as a percentage of the total number of valid votes in the constituency (Bouteca et al., 2019; Maddens et al., 2006; Trumm, 2022).⁶ In this way, we control for differences in constituency size, but not for differences in party size. The latter variable can be controlled for by calculating the number of preferential votes as a percentage of the valid votes for the list only. However, this distorts the dependent variable, as the electoral result for candidates on electorally marginal lists is inflated. A better way to control for party size is to include this variable in the model (cf. infra). As the percentage of preferential votes is highly skewed and its values are all positive, we apply a natural logarithmic transformation (Schroeder et al., 2017: 62).

Main Independent Variables

The detailed information on the expense declarations of the candidates allows us to distinguish between three types of expenses: (1) expenses for innovative paid digital media, that is, ads and sponsored posts on social media, (2) expenses for traditional owned digital media, that is, websites and email, and (3) expenses for traditional non-digital media.⁷ These three expense types add up to the total expenses of the candidate, as in Belgium administrative or legal campaign costs are normally covered by the party.

To check the robustness of our findings, we opt for three different operationalizations of the spending variables. First, we take into account the actual investment for each of the three expense types. To control for the substantial differences in constituency size, the expenses per registered voter in the constituency are calculated, expressed in Eurocents. Second, we include the total expenses of a candidate in the model (per thousand Euros), while adding two binary variables reporting whether a candidate invested in Web 1.0 and Web 2.0 tools, respectively. In this way, we can estimate to what extent using either of these digital tools in the campaign is electorally beneficial for a candidate, in addition to the overall spending effect. Finally, we calculate an alternative relative measure capturing the investment in the different tools by a candidate relative to the total expenditure for the specific tool by all the candidates in the constituency. This measure controls for the differences in constituency size and is calculated as follows, for each of the three expense types separately. We first calculate the share that each candidate spends out of all the propaganda expenses made by all the candidates in the district. As this share is highly dependent on the number of candidates, we divide it by the expected share if expenses were distributed equally across all candidates in the district. Take for instance a district with only two candidates, the first of which spends 70% of all the expenses on traditional media in the district and the second 30%. The standardized expense for traditional media of the first candidate at the district level is then 70/50 = 1.4, and of the second candidate 30/50 = 0.6. In other words, this second candidate only spends 60% of what we would expect if traditional propaganda expenses were distributed equally across all the candidates of the district. In order to check for multicollinearity, we calculated the correlations between the different expenditure variables for each operationalization. The different correlation tables can be found in the Online Appendices Section B.1 and show only moderate correlations.⁸

Previous research has shown that the relationship between the expenses and the vote is nonlinear, in the sense that the marginal returns of spending gradually decrease with the amount spent. This ceiling effect can be captured by also including the squared expenses in the model, which are expected to have a negative coefficient (e.g. Maddens et al., 2006; Samuels, 2001).

A candidate is considered as an incumbent if he or she was an MP or a minister at the federal, regional, or European level at the time of the campaign. In Belgium, there is no hierarchy between these three legislative assemblies and politicians regularly switch between levels (Vanlangenakker et al., 2013). As a result, there is no need to distinguish between the incumbents from the three assemblies.

Control Variables

To estimate the number of preferential votes, we need to control for some mechanical effects at the list level. A candidate running for a larger party will normally obtain more preferential votes, as the pool of available votes is larger. We therefore include in the models the total number of votes for the list as a percentage of the valid votes in the constituency. The availability of votes also varies in function of the percentage of the voters who cast a preferential vote instead of a list vote. This percentage varies both across constituencies and across parties. Voters for traditional parties are more inclined to cast a preferential vote in comparison to the less personalized ecologist and far right parties (Wauters et al., 2019). This difference regarding personalization is also reflected in the average number of preferential votes cast by the preference voters for a list. As mentioned above, voters can cast as many preferential votes as they like, as long as they remain within the list. The larger the number of votes cast by the average preference voter for a list, the larger the pool of available votes for an individual candidate. These three variables determine the total pool of votes available for a candidate. But if these votes are to be distributed across a larger number of candidates, each single candidate will on average receive a smaller number. We therefore also include the total number of (effective and successor) candidates on the list. This variable correlates strongly with the district magnitude, but it is more precise due to the varying rules on the minimum number of successor candidates.

At the level of the individual candidates, it is important to control for list position. Candidates at the most visible positions of the list have been shown to receive more preferential votes, even when other variables such as incumbency are controlled for (Geys and Heyndels, 2003). Taking into account the substantial variation in district magnitude, ranging from 4 to 33, it is not advisable to include the list position as a quantitative variable. Instead, we include dummy variables for the most critical positions: the first, second, and third positions, the last position, and the first position on the list of successor candidates. The candidates on all other positions form the reference category. In addition, we will also include a dummy indicating whether or not a candidate is the highest ranking candidate, in the formal sense of having a substantially higher spending cap than the ordinary candidates (cf. supra). In this way, we can account for the fact that, particularly for the larger parties in the larger constituencies, the number of important positions at the top of the list is larger than three.

Earlier research on Belgium has shown that the media coverage for candidates has a strong effect on the election result. Candidates who receive more media attention on average obtain more preferential votes, controlling for various other variables including campaign expenses. In other words, the direct communication of the candidates to the electorate and the mediated communication each have a separate effect on the election result (Bouteca et al., 2019; Maddens et al., 2006; Wauters et al., 2010). While the number of television appearances has a strong effect for a limited number of top politicians, most candidates are hardly mentioned on national television and are dependent on newspaper coverage for obtaining preferential votes (Van Aelst et al., 2006). Capturing this broader media effect requires measuring the number of times a candidate is mentioned in the newspapers. We counted the number of articles in which a candidate was mentioned in 17 Belgian newspapers, during the official campaign period of 4 months before the election.⁹

It has also been shown that occupying an office at the local level increases the number of preferential votes (Jankowski, 2016; Maddens and Put, 2013; Put and Maddens, 2015; Tavits, 2010). To control for this effect, we include three dummies for occupying the office of local councilor, alderman, and mayor, with candidates without local office as reference category. Gender also needs to be controlled for. The average woman candidate obtains fewer preferential votes, but this difference decreases and sometimes even disappears when crucial variables such as political position, media coverage and campaign expenses are accounted for (Put and Maddens, 2015; Wauters et al., 2010). Finally, we include three dummies to distinguish between the four assemblies for which the candidates run (federal parliament, Walloon Parliament, Flemish Parliament, and European Parliament). In this way, we also control for the fact that the number of voters per seat is relatively large for the European Parliament, and relatively small for the regional parliaments. Particularly the large region-wide scale of the European election may make it more difficult for a candidate to attain a certain percentage of preferential votes.

Analyses and Results

Together, the 1253 high-quality candidates spent €14,104,880 on the 2019 electoral campaign: 17% of that budget, or €2,397,187 was spent on digital tools. This digital budget is almost exclusively spent on paid Web 2.0 media: expenses for owned Web 1.0 tools account for a mere 4.3% of the total digital expenses; 83% of the total campaign budget is still reserved for the more traditional tools such as newspaper ads, posters, flyers, and personalized campaign letters. This dominance of traditional campaign tools is also reflected in the fact that an average candidate spends 87.1% of his or her budget on traditional campaigning, while 12.4% is spent on Web 2.0 online tools and only 0.5% on the older Web 1.0 instruments. In absolute, numbers, the average candidate spent €9354 on traditional tools, €1837 on Web 2.0 instruments, and only €82 on Web 1.0 tools.

At the same time, the fact that a large majority of the candidates (77.9%) reported digital expenses shows that the use of digital campaigns is widespread. Table 2 summarizes these descriptives and shows that this percentage is not vastly inferior to the percentage reporting traditional expenses (91.3%); 74.5% of all the 1253 candidates made expenses for Web 2.0 digital tools, against 40.9% for Web 1.0 instruments. The number of candidates investing in digital tools is actually quite high given the relatively small part of digital expenses in the total budget. Clearly, a lot of candidates took the step of engaging in digital campaigning, but did so only half-heartedly with relatively small amounts. This is particularly the case for the Web 1.0 tools. The candidates who used these tools on average spend only €89.9 or 0.5% of their budget on them. Candidates who used Web 2.0 tools spend on average €2001.9 on these, amounting to 12.4% of their budget.

Table 2.

Descriptive Statistics.

	Percentage reporting digital expenses	Percentage reporting Web 1.0 expenses	Percentage reporting Web 2.0 expenses	Percentage reporting traditional tool expenses	Average expenditure spent on Web 1.0in Euro (SD)	Average expenditure spent on Web 2.0in Euro (SD)	Average expenditure spent on traditionalin Euro (SD)	N
All candidates	77.9%	40.9%	74.5%	91.3%	82.3 (257.1)	1833.6 (4275.6)	9354.4 (12,599)	1253

With 1253 candidates nested in 191 electoral lists and electoral lists nested in 30 constituencies, our data clearly have a hierarchical structure. This is further confirmed by both the null model,¹⁰ which shows the presence of variance on both the constituency and the list level, and the estimation of the intraclass correlation on both Level 3 (constituency) and Level 2 (list), which amounts to 0.159 and 0.527, respectively. Due to the limited number of Level 3-units (N = 30), we use restricted maximum likelihood (RML) to estimate our hierarchical models.¹¹

Table 3 summarizes the results of the different RML models.¹² Our first model only contains the higher-level mechanical effects. As expected, all these mechanical effect variables have a statistically significant effect on the number of preferential votes, and they explain almost all higher-level variance. In the second model, we included the variables reporting the expenses for the different types of campaigning tools under scrutiny relative to the number of registered voters in the candidate’s constituency and the different control variables. The results show that only the investments in traditional (offline) tools have a statistically significant positive effect on the number of preferential votes, providing strong evidence for H3. The statistically significant negative effect of the quadratic term confirms that this relationship is nonlinear and that the positive effect gradually decreases with the amount spent. On the contrary, investing in either Web 1.0 or Web 2.0 digital tools does not significantly increase the electoral result, contrary to what was expected under H1, H2, and H4.

Table 3.

Restricted Maximum Likelihood Estimation with the Logged Number of Preference Votes (As the Percentage of Valid Votes in the Constituency) as Dependent Variable (Unstandardized Coefficients).

	Model 1	Model 2	Model 3	Model 4
Percentage of list in constituency	5.328***(0.311)	6.104***(0.311)	6.151*** (0.322)	6.019***(0.307)
Percentage of pref. votes list	2.201***(0.246)	2.098***(0.220)	2.045*** (0.214)	2.080***(0.213)
Average #pref. votes/list	0.359***(0.045)	0.396***(0.042)	0.382*** (0.041)	0.392***(0.042)
# candidates on list	−0.043***(0.003)	−0.033***(0.004)	−0.035*** (0.004)	−0.034***(0.004)
Web1 exp. reg. voter		0.189 (0.949)
Web1 exp. reg. voter²		−0.685 (2.969)
Web2 exp. reg. voter		−0.046 (0.083)
Web2 exp. reg. voter²		0.003 (0.033)
Trad. exp. reg. voter		0.160***(0.029)
Trad. exp. reg. voter²		−0.025***(0.003)
Web 1.0			−0.068 (0.039)
Web 2.0			0.045 (0.032)
Total expenditure^a			0.016***(0.003)
Total expenditure²			−0.0001* (0.0001)
Rel. exp. Web1				−0.002 (0.004)
Rel. exp. Web1²				0.00004 (0.00005)
Rel. exp. Web2				−0.004 (0.007)
Rel. exp. Web2²				0.0001 (0.0002)
Rel. exp. Trad.				0.085***(0.022)
Rel. exp. Trad²				−0.005* (0.002)
Incumbent		0.130*** (0.028)	0.104*** (0.028)	0.114*** (0.029)
Mayor		0.242*** (0.035)	0.240*** (0.035)	0.241*** (0.036)
Alderman		0.133*** (0.032)	0.136*** (0.032)	0.138*** (0.032)
Local councilor		0.104*** (0.025)	0.106*** (0.025)	0.106*** (0.026)
First listposition		1.315*** (0.046)	1.315*** (0.045)	1.305*** (0.047)
Last listposition		0.342*** (0.041)	0.358*** (0.041)	0.363*** (0.041)
Second listposition		0.597*** (0.039)	0.596*** (0.039)	0.597***(0.040)
Third listposition		0.450*** (0.040)	0.471*** (0.040)	0.481*** (0.041)
First successor		0.207*** (0.045)	0.225*** (0.045)	0.216*** (0.046)
Higher spending cap		0.210*** (0.047)	0.051 (0.047)	0.104* (0.051)
Federal parliament		−0.148 (0.116)	−0.046 (0.111)	−0.124 (0.114)
Walloon parliament		−0.031 (0.116)	0.097 (0.111)	−0.022 (0.113)
Flemish parliament		−0.020 (0.142)	0.096 (0.137)	0.003 (0.140)
Gender (male)		−0.102***(0.020)	−0.105*** (0.020)	−0.106***(0.020)
Media exposure		0.001***(0.000)	0.0009*** (0.00009)	0.001***(0.0001)
Intercept	−1.923***(0.205)	−2.881***(0.199)	−2.866*** (0.191)	−2.783***(0.196)
N (level 1)	1253	1251	1253	1253

European parliament constitutes the ref. category; N (Level 3) = 30 constituencies; N (level 2) = 191 lists.

Per thousand Euro (the absolute amount has been divided by 1000).

p < 0.05; **p < 0.01; ***p < 0.001.

The third model includes the total expenses of the candidates, together with two dummies expressing whether a candidate made expenses for Web 1.0 and Web 2.0 tools, respectively. Is this way we can check whether there is, on top of the overall spending effect, a bonus for candidates who invest in digital tools. The result shows that the total expenses, together with the quadratic term, have the expected effect. Candidates who spend more obtain a larger share of preferential votes, but this effect decreases with higher spending as indicated by the negative effect of the quadratic term. Yet, we do not find an additional effect of digital tools, as neither of the two digital dummies have a significant effect. Again, we do not find support for hypotheses H1, H2, and H4.

The fourth model is yet another way to check the robustness of our findings. It includes the relative spending measures, capturing an individual candidate’s share of the total expenses in the constituency, as explained above. Again, we find a statistically significant positive effect for investments in traditional campaign instruments, together with a significant negative effect of the quadratic term. But no such effect is found for the relative investment in either Web 1.0 or Web 2.0 digital tools. Thus, in whatever way the expense variables are operationalized, no support is found for H1, H2, and H4.

Incidentally, the effects of the control variables at the individual level in these models are perfectly in line with previous research. Occupying a critical list position boosts the number of preferential votes. Incumbents and local office holders receive significantly more votes. The electoral result also strongly depends on the amount of newspaper coverage for a candidate. Although women candidates on average receive less preferential votes than men, when all the above-mentioned variables are controlled for, being a woman appears to be an advantage, as the male dummy in the model has a significantly negative effect. Finally, there appears to be no difference between the various assemblies with respect to the number of preferential votes.

We also hypothesized, based on the equalization logic, that challengers would benefit more from digital Web 2.0 expenses than incumbents. Given the above reported null finding regarding digital expenses, this would imply that the effect is only positive and significant for challengers. However, as shown in Table 4, there is no robust interaction effect between either of the spending variables and incumbency. Model 5 suggests a reverse effect, in the sense that investing in Web 2.0 tools is particularly effective for incumbents. But we should not make much of that, as this result is not confirmed by Models 6 and 7, with the alternative operationalizations of the spending variables. The result of Model 6, with the two digital dummies, even points in the opposite direction, as the interaction between incumbency and the Web 2.0 dummy is now negative. All in all, these results confirm the absence of a robust incumbency effect in proportional systems, in line with earlier research. We find no support for the hypothesis (H5) that such an incumbency effect materializes if we focus specifically on digital Web 2.0 expenses.

Table 4.

Restricted Maximum Likelihood Estimation with the Logged Number of Preference Votes (As the Percentage of Valid Votes in the Constituency) as Dependent Variable (Unstandardized Coefficients).

	Model 5	Model 6	Model 7
Incumbent	0.205*** (0.041)	0.190** (0.056)	0.176*** (0.044)
Web1 exp. reg. voter	0.306 (0.985)
Web1 exp. reg.voter × incumb	−0.198 (0.741)
Web1 exp. reg. voter²	−0.691 (3.032)
Web2 exp. reg. voter	−0.140 (0.094)
Web2 exp.reg. voter × incumb	0.111* (0.050)
Web2 exp. reg. voter²	0.018 (0.034)
Trad. exp. reg. voter	0.187*** (0.030)
Trad. Exp. reg. voter × incumb	−0.054*** (0.014)
Traditional exp. reg. voter ²	−0.024 *** (0.003)
Web 1.0		−0.079 (0.042)
Web 1.0 × incumbent		0.001 (0.049)
Web 2.0		0.098** (0.036)
Web 2.0 × incumb		−0.191** (0.061)
Total expenditure^a		0.016*** (0.003)
Tot exp × incumbent		0.004* (0.002)
Total expenditure²		−0.0001** (0.00005)
Rel. exp. Web1			−0.004 (0.004)
Rel. exp. Web1 × incumb			0.003 (0.003)
Rel. exp. Web1²			0.0001 (0.0001)
Rel. exp. Web2			−0.008 (0.007)
Rel. exp. Web2 × incumb			0.006 (0.004)
Rel. exp. Web2²			0.0001 (0.0002)
Rel. exp. traditional			0.092*** (0.022)
Rel. exp. traditional × incumb			−0.027** (0.010)
Rel. exp. Traditional²			−0.004 (0.002)
Intercept	−2.872*** (0.196)	−2.869*** (0.192)	−2.786*** (0.197)
N (Level 1)	1251	1253	1253

Models including the interaction terms. N (Level 3) = 30 constituencies; N (Level 2) = 191 lists. All the control variables have been included but not reported. A table which includes the control variables can be found in the Online Appendices C.4.

Per thousand Euro (the absolute amount has been divided by 1000).

p < 0.05; **p < 0.01; ***p < 0.001.

Conclusion and Discussion

Building on the argument that digital campaigning tools need to result in electoral gains to have the potential to equalize or normalize the electoral playing field (Gibson and Mcallister, 2015), this article looked at the equalization/normalization debate from the perspective of campaign expense effect research. More specifically, we looked into the effects of investing financial resources in three different types of campaign instruments: owned Web 1.0 digital media (such as websites), paid Web 2.0 digital media (such as sponsored posts and targeted ads on social media), and traditional offline campaigning (such as flyers, posters, and ads in the traditional media). In a context of increasing personalization, it can be expected that Web 2.0 tools, allowing to direct seemingly personal messages to individual voters, will be more effective than both Web 1.0 and traditional tools. As long as Web 2.0 tools are relatively cheap, they will arguably be particularly effective for challengers, according to the equalization model. These tools allow the challengers to make up for their competitive disadvantage in name recognition at a low cost.

However plausible these hypotheses may sound, none of them was confirmed by our analysis of 1253 high-quality candidates in the Belgian federal, regional and European elections of 2019. We found a very robust null effect for both Web 1.0 and Web 2.0 digital expenses. In whichever way these variables were operationalized, they did not have a significant effect of the electoral outcome. They are only the expenses for traditional campaign tools which are positively correlated with the election result. Neither does such a digital campaign expense effect materialize specifically for challengers. We did not find a robust and consistent interaction effect between incumbency and either of the expense variables. This is in line with earlier research, showing that the lower spending efficacy for incumbents is typical for majoritarian systems, but is much weaker or absent in proportional ones.

These results contradict an apparently growing belief among politicians and analysts that Web 2.0 campaigning is the magic formula to electoral success. It is taken for granted that elections are no longer won with billboards and flyers, but with ads on Facebook. Yet this conviction is hardly evidence based. As mentioned above, research on the effect of digital tools on the election result of individual candidates in PR systems is scarce. There are only a few studies, with mixed results. In this sense, our results row less against the current than might seem at first sight.

Given these results, might it not be advisable for politicians to abandon Facebook and return to the good old billboard and flyer? A first reason to be cautious in drawing such far-reaching conclusions is that our result is based on a snapshot of the 2019 election. While a large majority of 77.9% of the analyzed Belgian candidates invest in digital tools, they generally do so with small amounts. Belgium appears to be in a phase where, on the one hand, the use of digital tools has become widespread but, on the other hand, candidates spend only half-heartedly on these tools. They are ready to experiment with modern techniques, but do not yet dare to take the leap to a full digital campaign. This is not without consequences as prior research by Sudulich and Wall (2011) shows that in order for a campaign tool to be effective, it seems necessary that the spending on said tool passes a critical threshold. Furthermore, the study also indicates that campaign diversification can only positively influence electoral performance when enough budget is available for every tool included in the campaign. If not, it is better to invest in a smaller number of campaign instruments. We also have to bear in mind that our measure focusses on spending. It does not evaluate the quality of each candidate’s campaign, nor does it consider the candidate’s personal characteristics, such as rhetorical skills, tech savviness, celebrity status, physical looks, or the size and quality of one’s network.

Using the full vote generating potential of these paid Web 2.0 tools requires candidates to make expensive investments in data-driven targeted ads. Taking into account the legal expenditure cap, this would also imply spending the larger part of the campaign budget on digital tools. Yet, this is obviously still a bridge too far for most candidates. This reluctant and experimental use of Web 2.0 tools is probably typical of what Gibson (2020) calls the third phase of digital campaigning.

The implication is that the expected strong effect of Web 2.0 tools will appear once a country enters the fourth phase, and the use of these tools becomes more professional and more expensive. But this development might be slowed down by legal restrictions. As already mentioned, the spending cap puts a brake on a boost of digital expenses. Also, ethical concerns with regard to data-driven (micro) targeted campaigning might lead institutions such as the European Union to regulate online political messages more strictly. As a result, it remains to be seen whether a country such as Belgium will soon enter the fourth phase and digital campaigning will really become effective at the level of individual candidates. In the meantime, the impact of digital campaigning on the individual election result should not be exaggerated, to say the least.

Supplemental Material

sj-docx-1-psx-10.1177_00323217231179189 – Supplemental material for Does Digital Campaigning Make a Difference for Individual Candidates in an Open List Proportional Representation System? The Case of the 2019 Election in Belgium

Supplemental material, sj-docx-1-psx-10.1177_00323217231179189 for Does Digital Campaigning Make a Difference for Individual Candidates in an Open List Proportional Representation System? The Case of the 2019 Election in Belgium by Gunther Vanden Eynde and Bart Maddens in Political Studies

Footnotes

Acknowledgements

The authors thank Théo Kraakman for his assistance with the data collection.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the Research Foundation Flanders (FWO) under PhD Grant No. 11H1920N.

ORCID iD

Gunther Vanden Eynde

Supplemental Material

Additional Supplementary Information may be found with the online version of this article.

Contents

A. Additional Information

A1. Allocation of seats to candidates in the Belgian PR system

A2. District level descriptives

Table A1. Electoral Districts for the Elections of the Federal Parliament.

Table A2. Electoral Districts for the Elections of the Flemish Parliament.

Table A3. Electoral Districts for the Elections of the Walloon Parliament.

Table A4. Electoral Districts for the Elections of the European Parliament.

A3. Expense declaration document

A4. List of newspapers used for data collection

B. Variables

B1. Correlation tables expense variables

Table B1. Correlation Table Expenditure per Registered Voter Measures.

Table B2. Correlation Table Relative Measures.

C. Analyses

C1. Null model—Random intercept

Table C1. Null Model.

C2. Restricted Maximum Likelihood with Kenward-Roger correction

Table C2. Restricted Maximum Likelihood Estimation With Kenward-Roger Correction and the Logged Number of Preference Votes (As the Percentage of Valid Votes in the Constituency) as Dependent Variable (Unstandardized Coefficients).

Table C3. Model Fit Statistics of the Models Without Interaction Effects (Restricted ML).

Table C4. Model Fit Statistics of the Models With Interaction Effects (Restricted ML).

C3. Model fit statistics

C4. Interaction Models with control variables

Table C5. Interaction Models With Control Variables.

References

Notes

Author Biographies

Gunther Vanden Eynde is a Researcher at the KU Leuven Public Governance Institute (Belgium) and Research Fellow of the Research Foundation Flanders (FWO). He is preparing a PhD in political science on digital campaign expenses. His research focuses on political finance, campaign spending, and the social media campaigns of political parties and their candidates.

Bart Maddens is full Professor of Political Science at the KU Leuven Public Governance Institute. His research interests include political finance, elections, and multilevel systems. His work has been published, among others, in the European Journal of Communication, Journal of Information Technology and Politics, West European Politics, Party Politics, Electoral Studies, and Political Psychology.

References

Abramowitz

(1988) Explaining Senate Election Outcomes. The American Political Science Review 82 (2): 385–403.

Andre

DePauw

Shugart

, et al. (2017) Party Nomination Strategies in Flexible-List Systems : Do Preference Votes Matter ? Party Politics 23 (5): 589–600.

Benoit

Marsh

(2003) For a Few Euros More: Campaign Spending Effects in the Irish Local Elections of 1999. Party Politics 9 (5): 561–582.

Benoit

Marsh

(2010) Incumbent and Challenger Campaign Spending Effects in Proportional Electoral Systems: The Irish Elections of 2002. Political Research Quarterly 63 (1): 159–173.

Bouteca

Smulders

Maddens

, et al. (2019) “A Fair Day’s Wage for a Fair Day’s Work”? Exploring the Connection between the Parliamentary Work of MPs and Their Electoral Support. The Journal of Legislative Studies 25 (1): 44–65.

Cheibub

Sin

(2020) Preference Vote and Intra-Party Competition in Open List PR Systems. Journal of Theoretical Politics 32 (1): 70–95.

Copeland

Römmele

(2014) Beyond the Base? Political Parties, Citizen Activists, and Digital Media Use in the 2009 German Federal Election Campaign. Journal of Information Technology & Politics 11 (2): 169–185.

Corrado

Firestone

(1996) Elections in Cyberspace: Toward a New Era in American Politics. Washington, DC: Aspen Institute.

Dodeigne

Pilet

J-B

(2021) Centralized or Decentralized Personalization? Measuring Intra-Party Competition in Open and Flexible List PR Systems. Party Politics 27 (2): 234–245.

10.

Dodeigne

Put

G-J

Teuber

(2023) Is Electoral Intraparty Competition More Intense in Urban Areas? Electoral Studies. Epub ahead of print 24 February 2023. DOI: 10.1016/j.electstud.2022.102571

11.

Duggan

(2020) Campaign Spending and Electoral Integrity: Assessing the Plausibility of the Challenger Spending Efficacy. PhD Thesis, Trinity College, Dublin.

12.

Folke

Rickne

(2020) Who Wins Preference Votes? An Analysis of Party Loyalty, Ideology, and Accountability to Voters. Journal of Theoretical Politics 32 (1): 11–35.

13.

Geys

Heyndels

(2003) Influence of “Cognitive Sophistication” on Ballot Layout Effects. Acta Politica 38 (4): 295–311.

14.

Gibson

Ward

(2000) An Outsider’s Medium? The European Elections and UK Party Competition on the Internet. British Elections & Parties Review 10 (1): 173–191.

15.

Gibson

(2020) When the Nerds Go Marching in: How Digital Technology Moved from the Margins to the Mainstream of Political Campaigns. Thousand Oaks, CA: Oxford University Press.

16.

Gibson

McAllister

(2006) Does Cyber-Campaigning Win Votes? Online Communication in the 2004 Australian Election. Journal of Elections, Public Opinion and Parties 16 (3): 243–263.

17.

Gibson

McAllister

(2011) Do Online Election Campaigns Win Votes? The 2007 Australian “YouTube” Election. Political Communication 28 (2): 227–244.

18.

Gibson

Mcallister

(2015) Normalising or Equalising Party Competition? Assessing the Impact of the Web on Election Campaigning. Political Studies 63 (3): 529–547.

19.

Green

Krasno

(1988) Salvation for the Spendthrift Incumbent: Reestimating the Effects of Campaign Spending in House Elections. American Journal of Political Science 32 (4): 884–907.

20.

Green

Krasno

(1990) Rebuttal to Jacobson’s “New Evidence for Old Arguments.” American Journal of Political Science 34 (2): 363–372.

21.

Hansen

Kosiara-Pedersen

(2014) Cyber-Campaigning in Denmark: Application and Effects of Candidate Campaigning. Journal of Information Technology & Politics 11 (2): 206–219.

22.

Jacobs

Spierings

(2016) # Social Media, Parties, and Political Inequalities. Basingstoke: Palgrave Macmillan.

23.

Jacobson

(1978) The Effects of Campaign Spending in Congressional Elections. American Political Science Review 72 (2): 469–491.

24.

Jacobson

(1990) The Effects of Campaign Spending in House Elections: New Evidence for Old Arguments. American Journal of Political Science 34 (2): 334.

25.

Jankowski

(2016) Voting for Locals: Voters’ Information Processing Strategies in Open-List PR Systems. Electoral Studies 43: 72–84.

26.

Johnston

Pattie

(2006) Candidate Quality and the Impact of Campaign Expenditure: A British Example. Journal of Elections, Public Opinion and Parties 16 (3): 283–294.

27.

Karvonen

(2010) The Personalization of Politics. A Study of Parliamentary Democracies. Colchester: ECPR Press.

28.

Koc-Michalska

Lilleker

Surowiec

, et al. (2014) Poland’s 2011 Online Election Campaign: New Tools, New Professionalism, New Ways to Win Votes. Journal of Information Technology & Politics 11 (2): 186–205.

29.

Kruschinski

Bene

(2022) In Varietate Concordia?! Political Parties’ Digital Political Marketing in the 2019 European Parliament Election Campaign. European Union Politics 23 (1): 43–65.

30.

Lusoli

(2005) The Internet and the European Parliament Elections : Theoretical Perspectives, Empirical Investigations and Proposals for Research. Information Polity 10 (34): 153–163.

31.

Maddens

Put

(2013) Office Effects and Campaign Spending in a Semi-open List PR System: The Belgian/Flemish Federal and Regional Elections 1999-2010. Electoral Studies 32 (4): 852–863.

32.

Maddens

Smulders

Wolfs

(2019) De Prijs van Politiek: Over de Portefeuille van de Partijen. Leuven: Lannoo Campus.

33.

Maddens

Wauters

Noppe

, et al. (2006) Effects of Campaign Spending in an Open List PR System: The 2003 Legislative Elections in Flanders/Belgium. West European Politics 29 (1): 161–168.

34.

Margolis

Resnick

(2000) Politics as Usual: The Cyberspace `Revolution′. Thousand Oaks, CA: Sage.

35.

Margolis

Resnick

Wolfe

(1999) Party Competition on the Internet in the United States and Britain. Harvard International Journal of Press/politics 4 (4): 24–47.

36.

Norris

(2003) Preaching to the Converted?: Pluralism, Participation and Party Websites. Party Politics 9 (1): 21–45.

37.

Palda

(1998) The Impact of Campaign Expenditures on Political Competition in the French Legislative Elections of 1993. Public Choice 94 (1): 157–174.

38.

Pattie

Johnston

Fieldhouse

(1995) Winning the Local Vote: The Effectiveness of Constituency Campaign Spending in Great Britain, 1983–1992. American Political Science Review 89 (4): 969–983.

39.

Put

Maddens

(2015) The Effect of Municipality Size and Local Office on the Electoral Success of Belgian/Flemish Election Candidates: A Multilevel Analysis. Government and Opposition 50 (4): 607–628.

40.

Römmele

Gibson

(2020) Scientific and Subversive: The Two Faces of the Fourth Era of Political Campaigning. New Media & Society 22 (4): 595–610.

41.

Samuels

(2001) Incumbents and Challengers on a Level Playing Field: Assessing the Impact of Campaign Finance in Brazil. The Journal of Politics 63 (2): 569–584.

42.

Sandberg

LAC

Öhberg

(2017) The Role of Gender in Online Campaigning: Swedish Candidates’ Motives and Use of Social Media during the European Election 2014. Journal of Information Technology & Politics 14 (4): 314–333.

43.

Schroeder

Sjoquist

Stephan

(2017) Understanding Regression Analysis: An Introductory Guide, 2nd edn.

Quantitative Applications in the Social Sciences 57. Thousand Oaks, CA: Sage.

44.

Southern

(2015) Is Web 2.0 Providing a Voice for Outsiders? A Comparison of Personal Web Site and Social Media Use by Candidates at the 2010 UK General Election. Journal of Information Technology and Politics 12 (1): 1–17.

45.

Southern

Lee

(2018) Politics as Usual? Assessing the Extent and Content of Candidate-Level Online Campaigning at the 2015 UK General Election. Journal of Elections, Public Opinion and Parties 29 (2): 1–20.

46.

Spierings

Jacobs

(2014) Getting Personal? The Impact of Social Media on Preferential Voting. Political Behavior 36 (1): 215–234.

47.

Sudulich

Wall

(2009) Keeping Up with the Murphys? Candidate Cyber-Campaigning in the 2007 Irish General Election. Parliamentary Affairs 62 (3): 456–475.

48.

Sudulich

Wall

(2011) How Do Candidates Spend Their Money? Objects of Campaign Spending and the Effectiveness of Diversification. Electoral Studies 30 (1): 91–101.

49.

Suiter

(2015) Political Campaigns and Social Media: A Study of #mhe13 in Ireland. Irish Political Studies 30 (2): 299–309.

50.

Tavits

(2010) Effect of Local Ties On Electoral Success and Parliamentary Behaviour: The Case of Estonia. Party Politics 16 (2): 215–235.

51.

Trumm

(2022) Online Versus Offline: Exploring the Link between How Candidates Campaign and How Voters Cast Their Ballot. European Journal of Political Research 61 (2): 485–502.

52.

Vaccari

(2013) Digital Politics in Western Democracies: A Comparative Study. Baltimore, MD: John Hopkins University Press.

53.

Van Aelst

Maddens

Noppe

, et al. (2006) Van de studio naar de kamer? Over de relatie tussen media en voorkeurstemmen bij de verkiezingen van 2003. In: Fiers

Reynaert

(eds) Wie zetelt? De gekozen politieke elite in Vlaanderen doorgelicht. Tielt, Belgium: Lannoo Campus, pp.163–176.

54.

Van Aelst

Sheafer

Stanyer

(2012) The Personalization of Mediated Political Communication: A Review of Concepts, Operationalizations and Key Findings. Journalism 13 (2): 203–220.

55.

Van Aelst

Strömbäck

Aalberg

, et al. (2017) Political Communication in a High-Choice Media Environment: A Challenge for Democracy? Annals of the International Communication Association 41 (1): 3–27.

56.

Vanden Eynde

Maddens

(2021) Explaining Digital Campaign Expenses: The Case of the 2018 Legislative Elections in Colombia. Journal of Information Technology & Politics 19 (3): 302–315.

57.

Vanden Eynde

Put

G-J

Maddens

(2022) The Coming of Age for Paid Digital Campaigning: Equalization or Normalization in the 2019 Belgian Federal Elections? Online Information Review. Epub ahead of print 3 October 2022. DOI: 10.1108/OIR-12-2021-0679

58.

Vanlangenakker

Maddens

Put

G-J

(2013) Career Patterns in Multilevel States: An Analysis of the Belgian Regions. Regional Studies 47 (3): 356–367.

59.

Vieira

de Almeida

MIS

Agnihotri

, et al. (2019) In Pursuit of an Effective B2B Digital Marketing Strategy in an Emerging Market. Journal of the Academy of Marketing Science 47 (6): 1085–1108.

60.

Wauters

Pittoors

Moens

(2019) Het gebruik van de voorkeurstem bij de parlementsverkiezingen van 26 mei 2019. Report, Gent University, Belgium, 26 May.

61.

Wauters

Thijssen

Van Aelst

, et al. (2018) Centralized Personalization at the Expense of Decentralized Personalization. The Decline of Preferential Voting in Belgium (2003–2014). Party Politics 24 (5): 511–523.

62.

Wauters

Weekers

Maddens

(2010) Explaining the Number of Preferential Votes for Women in an Open-List PR System: An Investigation of the 2003 Federal Elections in Flanders (Belgium). Acta Politica 45 (4): 468–490.

63.

Xenos

Foot

(2005) Politics As Usual, or Politics Unusual? Position Taking and Dialogue on Campaign Websites in the 2002 U.S. Elections. Journal of Communication 55 (1): 169–185.

64.

Zarouali

Dobber

De Pauw

, et al. (2022) Using a Personality-Profiling Algorithm to Investigate Political Microtargeting: Assessing the Persuasion Effects of Personality-Tailored Ads on Social Media. Communication Research 49 (8): 1066–1091.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.07 MB