Campaigning Online and Offline: Different Ballgames?

This research note investigates Flemish parties’ advertising on social media (Facebook) and in the written press during the final campaign weeks of the May 2019 general elections in Belgium. Since people had to vote for the federal, regional and European parliament, the stakes for parties were high, and prior evidence suggests this was amongst others reflected in campaign spending of over € 20 million.
This research note deals with two broad questions. First, to what extent have parties embraced Facebook ads as a new campaign tool? There is ample research about how social media have gradually become a crucial aspect of political campaigning (Jungherr 2016; Kreiss and McGregor 2018). We compare what Flemish parties have spent on Facebook ads and to what extent this has gone at the expense of newspaper ads. In addition, based on ad metadata, we present a basic analysis of the demographics of the groups that were exposed to social media advertisements.
Second, we ask whether this form of online advertising is substantially changing the way parties campaign. Based on a quantitative content analysis, we investigate several characteristics of parties’ campaign advertisements on Facebook and in newspapers. In this way, the content of a traditional offline campaign channel is contrasted with a relatively new channel that parties have added to their campaign toolkit. More concretely, we examine three aspects of the ads that are related to important topics in the campaign literature: the level of presidentialization, the issues of the campaign and the degree of negativity of the campaign ads.

• 1 Method

  We analyse parties’ campaign communication by examining two types of advertising: social media (Facebook) advertising and advertising in the written press (newspapers). As party advertising on broadcast media is limited, newspaper advertising is a mainstay in Flemish party campaigns. That said, social media are becoming, also in the Belgian context, increasingly important: not only do we see a drop in newspaper advertising (see Results), digital advertising also constitutes roughly one fourth of the total campaign expenditures of parties, totalling over 5 million euro (Vanden Eynde et al. 2019). Moreover, Facebook has disclosed party advertising through its Facebook ad library, providing researchers with a comprehensive overview of the advertisements that were published.
  The Facebook advertisements were collected through the Facebook ad library: we scraped all ads (N = 1,683) that appeared on the accounts of the party and the party chairman during the final three weeks of the campaign (4-25 May). We chose to focus on these two accounts for several reasons. First, we opted to include the party chairman’s account because many ads, while technically being posted by the party chairman, were funded by the party itself. As such, we consider these ads as equally representing party messaging. Second, although this selection excludes ads that were posted by other politicians of the party, it allows for a more unified selection criterion, since all major Flemish parties and party chairmen have active Facebook accounts, whereas this is not the case for all other politicians. Moreover, it is important to note that these ads, in contrast to regular posts, reach an audience beyond the followers of the party or party chair. Third, as our results show, the ads that were published on these two pages capture a substantial number of ads and represent a sizeable budget. Thus, by including the party leader Facebook account, the sample offers a better (but not complete) representation of the parties’ advertising. However, many ads were duplicates – identical ads that were either fielded on different days or shown to different publics. Therefore, we only use the full scraped set (N = 1,683) for our dissemination and cost analysis. The N = 807 unique advertisements – i.e. those advertisements published by a party that had a unique advertisement text – were coded in depth to track issue mentions, negativity and candidate appearances (see below), allowing us to appraise these elements.
  The newspaper sample consists of 756 advertisements that were published during the final week of the campaign (21-25 May) in six Flemish newspapers (De Morgen, De Standaard, De Zondag, Deze Week, Het Laatste Nieuws and Het Nieuwsblad). Although the periods differ, our results do not substantially differ if we restrict the social media analysis to the final campaign week, so we report the results for the full period. The newspaper advertisements were collected by student coders who went through the newspapers and marked the various advertisements in the editions of the newspapers.
  The advertisements were coded by two student coders. To ascertain the reliability of coding, several training sessions were organized, and a subset of the sample (N = 20) was double coded. Krippendorff’s alpha was at least 0.7 for all variables included in the analysis (Hayes and Krippendorff 2007).
  We focus on four sets of variables: those pertaining to presidentialization, issues, negativity and data on dissemination of the advertisements. First, to examine the level of presidentialization, we track the highest per cent of candidate mentions/features taken up by any single candidate. This gives us an indication of the extent to which parties centralized their campaigns around a single political leader, rather than a dispersed set of candidates. For each advertisement, up to five candidates were coded. If the advertisement contained more than five politicians, which was the case in less than 1% of the ads, the five that were most prominently displayed (as determined by size) were coded. We also track the dispersion of attention across candidates through the normalized Shannon H, which is a common measure of entropy that has been successfully used to measure scope of attention (Boydstun 2008). We use the normalized version of H, as this accounts for the fact that the maximum value of the non-normalized version increases as the number of unique politicians per party increases, and we know this varies substantially between parties. The normalized version of Shannon H corrects for this, resulting in a measure that ranges from 0 to 1, with higher values indicating greater dispersion of attention across politicians, and lower values indicating greater concentration – i.e. presidentialization. Second, we also track the issues that were mentioned in the advertisements by assigning up to three major topic codes based on the Belgian Comparative Agendas Project codebook (www.comparativeagendas.net). In total, there exist 25 major topic codes related to policy.¹x Economy, civil rights, health, agriculture, education, environment, energy, migration, mobility, justice, social affairs, housing, entrepreneurship, defence, science, foreign trade, foreign affairs, government affairs, spatial planning, culture, local and provincial governance, weather and natural disasters, fires and accidents, sport, church and religion. This allows us to examine to what extent parties use these ads to stress issues rather than persons and second whether parties stress different issues on different platforms. Third, with regard to negative campaigning, for each ad we assess whether the ad criticizes another party or a candidate of that party (1) or not (0), and we also examine both the party that is criticized and the type of attack (issue, person or both). Fourth and finally, based on the publicly available data on the Facebook ad library, we assess the demographic distribution of the social media advertisements in terms of gender and age groups.
  Unless otherwise noted, we always weight our analysis based on the estimated number of impressions (Facebook) or size (newspaper) of an advertisement. This helps us account for the fact that some ads had greater dissemination or were published much larger than other ads. Note that the Facebook ad library only provides brackets indicating the number of impressions (e.g. between 1,000 and 4,999), so we assign the mean value of the bracket to each advertisement. Although imperfect, it allows us to account for the fact that some ads reach much larger audiences than others. For newspapers, the size of an advertisement is expressed as the per cent of the page that is occupied by the advertisement, and ranges from 0 (0%) to 1 (100%).

• 2 Results

  2.1 Are Parties Changing from Offline to Online Advertising?

  Before we look at the use of parties’ ads in the 2019 election campaign, it is relevant to mention the shift in ad usage compared to the previous election. Comparing the newspaper ads that appeared in the six newspapers between 2014 and 2019 reveals that in 2014 there were substantially more newspaper advertisements in the final campaign week (N = 1,568) compared to 2019 (N = 756) (see Lefevere et al. 2019 for an analysis of 2014 newspaper ads). Although part of the drop has to do with shifts in the newspaper offering, particularly a reduction in the number of editions of De Zondag, all newspapers had less ads in 2019 compared to 2014. In short, there seems to be a general decline of party advertising in Flemish newspapers – presumably in favour of social media advertising.
  As a further exploration, Table 1 presents the number of ads per party on both Facebook and in newspapers. The table lists both the unweighted per cent of ads and the weighted per cent (based on the number of impressions for Facebook ads and size for newspaper ads), as comparing both numbers provides a first glance in the type of ads – i.e. few but large and/or highly disseminated ads versus many but smaller ads/ads with less reach. It is clear that parties differ in the relative attention they give to various channels. We can roughly see three patterns. First, the small opposition parties Groen and PVDA hardly use any type of ad, probably due to a lack of resources. Although Groen published quite a lot of ads on social media (18%), when accounting for their dissemination the number drops to 7%, indicating that their ads had more limited dissemination compared to other parties. Second, the three government parties have started to use Facebook ads, but still consider newspaper ads as relatively more important. N-VA as the biggest party, with the largest campaign budget, is present on both platforms, whereas Open VLD has the highest level of Facebook ads of the incumbent government parties. Yet, the relative difference between the two types of ads is similar for the three parties. N-VA dominated the newspaper advertisements, oftentimes with large, full-page ads featuring their party chairman, Bart De Wever. On social media, the N-VA took a similar strategy, focusing on fewer ads with high dissemination (as indicated by the small number of unweighted ads – 4% – which jumps to 12% if we account for their level of dissemination). Third, the opposition parties Vlaams Belang (VB) and s.pa clearly favour social media advertising. For VB this makes sense as they are blocked from advertising in many subscription newspapers, and their ads only appear in the free newspaper De Zondag. sp.a had a decent number of newspaper ads, and also had a few, but highly disseminated ads on Facebook: although they had 353 ads on Facebook, many of them were duplicates (our analysis indicates 169 unique ads) that were either fielded on different days or disseminated to different groups.

  Table 1. Number of ads in newspapers (weighted for ad size) and Facebook (weighted for impressions)

  Party	Newspaper ads		Facebook ads
  	Unweighted %	Weighted %	Unweighted %	Weighted %
  CD&V	14	22	18	8
  Groen	<1	<1	18	7
  N-VA	50	45	4	12
  Open VLD	25	26	23	17
  sp.a	9	14	21	24
  Vlaams Belang	2	2	10	27
  PVDA	0	0	7	5
  Total	756		1,683

  For social media, the Facebook ad library allows us to estimate the cost per ad, which enables us to compare the party spending on Facebook ads to the overall cost of parties’ digital campaigns. We rely on the VIVES data (KULeuven) on overall digital campaign spending to have an impression of the relative emphasis on social media campaigning within the overall digital campaign spending of the parties (Vanden Eynde et al. 2019). The left pane of Figure 1 has the overall digital campaign spending of parties based on the VIVES data during the four months before the election, whereas the right pane shows the estimated cost of the Facebook ads per party on the pages of the party leader and party itself during the short campaign of three weeks. Note that our measure of social media spending is also based on the ‘brackets’ information of the Facebook ad library, so absolute numbers should be interpreted with caution. That said, these figures do give us an indication of the spending of parties, relative to other parties and to their overall digital campaign budgets.

  Facebook (short campaign; party + leader) and overall digital campaign spending (VIVES data) for 2019 election campaign, per party
  

  For the most part, the patterns for the overall digital campaign and Facebook spending line up. What is immediately clear is that VB spent a lot more on digital media and Facebook advertising compared to other parties. During both the short campaign and the overall campaign, about half of its total campaign budget was assigned digital campaigning (Van den Eynde et al. 2019). Relatively speaking, VB’s Facebook spending estimated by our sample took up 22% of its digital campaign budget. Our findings indicate that N-VA spent comparatively less on Facebook ads (14%). We suspect that this is the case because the party decided to focus more on Google ads.
  Finally, the metadata from the Facebook ad library allows us to examine the demographic distribution of the party ads. This data only contains information on age and gender,²x Although technically information on the region is also available, given the split party system this is not very informative, as most of the ads are only shown to Flemish Facebook users. so admittedly our analysis is only a limited examination of this phenomenon. That said, given the fact that in the interpretation of the elections reference was regularly made to the (assumed) targeting of social media ads, even a cursory examination seems worthwhile. Figure 2 presents the per cent of impressions of Facebook ads, broken down by age and gender groups. We further show the distribution across age groups for all parties combined (black bars) and per party (gray bars).

  Per cent impressions of party advertisements amongst age/gender groups
  

  The figure shows the per cent of impressions amongst different age groups, with the lighter shaded area in each column indicating the per cent of females in that group, the darker shaded area the per cent of males. Given the Facebook demographic, it should not surprise that overall the younger age cohorts tended to be targeted more (black bars). The targeting was moreover not solely based on age groups, as we notice ‘spillover’ into the 13- to 17-year-old bracket, which seems odd given their inability to vote (now). As such, these users probably have been exposed to the ads because of membership to another segment that was targeted.
  In the interpretation of the successful campaign of VB, oftentimes the suggestion was made that they targeted young people. Our analysis indicates that although the party did target young people, other party ads had an even greater dissemination amongst younger age cohorts. For example, Groen had relatively more attention to voters in the bracket 18-24, and sp.a/CD&V in the age bracket 25-34. Of course, given the number of VB ads, in absolute terms the party still reached this segment comparatively more than other parties: 27.2% of VB ad impressions occurred amongst the age bracket of 18-24. Knowing that VB ads accounted for 27% of the overall impressions (see Table 1), this amounts to 7.3% of VB impressions amongst these age groups (27.2% * 27%). While Groen had, relatively speaking, more impressions amongst the same age bracket (31.8%), its ads only accounted for 7% of overall impressions (see Table 1). As such, Groen’s higher relative percentage only amounts to 2.2% of impressions, less than one third of the VB impressions amongst this group. Moreover, there does not seem to be an outspoken focus on men over women in the VB advertising demographics either, as it had the same per cent of reach amongst male voters as N-VA (both 57% male, 43% female). That said, another perspective on the same graph is that amongst the right-wing parties, the VB ads had much greater reach amongst younger demographics. Both N-VA and Open VLD had the lowest dissemination amongst the 18-34 age brackets. This suggests that on the right-hand side of the political spectrum, VB had greater dissemination amongst first-time voters, especially compared to Open VLD and N-VA.

  2.2 Is the Content and Style of Facebook Ads Different from Newspaper Ads?

  First, we look at the extent to which parties ‘presidentialized’ the elections – i.e. whether they focused attention on a single candidate, or on a wide array of candidates (Van Aelst et al. 2012). There are various ways of operationalizing presidentialization, but we opt here for two key measures. First, the per cent of candidate appearances in advertisements taken up by the most visible candidate: this gives us an indication of the concentration of candidate visibility around the ‘top’ candidate. Second, we report Shannon H’s measure of entropy, which provides a more generalized measure of the dispersion of candidate visibility across candidates (Boydstun 2008). Lower values indicate greater concentration of attention to fewer candidates, higher values a more evenly spread attention. Table 2 presents the results per party, for both social and newspaper advertisements. Note that because Groen (N = 2) and PVDA (N = 0) had no or very little newspaper ads, we do not report figures for these parties’ newspaper advertising.

  Table 2. Presidentialization in Flemish party ad campaigns

  Party	Facebook advertisements		Newspaper advertisements
  	Most mentioned candidate (%)	Shannon H (normalized)	Most mentioned candidate (%)	Shannon H (normalized)
  CD&V	Hilde Crevits (44)	0.69	Hilde Crevits (23)	0.56
  Groen	Meyrem Almaci (47)	0.65	–	–
  N-VA	Bart De Wever (38)	0.67	Bart De Wever (33)	0.66
  Open VLD	Maggie De Block (22)	0.72	Guy Verhofstadt (18)	0.76
  sp.a	John Crombez (64)	0.54	John Crombez (18)	0.84
  VB	Tom Van Grieken (64)	0.58	Tom Van Grieken (89)	0.29
  PVDA	Peter Mertens (78)	0.48	–	–

  The findings reveal substantial differences both between and within parties: VB most consistently presidentialized its campaign by drawing the vast majority of attention to the top candidate, party leader Tom Van Grieken. Across both newspaper and social media advertisements the party put forward the party chairman, leaving almost no attention for other candidates. PVDA had similarly focused attention on its party chairman, but our data does not allow us to test whether this presidentialization also held up in newspaper advertisements because the party simply had no advertisements in newspapers. The highly focused campaigns of these smaller parties are not uncommon, as they have less well-known candidates and no or few people in (local) government. sp.a focused its attention on its chairman as well, but interestingly only on social media (64%), and much less so in newspaper ads (18%). Part of the explanation is that many parties use newspaper ads seemingly more to increase the visibility of their leading candidates in different constituencies, leading to a plethora of candidates getting a bit of attention (with the exception of VB).
  CD&V, Groen and N-VA form the ‘middle’ of the pack, with leading candidates that take up most attention by a good margin, but not anywhere close to the numbers for VB and PVDA. Again, the leading candidate’s visibility bonus is less pronounced in newspaper ads. Finally, Open VLD presents an interesting case: the party had extremely dispersed attention, fielding a lot of different candidates in their advertising. It is hard to designate a clear ‘top’ candidate, as the difference with the other candidates is only 1 or 2%. Although the Facebook ad library does not allow us to test this, the strategy seems to also diversify the candidate offering geographically. This is picked up in the newspaper ads especially, with a very high dispersion index: unsurprisingly, in this setting it is the EP candidate Verhofstadt who manages to get most attention, as he is featured throughout Flanders given the single constituency in which EP elections are held. The other candidates tend to be visible only in newspaper editions for ‘their’ constituency, resulting in a high entropy (0.76).
  To provide insight into the issues that dominated the advertisement campaigns, we first look at the number of ads that could be considered as substantial, meaning that they received an issue code. Only 41% of newspaper ads mention any issue, while in contrast, 80% of Facebook ads had at least one issue mentioned. This indicates that many newspaper ads tend to serve a different purpose, and merely serve to present the (regional) candidates. The use of such ads is rampant in newspapers, but almost completely absent from social media. Table 3 presents the per cent of advertisements that mention various issues, for ads on social media and newspapers. Note that we omit issues that were not mentioned by any parties (e.g. sport, religion) from the table, and that because for each ad up to three issues could be coded the percentages sum to over 100%.

  Table 3. Per cent of social media and newspaper advertisements that mention issues

  Issue	Facebook ads (%)	Newspaper ads (%)
  Employment	25	17
  Migration	12	20
  Government affairs	10	2
  Social affairs	8	45
  Economy	8	43
  Health	7	0
  Justice	7	19
  Environment	5	1
  Energy	4	17
  Education	4	17
  Mobility	3	5
  Civil rights	2	0
  Housing	2	0
  Agriculture	1	0
  Foreign affairs	1	0
  Foreign trade	1	0

  Overall, the issues in social and newspaper advertising do not seem to have much correlation as newspaper and Facebook ads seem to stress different issues. That said, the economic conditions were important in both campaigns, either through discussions of the economy (8% Facebook, 43% newspapers) or employment, including the prominent debate on pension reform (25% Facebook, 17% newspapers). Migration was also discussed quite intensively in both Facebook and newspaper ads, as was justice (and crime). But several issues, for example social affairs, education, energy and government affairs, were much more dominant on either Facebook or newspaper ads. To some extent, this may be due to the different emphasis of parties on one medium or the other. For instance, VB pushed migration, but had less newspaper ads. To examine the extent to which the issue agendas of individual parties aligned, we calculated the correlation between parties’ issue attention in their Facebook and newspaper ads. Table 4 shows the correlation, as well as the most mentioned issue in a party’s social media and newspaper ads.

  Table 4. Alignment of parties’ issue agendas in Facebook and newspaper advertisements

  Party		Most mentioned issue on…
  	Correlation	Facebook ads (%)	Newspaper ads (%)
  CD&V	0.61	Employment (31)	Social affairs (87)
  Groen	–	Environment (61)	–
  N-VA	0.26	Funct. of Democracy (18)	Economy (40)
  Open VLD	0.24	Health (17)	Social affairs (73)
  sp.a	0.17	Employment (50)	Social affairs (77)
  VB	0.59	Migration (32)	Migration (96)

  What is immediately clear is that even at the individual party level, issue agendas are quite different depending on the medium involved. For two parties the correlations are relatively high (CD&V and VB), suggesting some alignment. For Groen, the data also suggests a clear focus on the environmental issue but due to the low number of newspaper ads it is hard to speak of an ‘agenda’ of newspaper ads. Yet, for N-VA, Open VLD and sp.a, the correlations are quite low. Looking at the raw data, we noticed that this seems to be due to the different nature of both type of ads. As mentioned before, newspaper ads tend to serve a different purpose and are less about issues and more about promoting individual candidates. In addition, the nature of the issue-based ads differs between social media and newspapers: whereas the newspaper ads are often broadly targeted and discuss the party platform as a whole (i.e. a letter from the leading candidate(s), see Figure 3), the Facebook ads deal with more specific and current policy issues, policy gaffes of opponents, specific issue positions and so forth – presumably because they can be much easier made and related to current events. In sum, the result of these two patterns is that parties’ issue attention on Facebook is much more dispersed compared to newspaper ads. Not only are there less issue-focused ads in newspapers, those that are published tend to deal with the broad, overarching issues that the party wants to push.

  Examples of letter style, broad-based newspaper ads (left: CD&V; middle: N-VA; right: Open VLD)
  

  Next, we discuss the negativity of the campaign: to what extent did parties go on the attack and who criticized whom? In terms of overall negativity, we find that 20% of social media advertisements contained an attack against another party. Yet, these advertisements had systematically more impressions – we cannot be sure whether this is because parties were pushing the attack ads more, or whether they were simply viewed more – because if we weight the data by the number of impressions, 29% of the advertisements was negative. As such, at least on social media the short campaign had substantial negativity. In contrast, in the newspaper advertisements, negativity was present in 18% of advertisements, accounting for ad size. Moreover, only N-VA, who warned of the danger of a ‘red-green front’ in about one third (35%) of its newspaper ads, used ads to ‘go negative’. While other parties did make very indirect references to other parties (i.e. Open VLD hinted at the fact that for them it is not your background that matters, but your future), or the fact that current policies are inadequate (e.g. VB’s argument that Flanders is providing welfare for the entire world), no direct attacks were made against other parties. Even N-VA’s attacks were somewhat unclear, as it is uncertain to what extent these attacks singled out the Flemish parties (Groen/sp.a) or the Walloon ones (Ecolo/PS).
  As such, we focus on negative campaigning in social media ads here. On social media, negative advertising was much more evenly spread across parties – and it was also much more direct. 54% of the attack ads were published by VB, followed by 20% sp.a, 12% N-VA, 6% by PVDA and Groen, and almost no attack ads by Open VLD (1%) and CD&V (less than 1%). Although it is sensible to expect VB to have the most attack ads since it had the most ads overall, VB’s advertising was amongst the most negative, with more than half of its ads (52%) being critical of other parties. PVDA (45%) had a similar level of negativity, whereas Groen (37%) and especially N-VA (23%) and sp.a (20%) had only a minority of attack ads on social media.
  To understand who criticized who, Table 5 presents the percentage of social media ads that were targeted at other parties. Because of the low number of negative ads by CD&V and Open VLD, we omit them from the table. Table entries indicate the per cent of a party’s attack ads that mention another party. For example, the 38% entry in the top left corner indicates that 38% of Groen’s attack ads attacked CD&V. Because attack ads could be targeted at multiple parties (e.g. all government parties), percentages sum up to over 100%.

  Table 5. Per cent of attack ads of a party targeting other parties

  Attacker	Target
  	CD&V (%)	Groen (%)	N-VA (%)	Open VLD (%)	sp.a (%)	VB (%)	PVDA (%)
  Groen (N = 20)	38	–	76	31	0	0	0
  N-VA (N = 9)	1	67	–	0	43	0	0
  sp.a (N = 34)	29	11	75	26	–	3	0
  VB (N = 51)	12	25	86	20	17	–	15
  PVDA (N = 37)	54	31	84	68	28	23	–

  A first pattern is that the incumbent coalition parties (CD&V, Open VLD and N-VA) were targeted by all opposition parties: Groen, sp.a, VB and PVDA all focused the majority of their attacks on the incumbent coalition. In this, the N-VA was the ‘primus inter pares’: as the largest party in the coalition, it received most attacks. During the campaign, the rivalry between N-VA and Groen was emphasized in the media: indeed, our data indicates a mutual dislike indicated by the high number of negative ads from Groen to N-VA (76%) and vice versa (67%). The second, and more surprising, pattern is that the extreme right VB attacked all other parties, with a particular focus on N-VA (86%), yet was almost completely left alone by the other parties, save for PVDA. In other words, while VB was free to launch attacks on other parties, it enjoyed relatively still waters itself.

• 3 Conclusion

  This research note presented an analysis of Flemish online and offline party advertising during the last weeks of the 2019 general election campaign. The election resulted in a win for VB, and prior analyses have indicated that important shifts occurred during the campaign (van Erkel et al. 2019), yet thus far no work has sought to lay bare the patterns in the campaign content itself. Based on a quantitative content analysis of Facebook and newspaper ads, we examined the use, (demographic) reach and content of parties’ advertising campaigns.
  Our results reveal several relevant findings. First, our analysis indicates that, as suggested by our note’s title, online and offline campaigning are indeed different ballgames. At a fundamental level, we simply observe different types of ads. Ads on social media tend to address current issues, more focused policy stances and direct attacks on competitors. In contrast, newspaper ads are much more about self-praise and self-presentation. Many newspaper ads simply feature an overview of candidates or a broadly aimed ‘letter’ by the leading candidate that is clearly meant to appeal to a broad audience. The audience characteristics are presumably driving parties to more ‘generic’ appeals in legacy media, as the message cannot be tailored to specific audiences. Newspaper ads basically only differ between regional editions as parties mainly want to promote the candidates who are electable in that region. In contrast, Facebook ads allow for so-called micro targeting, where fine-tuned ads are delivered to just the audience that will resonate with the message. However, the extent to which this is actually used in the Flemish context as the information of the Facebook ad library in this respect is limited remains unclear. Our demographic analysis shows that at least on the aggregate level all groups are targeted and that focusing overall campaign attention on specific socio-demographic audiences is not a very common practice. Of course, there are two important caveats to this claim: first, it only relates to overall attention of ads. Our goal here was to investigate broad patterns in party advertising during the 2019 elections, and as such we did not investigate whether specific (single) advertisements were tailored to specific demographic groups. Second, due to the data available on the Facebook ad library, we could only examine the relative exposure of age and gender groups, but it is likely that subsegments were selected based on other criteria as well. Our analysis does show that Facebook ads are used in a much more focused and flexible way, allowing them to interact with recent events or mistakes of opponents. In sum, our analysis indicates that an analysis of contemporary campaign communications should account for communication in various settings which is in line with prior research investigating party communication through different media (see, e.g. Walter and Vliegenthart 2010; Tresch et al. 2018).
  Second, the level of presidentialization differed markedly between parties, with VB and PVDA having more attention for their leading candidates than the other parties. CD&V, Groen and sp.a still had markedly more attention for their leading candidates, whereas Open VLD had by far the most evenly spread attention for candidates. An interesting pattern was the fact that newspaper ads are more diverse in general, because a substantial proportion seem to be meant to introduce local candidates, whereas these sorts of ads are much less prevalent on Facebook – in line with our first conclusion.
  Third, roughly one in three social media advertisements was negative, whereas this was less than one in five for newspaper ads – and then it was only N-VA that (briefly) warned of a ‘red-green front’. In social media ads, the negativity was targeted most at the government parties N-VA, Open VLD and CD&V, yet Open VLD and CD&V did not go on the attack themselves. The opposition party VB enjoyed a unique situation: although it criticized a lot of different parties, particularly N-VA, at least in advertising it was not subject to much criticism.
  In sum, this research note provided a first insight into the patterns of party communication both online and offline during the last stretch of the general election campaign of 2019. Although in some respects the findings indicate ‘business as usual’, for example with respect to leading candidates commanding more attention (Lefevere and Dandoy 2011), in other respects our findings reveal sometimes surprising patterns, for example regarding the difference between the type of ads launched on Facebook as compared to those that parties field in legacy media.

References

• Boydstun, Amber E. (2008). ‘Measuring the Scope of Political Communication’, Ohio State University.

• Erkel, Patrick F. A. van, et al. (2019). ‘Verschuivingen in Partijvoorkeur. Veranderingen in Partijvoorkeur 2014 - 2019, en tijdens de verkiezingscampagne van 2019’, Antwerp, Belgium.

• Hayes, Andrew F., and Klaus Krippendorff (2007). ‘Answering the Call for a Standard Reliability Measure for Coding Data’, Communication Methods and Measures, 1:1, 77-89.

• Jungherr, Andreas (2016). ‘Twitter Use in Election Campaigns: A Systematic Literature Review’, Journal of Information Technology & Politics, 13:1, 72-91.

• Kreiss, Daniel, and S. C. McGregor (2018). ‘Technology Firms Shape Political Communication: The Work of Microsoft, Facebook, Twitter, and Google with Campaigns During the 2016 U.S. Presidential Cycle’, Political Communication, 35:2, 155-177.

• Lefevere, Jonas, and Régis Dandoy (2011). ‘Candidate Choice in Political Advertising: What Determines Who Gets Attention?’, World Political Science Review, 7:1, http://www.bepress.com/wpsr/vol7/iss1/art5 (Accessed 19 May 2011).

• Lefevere, Jonas M., Knut De Swert, and Artemis Tsoulou-Malakoudi (2019). ‘Sub-constituency Campaigning in PR Systems. Evidence from the 2014 General Elections in Belgium’, Politics of the Low Countries, 1:3, 237-256.

• Tresch, Anke, Jonas Lefevere, and Stefaan Walgrave (2018). ‘How Parties’ Issue Emphasis Strategies Vary Across Communication Channels: The 2009 Regional Election Campaign in Belgium’, Acta Politica, 53:1, 25-47.

• Van Aelst, Peter, Tamir Sheafer, and James Stanyer (2012). ‘The Personalization of Mediated Political Communication: A Review of Concepts, Operationalizations and Key Findings’, Journalism, 13:2, 203-220.

• Vanden Eynde, Gunther, Gert-Jan Put, and Bart Maddens (2019). ‘Hoeveel kostte de digitale campagne van de Vlaamse partijen voor de federale, regionale en Europese verkiezingen van 26 mei 2019?’, KU Leuven: VIVES Research Center for Regional Economics.

• Walter, Annemarie S., and Rens Vliegenthart (2010). ‘Negative Campaigning Across Different Communication Channels: Different Ball Games?’, The International Journal of Press/Politics, 15:4, 441-461.

Noten

• 1 Economy, civil rights, health, agriculture, education, environment, energy, migration, mobility, justice, social affairs, housing, entrepreneurship, defence, science, foreign trade, foreign affairs, government affairs, spatial planning, culture, local and provincial governance, weather and natural disasters, fires and accidents, sport, church and religion.

• 2 Although technically information on the region is also available, given the split party system this is not very informative, as most of the ads are only shown to Flemish Facebook users.