Policy issues remain at the core of electoral choices and parties’ behaviour (Jocker et al. Reference Jocker, van der Brug and Rekker2025). Voters are inclined to support parties or candidates they associate most closely with the issues they consider most important (Bélanger and Meguid Reference Bélanger and Meguid2008; Lachat Reference Lachat2014; Petrocik Reference Petrocik1996). Since evaluating all political options is cognitively demanding, voters tend to associate certain parties with specific issues: social-democratic parties with the welfare state, or Greens with the environment, to name a few. This phenomenon has been theorised through the concept of associative issue ownership (van der Meer and Damstra Reference van der Meer and Damstra2024; Walgrave et al. Reference Walgrave, Lefevere and Tresch2012; Walgrave et al. Reference Walgrave, Tresch and Lefevere2015). It implies that parties and candidates campaign to maintain their reputational advantage on ‘owned issues’, and try to influence voters’ perceptions accordingly (Stubager and Seeberg Reference Stubager and Seeberg2016). Another common strategy is negative campaigning, which involves attacking a political rival to undermine its credibility in the eyes of voters, while at the same time trying to present oneself as more competent (Seeberg and Nai Reference Seeberg and Nai2021). It is well-established that parties and candidates usually attack on issues that they (or their party) own (Damore Reference Damore2002; Elmelund-Præstekær Reference Elmelund-Præstekær2011; Poljak and Seeberg Reference Poljak and Seeberg2024; Seeberg Reference Seeberg2020), also because it helps advance their owned issues on the agenda. In the so-called ‘digital age’, such attacks are increasingly integrated into parties’ social media strategies – for instance on X (formerly known as Twitter) – aiming to shape the media agenda and, in turn, influence public opinion (Gilardi et al. Reference Gilardi, Gessler, Kubli and Müller2022). On social media, negative messages are more likely to go viral than neutral or positive ones (Jacobs et al. Reference Jacobs, Kins, Close and Marie2025; Peeters et al. Reference Peeters, Opgenhaffen, Kreutz and Van Aelst2023). This means that if parties want to set ‘their’ issues high on the agenda and maintain a grip on their image as an issue owner, going negative might just be the ideal strategy.
However, in highly fragmented party systems where voters have more parties to choose from (van der Meer and Damstra Reference van der Meer and Damstra2024), issue ownership can be ambiguous and contested: multiple parties can be associated with the same issue(s) and can therefore stand as competitors for issue ownership. In such a setting, it is unclear which strategy parties opt for: do they attack when they own the issue, or when they are a competitor for ownership of that issue, or both? Conversely, do they target the parties that own the issue, or do they go after competitors? The dynamics of that relationship remain unclear. Besides, the interplay between issue ownership and negative campaigning has mostly been studied in electoral settings (Elmelund-Præstekær Reference Elmelund-Præstekær2011; Seeberg and Nai Reference Seeberg and Nai2021) and less often outside of campaign times (Poljak and Seeberg Reference Poljak and Seeberg2024). However, routine and campaign periods can lead to very different communication strategies, especially in multi-party systems with coalition governments (Poljak and Van Aelst Reference Poljak, Van Aelst, Nai and Walter2025). This article seeks to explore how the interactional nature of ‘issue ownership competition’ drives negative campaigning in such settings.
In a nutshell, parties should make sure that they are visible on ‘their’ issues – especially those on which they stand a chance – and attract attention towards these issues, which they can best do by attacking in their (nowadays increasingly digital) communication. To disentangle how issue ownership and issue competition drive attacks, we investigate the social media communication of parties and party leaders in Belgium. Belgium exhibits a highly fragmented party system, with fierce competition for issue ownership between traditional and challenger political forces (Legein and van Haute Reference Legein, van Haute, Delwit and Van Haute2021), but also among traditional parties. Confronted with high levels of electoral volatility and intense competition for a small voter pool, parties are likely to challenge their rivals’ issue ownership. We argue that in such a context, issue competition rather than ownership alone – on the side of both the sender and the target of an attack – is a better predictor of issue-based negative campaigning. Hence, we consider both the content of negative campaign messages (issues) and the context (competition for ownership) and demonstrate that political actors’ attack strategy is not necessarily determined by issue ownership, but rather by the competition for ownership of the issues they talk about. In addition, by analysing two-and-a-half years of posts on X, comprising both routine and campaign periods, we determine whether these dynamics vary depending on the electoral context.
Our findings indicate that, in the Belgian context, issue-based attacks are not more likely to emanate from parties that own the issue at stake, while the effect of issue competition for the sender is only visible during the campaign period. Considering targets’ characteristics appears more crucial. Issue-based attacks are more likely to occur against parties that compete for ownership of the issue at stake than towards those which do not, with this effect also being slightly stronger during rather than outside the campaign. Interestingly, issue-based attacks expressed during the electoral campaign are more likely to be directed against the owner of the issue at stake – although this is not found during routine times. Overall, our findings suggest that issue ownership and issue ownership competition on the target side function as significant predictors of issue-based attacks in a multi-party and fragmented party system and also hint at the influence of the electoral context on political actors’communication strategy.
Party competition, issue ownership and negative campaigning
Party competition and issue ownership
According to issue ownership theory, political actors selectively emphasise certain issues in their communication, based on the reputation they have built in voters’ minds through their track record dealing with these issues (Budge and Farlie Reference Budge and Farlie1983; Lefevere et al. Reference Lefevere, Seeberg and Walgrave2020a). Aligning with the literature on issue voting, issue ownership therefore contributes to making the partisan landscape more intelligible to voters (Walgrave and Lefevere Reference Walgrave and Lefevere2013). Subsequently, it affects both political parties and voters’ behaviour: parties have an incentive to draw attention to the issues they are perceived to own; and voters are more likely to choose a party that ‘owns’ the issue(s) that matter(s) to them (Walgrave et al. Reference Walgrave, Tresch and Lefevere2015), as they deem it most capable of handling that issue (Bouteca and Lefevere Reference Bouteca, Lefevere, Caluwaerts, Dodeigne, Pilet and Reuchamps2020; Petrocik Reference Petrocik1996).
As a result, it is generally safer for parties to build their communication around owned issues (Tresch et al. Reference Tresch, Lefevere and Walgrave2018). In fact, issues contain both fixed and variable aspects. The former includes their ‘scope’, which corresponds to the relevance of a given issue for large swaths of the population or for a smaller, more specific group of people, whereas the latter applies to parties’ reputational advantage with a specific issue (Green-Pedersen and Seeberg Reference Green-Pedersen and Seeberg2025). Željko Poljak and Henrik Bech Seeberg note that one of the main takeaways from the issue ownership literature is that ‘party behaviour differs across issues’ (Poljak and Seeberg Reference Poljak and Seeberg2024: 270). Namely, issue ownership – but also issue salience – is most central to explaining how political parties and their representatives address issues in their communication.
Negative campaigning and issue ownership
In its simplest definition, negative campaigning refers to ‘any criticism levelled by one candidate against another during a campaign’ (Geer Reference Geer2008: 23). The literature usually distinguishes between issue-based attacks, pertaining to the substance, and personal-based attacks, focusing on traits or characteristics of the political opponent (Nai and Mendoza Reference Nai, Mendoza, Grömping and Wirz2025). In this study, we only consider issue-based attacks because they are central to how parties contest issue ownership and hence align with our research goals. A key question in research on negative campaigning has been to investigate who goes negative and why, with extant scholarship identifying three core explanatory factors: (1) the characteristics of the sender of an attack (e.g. their political profile, personality traits, incumbency status, electoral strength or ideological leanings); (2) the characteristics of the target of an attack (e.g. competition, ideological distance); and (3) the context in which the attack takes place (e.g. the political system or media environment) (Nai Reference Nai2020; Nai et al. Reference Nai, Tresch and Maier2022). Most prior studies have focused on the former aspect, while the latter two have received comparatively less attention (Maier and Nai Reference Maier and Nai2022).
The content of campaign messages – such as the issues that political actors emphasise – has previously been suggested as a potential motive for parties and candidates to go negative, while being inextricably linked to the context (Elmelund-Præstekær Reference Elmelund-Præstekær2011). Issue ownership itself has been found to be a compelling driver of negativity (Gerstlé and Nai Reference Gerstlé and Nai2019). This argument is anchored in the issue ownership literature: parties would selectively emphasise the issues on which they have a reputational advantage (Lefevere et al. Reference Lefevere, Seeberg and Walgrave2020a; Petrocik Reference Petrocik1996) and use their own ‘frames’ when they attack competitors (Seeberg Reference Seeberg2020). However, in multi-party systems, ownership of issues has been tied to a more positive rhetorical style (Elmelund-Præstekær Reference Elmelund-Præstekær2011: 209). In fact, parties have limited control over the political and media agenda and must cope with variable levels of scrutiny for the issues they wish to talk about. Consequently, parties tend to attack one another not only on issues that they own but also on salient issues (Poljak and Seeberg Reference Poljak and Seeberg2024). In other words: ‘when a party has ownership over a salient issue, it will devote the greatest attention to attacking on that issue’ (Poljak and Seeberg Reference Poljak and Seeberg2024: 269).
Other perspectives consider that issue ownership is not static and can be contested. One way in which parties can undermine their rival’s issue ownership is through using negative campaigning. Scholars speak of ‘issue trespassing’ or issue convergence when parties step up their communication on issues that are associated with their rivals (Blomqvist and Green-Pedersen Reference Blomqvist and Green-Pedersen2004; Wagner and Meyer Reference Wagner and Meyer2014; Walgrave et al. Reference Walgrave, Tresch and Lefevere2015). In fact, both party communication and media coverage of issues can exert influence on the issue ownership perceptions of the public (Bouteca and Lefevere Reference Bouteca, Lefevere, Caluwaerts, Dodeigne, Pilet and Reuchamps2020; Dahlberg and Martinsson Reference Dahlberg and Martinsson2015; Seeberg and Nai Reference Seeberg and Nai2021; Tresch and Feddersen Reference Tresch and Feddersen2019), providing yet another incentive for political actors to leverage specific issues in their communication. Although contesting issue ownership is a long-term strategy and can be challenging, it is possible for parties to ‘make a dent’ in their rivals’ issue reputation, for instance by attacking their lack of commitment to the issue (Lefevere et al. Reference Lefevere, Seeberg and Walgrave2020a). This is where negative campaigning could come in handy, especially in high-stake periods such as election campaigns (Tresch et al. Reference Tresch, Lefevere and Walgrave2015).
In fact, most parties do not focus their communication efforts solely on owned issues. Even (long-time) single-issue parties ‘engage in issue trespassing and the claiming of unowned issues in day-to-day politics’ (Bos et al. Reference Bos, Lefevere, Thijssen and Sheets2017: 761). To understand these dynamics, and to elucidate how issue-based negative campaigning plays out in fragmented multi-party systems, it is thus essential to go beyond reductive conceptions of issue ownership.
How issue ownership (competition) drives attacks beyond elections
Overall, research on the link between negative campaigning and issue ownership has mainly addressed ownership on the side of the ‘attacker’ (Elmelund-Præstekær Reference Elmelund-Præstekær2011; Poljak and Seeberg Reference Poljak and Seeberg2024). In contrast, the interplay between the content of an attack (i.e. issues) and its context (i.e. competition for ownership), on the one hand, and the characteristics of both the sender and target of political attacks, on the other, has not yet been tackled, despite the centrality of target choice in the decision to attack (Walter Reference Walter2014). Furthermore, notwithstanding the documented differences between routine and campaign periods (Poljak and Van Aelst Reference Poljak, Van Aelst, Nai and Walter2025), there is little to no research on how issue competition unfolds publicly, on a day-to-day basis, as opposed to during election campaigns. The present article aims to fill these gaps by analysing the interplay of issue ownership competition and negative campaigning in a fragmented multi-party context with fierce competition over issues (Marié et al. Reference Marié, Kins, Close and Jacobs2024). Theoretically, we put forward the role of issues and interparty competition over them as potential explanatory factors of political actors’ decisions to resort to negative campaigning. We posit that an attack is deliberately directed at certain actors, based on the competition for ownership of the issues at stake.
We study this on social media, which act as central channels of ‘permanent communication’ (Severin-Nielsen et al. Reference Severin-Nielsen, Kruikemeier, Ohme and Gade Kjelmann2025), where political actors defend their issue positions and react to ongoing news events, often impulsively (Ecker Reference Ecker2017; Petkevic and Nai Reference Petkevic and Nai2022). If parliamentary speeches and party manifestos cater to a rather limited audience, and are ruled by a set of both explicit and implicit norms (Dolezal et al. Reference Dolezal, Ennser-jedenastik, Müller, Praprotnik and Winkler2018; Poljak and Seeberg Reference Poljak and Seeberg2024; Schwarzbözl et al. Reference Schwarzbözl, Fatke and Hutter2020), social media platforms act as direct channels between the general public, citizens and political actors, where the latter can communicate in a relatively unconstrained manner. Therefore, they provide an ideal setting for attacking and discrediting political adversaries (Masroor et al. Reference Masroor, Khan, Aib and Ali2019). Prior research has suggested that campaign dynamics on social media resemble those observed in other channels with regard to negative campaigning (Auter and Fine Reference Auter and Fine2016; Petkevic and Nai Reference Petkevic and Nai2022).
Nonetheless, some studies have found that social media may in fact be more prone to negativity (Klinger et al. Reference Klinger, Koc-Michalska and Russmann2023; Lehrner Reference Lehrner2021), as negative and emotional appeals typically reach a larger audience online (Fine and Hunt Reference Fine and Hunt2023), powered by mostly opaque platform algorithms. X, in particular, is well suited to hosting more aggressive forms of rhetoric (Yarchi et al. Reference Yarchi, Baden and Kligler-vilenchik2021). These dynamics have intensified since Elon Musk’s takeover of the platform, through his ‘free-speech absolutism’ approach to regulating discourse on the platform (Hardy et al. Reference Hardy, Salzano, Pfister, Danisch and Gronnvoll2024). Even though parties cannot pick the issues that are trending online, they can certainly choose which issues to highlight when they go after their opponents – and we believe they do. Hence, we argue that social media are well-suited to the investigation of political actors’ issue-based negative campaigning strategies (Haselmayer Reference Haselmayer2019). Besides, they also enable us to examine communication patterns across time and to compare between campaign and routine periods.
David Damore (Reference Damore2004) previously argued that it makes little sense for those who resort to negative campaigning to base their attacks on issues that are owned by a competitor, since they lack credibility on the issue. Looking at parliamentary debates, Poljak and Seeberg (Reference Poljak and Seeberg2024) recently confirmed that parties were indeed more likely to attack their adversaries on ‘owned’ issues. Our first hypothesis tests whether this link extends to the online arena:
Hypothesis 1 (H1): A social media post is more likely to contain an issue-based attack if the sender owns the issue(s) that is/are at stake in the post, compared to when the sender does not own the issue.
Our second hypothesis broadens the scope of issue ownership as a driver of negative campaigning by considering its ‘competitive’ aspect. We contend that political actors’ decision to attack will depend on whether there is something to be gained in terms of ownership of an issue. Following that line of reasoning, negative campaigning is used not only to defend ownership of an issue but also to leverage (or contest) a competitor status. In other words, we argue that the status of ‘ownership competitor’ matters just as much as that of owner, in the context of issue-based negative campaigning. We believe this to be particularly important to consider in multi-party systems, where multiple actors compete over the same issues, meaning that there is often not one sole, clear owner (see the Data, Method and Operationalisation section) for a given issue, leading us to expect that:
Hypothesis 2 (H2): A social media post is more likely to contain an issue-based attack when the sender is a competitor for ownership of the issue at stake, compared to when the sender is not such a competitor.
Third, we argue that a similar issue ownership competition logic applies to the choice of the target of an issue-based attack. Indeed, we believe the target’s status in relation to the issue(s) being raised (i.e. owner or competitor) also matters in political actors’ decisions to attack. Research shows that ownership reputations can vary across time (Christensen et al. Reference Christensen, Dahlberg and Martinsson2015) and that communication processes can partly contribute to these changes (Dahlberg and Martinsson Reference Dahlberg and Martinsson2015; Walgrave et al. Reference Walgrave, Lefevere and Nuytemans2009). From a strategic point of view, negative campaigning can be used by political actors to try to ‘steal’ ownership of an issue from a competitor. In a multi-party setting, it is therefore highly likely that issue owners will constitute the prime target for issue-based attacks:
Hypothesis 3 (H3): A social media post is more likely to contain an issue-based attack when the target owns the issue at stake, compared to when the target does not own the issue at stake.
However, targeting owners could be considered a risky and costly strategy, as the ownership of certain issues by parties can be considered almost ‘undisputable’ (Tresch et al. Reference Tresch, Lefevere and Walgrave2015). Instead, political actors might have more to gain from targeting competitors – those parties that are vying for ownership of an issue – having more room to compete. In other words, their chances of successfully making themselves visible and contesting ownership on a certain issue can be considered much greater in such a scenario. The rationale is the following: since attacks make messages more visible, and parties want to be associated with issues, when they attack, it will depend on their ‘chance of ownership’, which is greater in relation to competitors than owners. This gives us the following hypothesis:
Hypothesis 4 (H4): A social media post is more likely to contain an issue-based attack when the target is a competitor for ownership of the issue at stake, compared to when the target is not a competitor for the issue at stake.
Finally, we propose a more in-depth examination of potential differences between campaign and non-campaign (or routine) periods. This is justified by the long-standing scholarly debate over ‘permanent campaigning’, first coined by Blumenthal quite some time ago (1980). It refers to a mindset where electoral concerns are constantly guiding elected officials’ actions and has gathered increasing scholarly attention in recent years (Joathan and Lilleker Reference Joathan and Lilleker2023). According to this concept, the difference between electoral and routine periods has become blurred. This trend would have been boosted by the rise and generalisation of communication via digital means (Elmer et al. Reference Elmer, Langlois, Mckelvey and Kozolanka2018), where political actors compete ‘permanently’ in an attempt to dominate the news cycle (Larsson Reference Larsson2016). Accordingly, the expected effects listed earlier (H1–H4) should remain constant across the two periods investigated here – routine and campaign.
Nevertheless, some have argued that campaign periods are still exceptional and marked by intensified communication efforts by political actors as well as shifts in tone, style and content (Peeters et al. Reference Peeters, Opgenhaffen, Kreutz and Van Aelst2023; Poljak and Seeberg Reference Poljak and Seeberg2024). In multi-party systems with coalition governments, negative campaigning between government parties is generally limited – though not non-existent – during routine times, with most attacks originating from the opposition, and directed at incumbents (Close et al. Reference Close, Kins, Kumar and Jacobs2023; Haselmayer and Jenny Reference Haselmayer and Jenny2018). However, electoral campaigns provide an opportunity for government parties to challenge one another more intensely, differentiating themselves from their fellows in office. While it is assumed that campaigns lead to increased negativity (Auter and Fine Reference Auter and Fine2016; Damore Reference Damore2002), existing research on the social media communication of Belgian politicians has put that assumption to the test (Jacobs et al. Reference Jacobs, Kins, Close and Marie2025). To elucidate part of these debates, we set out to examine whether the impact of issue ownership (competition) on political actors’ attack behaviour changes once the campaign kicks in.
Case selection
We test our hypotheses on Belgium, a parliamentary democracy with a federal institutional set-up, characterised by a tradition of consociationalism (Deschouwer Reference Deschouwer2012). Several characteristics make it a most likely case to test our hypotheses. First, the consociational nature of the political system implies that parties from different ideological backgrounds within the two main linguistic communities must negotiate and forge hard-fought compromises to govern together. This would suggest that negative campaigning is less prevalent in Belgium than elsewhere – at least between members of the (federal) government coalition. However, previous studies of negative campaigning in Belgium, across both traditional and social media, seem to suggest the country is not immune to this type of communication (Close et al. Reference Close, Kins, Kumar and Jacobs2023; Lefevere et al. Reference Lefevere, Seeberg and Walgrave2020a; Poljak and Seeberg Reference Poljak and Seeberg2024). Nonetheless, attacks in the online realm have remained largely issue-focused, rather than personal-based (Jacobs et al. Reference Jacobs, Kins, Close and Marie2025). This makes Belgium an ideal case for studying the dynamics of issue-based attacks.
Second, the Belgian partisan landscape is extremely fragmented (Delwit and Lebrun Reference Delwit, Lebrun, Delwit and van Haute2021). During our period of investigation, the Chamber (i.e. the federal parliament) included 13 parliamentary parties, from the most left-wing to the most right-wing: PTB-PVDA, Groen, PS, Ecolo, Vooruit, Les Engagés, CD&V, DéFI, Open Vld, MR, N-VA and Vlaams Belang (Delwit and van Haute Reference Delwit and van Haute2021). Seven parliamentary parties competed in the Dutch-speaking community and six competed in the French-speaking one. Party system fragmentation has gradually increased the number of parties needed to reach a majority in parliament. From 2020 to 2024, the ‘De Croo I’ federal government included a record number of seven parties (four Dutch-speaking and three French-speaking), representing the four ‘mainstream’ party families: the liberals (MR, Open Vld), socialists (PS, Vooruit), Greens (Ecolo, Groen) and the Dutch-speaking Christian-democrats (CD&V).
Third, in terms of associative issue ownership in Belgium, there are clear ownership patterns for some ‘niche’ parties, such as the Greens (i.e. climate and the environment) or the radical right (i.e. migration, law and order) (see Tables 1 and 2 in the Supplementary Material) (Peeters and Coffé Reference Peeters and Coffé2025; Tresch et al. Reference Tresch, Lefevere and Walgrave2015; Walgrave et al. Reference Walgrave, Lefevere and Nuytemans2009, Reference Walgrave, Lefevere and Tresch2012; Walgrave and Lefevere Reference Walgrave and Lefevere2017). Issue ownership is less clear for other party families. Regionalists – especially on the Flemish side – are associated with the State Reform issue, but the N-VA has also become increasingly associated with other issues (e.g. migration, taxation), therefore challenging the main issue owner, Vlaams Belang. Liberals are mostly associated with taxation, and socialists with social security, but they are in competition for ownership of labour issues. These patterns can be attributed to the presence of several – sometimes overlapping – old and new cleavages (i.e. economic, community, identity) (Delespaul et al. Reference Delespaul, Meeusen, Abts and Swyngedouw2025; Kins Reference Kins2025). Parties’ issue emphasis in their online communication reflects this associative ownership and the competition over ownership, as seen in Tables 3 and 4 in the Supplementary Material (Marié et al. Reference Marié, Kins, Close and Jacobs2024).
Lastly, Belgian party actors have high rates of social media adoption (Kins and Claessens Reference Kins and Claessens2023) and have heavily invested in paid advertising (AdLens 2022). This means that despite its small size, Belgium’s digital communication landscape is comparable with that of larger Western European countries. On various platforms, communication from party organisations is still quite centralised around the official party accounts and/or the official party leader(s)’ accounts, reflecting the ‘partitocratic’ nature of the political system (De Winter Reference De Winter, Goetz and Meyer-sahling2013).Footnote 1 In other words, parties (and their leaders) exert significant influence and dominate the political life of the country, with campaigns being strongly centred around these actors (De Winter and Baudewyns Reference De Winter and Baudewyns2015). Hence, negativity should be especially common in these high-attention profiles, as their messages are more likely to be noticed and amplified.
Data, method and operationalisation
Data
This study examines the social media communication of Belgian parties and their leaders on X.Footnote 2 We believe this platform provides an ideal venue for observing direct interactions between elites and their agenda-setting strategies. During our fieldwork, the platform was still used daily by almost all political parties and their leaders.Footnote 3 Even though fewer than 10% of voters regularly use the platform (Close and Kins Reference Close and Kins2024), political elites use it intensively to send ‘signals’ to competitors, and as a way to get their messages across to journalists (Gilardi et al. Reference Gilardi, Gessler, Kubli and Müller2022). The period of observation started in January 2022, halfway between two electoral periods (2019 and 2024). This allows us to examine a non-electoral ‘routine’ period as well as the official 2024 election campaign, which started on 9 February 2024 and ended on election-day (9 June). This four-month period is referred to as the ‘period of caution’, which forces parties to comply with stricter rules, primarily connected to campaign spending (including online).
Our dataset includes all posts published through official party and party leader accounts between 1 January 2022 and 9 June 2024. Given Belgium’s partitocratic and leader-centric dynamics, these accounts provide the clearest window into party-level issue ownership (Wauters et al. Reference Wauters, Van Aelst, Thijssen, Rodenbach, Smulders, Pilet, Deschouwer, Delwit, Hooghe, Baudewyns and Walgrave2015; Lefevere et al. Reference Lefevere, Van Aelst and Peeters2020b). The official X accounts of all 13 parties (and party leaders) that were represented in the federal parliament at the time were included (see Table 5 in the Supplementary Material). ‘Posts’ (ex-‘tweets’) were collected through a combination of manual and automated techniques (i.e. web scraping in R using the API).Footnote 4 They were collected ex post, meaning that some deleted publications could not be included. A check indicated that deleted posts from party accounts in Belgium are rare and, therefore, unlikely to affect the validity of our analysis. The final dataset amounts to 30,377 posts.Footnote 5
Quantitative content analysis
Our unit of analysis is each post published by an official party or party leader account during the observation period. Reposts with comments (ex-‘cited tweets’) were also included; simple reposts of other users’ content without any commentary were not included, as they do not contain an individual claim from the actor. All posts were analysed in their original language. We considered each post as a political claim (Giugni et al. Reference Giugni, Koopmans, Passy and Statham2005) and resorted to quantitative content analysis to code the posts in terms of key variables of interest, according to a purpose-built codebook which can be consulted in Table 6 in the Supplementary Material. The coding was carried out by the authors – who have ample experience with quantitative content analysis – with the assistance of a student coder. All coders were trained to apply the coding scheme to the data. About 3% of the dataset was double-coded. Inter-coder reliability scores were satisfactory, meeting the minimal requirements prescribed by Klaus Krippendorff (Reference Krippendorff2018) (see Table 7 in the Supplementary Material and see Kins et al. Reference Kins, Jacobs and Close2026 for more details).
Dependent variable: issue-based negative campaigning
This study focuses on issue-based negative campaigning, drawing from the well-established distinction between issue- or policy-based and personal-based attacks (Auter and Fine Reference Auter and Fine2016; Lefevere et al. Reference Lefevere, Seeberg and Walgrave2020a; Walter Reference Walter2014). We began by assessing each post in its entirety to determine whether (or not) it contained an attack. This includes all textual information, but also any visible links, or articles attached to the post and pictures that were immediately discernible (i.e. that did not require further clicking to another page). We then assessed whether the attack was personal- or issue-based. On the one hand, we identified personal-based attacks in posts that focus on politicians’ character, criticism of their suitability and capability as an elected official and posts that were outright attacks on the personality, experience or knowledge of another party or political actor, without or with limited references to issues, policy and/or substantial politics (Garzia and Ferreira da Silva Reference Garzia and Ferreira da Silva2021). Personal attacks are not limited to individuals and can also encompass a whole party family, for instance when the radical right declares that ‘all greens are lunatics’.
On the other hand, issue-based attacks encompass criticism of a political opponent’s issue positions, or their inability to deal with a specific issue in the case of incumbents.Footnote 6 Our analysis uncovered that 28.3% of all posts (n = 8,593) contained an attack (of any type) and that issue-based attacks (n = 7,289; 24.0%) were much more frequent than personal-based attacks (n = 3,111, 10.2% of the whole dataset).Footnote 7 Negative campaigning is therefore a relatively frequent phenomenon, but it does not dominate partisan communication online. The resulting dependent variable is binary, with an issue-based attack being either absent from (0) or present in (1) each post.Footnote 8 Our emphasis on issue-based negative campaigning is justified by the fact that the focus of our theoretical puzzle rests on issue ownership competition between parties and that this type of attack is the most frequently used by political actors in our dataset (see Figure 1 in the Supplementary Material).
Independent variables
Issue ownership
Operationalising issue ownership is challenging. In this study, we focus on associative issue ownership, instead of relying on the issue emphasis of parties in their manifesto (Tresch et al. Reference Tresch, Lefevere and Walgrave2015). The latter is often regarded as less reliable, and insufficient on its own (Walgrave et al. Reference Walgrave, Tresch and Lefevere2015). To attribute ownership of issues to parties, we use data from the 2019 RepResent electoral survey (Michel et al. Reference Michel, Feitosa, Lefevere, Pilet, Van Erkel and Van Haute2024), in which respondents were asked to designate the party they associated with a battery of issues. We then matched the list of issue categories in our coding scheme with that of the RepResent survey. Note that our codebook contained more issues than those featured in the survey (see full codebook, Table 6 in the Supplementary Material).
We then created a dummy variable that captured whether a social media post published by a given political actor mentioned at least one issue for which their party is designated as owner in the survey (i.e. the party associated with the issue by the highest share of respondents) (dichotomous, 1 or 0).Footnote 9 Parties’ status as owner is indicated in Tables 1 and 2 in the Supplementary Material. For the attack target, we use a variable that describes whether the rival party mentioned in a post is the owner of at least one of the issues being discussed in this post (dichotomous, 1 or 0). For instance, if the radical left party PTB attacks the fiscal policy positions of the liberal MR – which owns the taxation issue – that variable will be coded as 1. Concrete examples of how this plays out in a post can be found in Table 8 in the Supplementary Material.
Issue competition
We adopted a two-step procedure to code our ownership competitor sender and target variables. First, we proceeded ‘vertically’, by looking at the second most associated party (issue competitor) for each individual issue. For example, the ‘law and order’ issue was attributed to the radical right VB by 35.5% of respondents, and to the Flemish-nationalist N-VA by 26.3% of respondents, meaning that we consider N-VA as competitor for ownership of that issue. Note that two contrasting realities can be observed in Dutch-speaking and French-speaking Belgium: while the competition for ownership among Francophones mostly plays out between PS and MR – the two most prominent parties at the time – issue ownership is much more scattered across Flemish parties, although N-VA (the largest party) stands out as the clear competitor for several issues. For instance, when the radical right VB accuses N-VA of ‘betraying the Flemish cause’ (i.e. the State Reform issue), it does so as a clear competitor on that issue. Conversely, when it attacks the integration policy of the N-VA-led Flemish government, it is targeting its main competitor on the issue of migration: the N-VA (for more examples, see Table 8 in the Supplementary Material).
In a second step, we examined the remaining parties that were neither owners of nor competitors on an issue, based on the initial procedure. Three parties were concerned: PVDA in Flanders, and Les Engagés and DéFI in French-speaking Belgium. For these, we proceeded horizontally, making them ‘competitor’ on the issue for which they had the highest score of association, while also drawing on previous work on these parties’ issue reputations (Delwit and van Haute Reference Delwit and van Haute2021). This approach ensured that issue ownership competition was evaluated for all parties.Footnote 10 The resulting variables are dichotomous (0 or 1).
Campaign
Our third key independent variable separates the routine and campaign periods (Vasko and Trilling Reference Vasko and Trilling2019). We created a binary variable that sorts all the posts that were published between 1 January 2022 up to (but excluding) 9 February, which is typically recognised as the official start of the campaign (see above). Posts published between 9 February and 9 June 2024 (inclusive) are considered part of the campaign period.
Control variables
Incumbency status
To ensure the robustness of our findings, we include a variable pertaining to the incumbency status of parties. The incumbency status of both the sender and the target is a key determinant of negative campaigning (Nai Reference Nai2020). This is also the case in Belgium, where attacks were found to mostly originate from opposition parties and to target government parties (Close et al. Reference Close, Kins, Kumar and Jacobs2023). Since Belgium is a federal country with multiple regional governments and a federal government, we have coded the level to which the post refers and whether the party is in the majority or opposition for that level (see Table 9 in the Supplementary Material). This variable has been coded for both the sender and the target, resulting in two dummy variables, based on whether the sender (or target) is incumbent at any level (1), or not (0).
Leader vs party account
We also include a dummy variable that captures whether a post was published by the party leader (1), or the official party account (0). Research has shown that party leaders’ accounts are more likely to engage in negative campaigning than official party accounts (Close et al. Reference Close, Kins, Kumar and Jacobs2023).
Ideological orientation
Further, since negative campaigning is often driven by more radical – especially right-wing – political actors in Belgium and elsewhere (Jacobs et al. Reference Jacobs, Kins, Close and Marie2025; Lehrner Reference Lehrner2021), we account for parties’ ideological position (for the sender of the post). We rely on the Chapel Hill Expert Survey to attribute scores that capture parties’ distance from the ideological centre. We make use of the version of the survey closest to our period of investigation (2019) and compute a continuous variable that corresponds to the absolute value of the subtraction of parties’ left–right score from the ideological centre (5), which we insert for all parties (Bakker et al. Reference Bakker, de Vries, Edwards, Hooghe, Jolly, Marks, Polk, Rovny, Steenbergen and Vachudova2015). The higher the score, the larger the distance from the ideological centre.Footnote 11
Issue variables
Finally, we control for issues. First, we created a dummy variable pertaining to whether a post deals with an issue for which issue ownership data are available (based on our operationalisation, see above) or not. Hence, this variable indicates whether the issue at stake in the post is owned by any party, or none.
Second, given that policy issues that are more salient are more likely to spark attacks (Poljak and Seeberg Reference Poljak and Seeberg2024), we include a dummy variable that also indicates whether the issue contained in a post is a salient issue (1) or not (0), during the month the post was emitted. Issues are considered salient if they were mentioned in at least 10% of all posts emitted throughout the month. Some issues are salient across the entire period: Economy, Justice, Social Security and Taxation. Rights and Liberties and Work were salient around 83–87% of the time (i.e. during 25 and 26 months over the 30-month period, respectively) and were discussed in more than 10% of the posts. Law and Order (salient for 16 months), Energy (13 months), Migration (11), Climate (10), Foreign Affairs (9) and Culture and Media (7) display more variation in terms of saliency over the period, (e.g. due to the energy crisis following the Russia–Ukraine war). Other issues were salient in a limited number of months, such as Education. Finally, some specific issues never reached the saliency threshold (e.g. State Reform, Defence, Mobility). Importantly, some of these (non)salient issues had owners, while some did not. The full list of issues and their frequency in the dataset can be found in Table 10 in the Supplementary Material.
By controlling for issues, we ensure a clean comparison, while also accounting for the fact that some issues are more salient at different moments. Intercoder reliability was relatively high across all issues. Finally, information about the issue density of each post can be found in Table 11 in the Supplementary Material.
Analysis and results
To test our hypotheses, we employ multilevel logistic regression models. We report odds ratios in Table 1 and beta coefficients in the coefficient plot (Figure 1). Multilevel modelling is particularly well suited for our analysis, as our data are hierarchically structured, with posts (level 1) nested within political parties (level 2).Footnote 12 Our null model shows that about 13.4% of the variance is located at this level, which validates our decision to apply multilevel modelling techniques. We adopt a stepwise approach. Model 1 tests our main effects (H1–H4), while Model 2 includes interaction terms to test whether the effects differ between the campaign and routine periods. The two models include the control variables discussed above.Footnote 13
Coefficient Plots – Model 1 and Model 2
Figure 1 Long description
The horizontal axis is labeled with coefficient values and shows tick labels negative 1, 0, 1, 2 and 3. The vertical axis lists predictors: Sender owns issue; Sender is competitor; Sender is incumbent; Target owns the issue; Target is competitor; Target is incumbent; Campaign; Part y leader account; Distance to center; Issue has an owner; Issue is salient; Campaign X Sender owns issue; Campaign X Sender is competitor; Campaign X Target owns the issue; Campaign X Target is competitor. Two plotted series are shown in a legend: Model1 and Model2. Each predictor row contains point markers with horizontal line segments. Most point markers appear close to the 0 tick on the coefficient axis. One point marker appears near the 3 tick on the coefficient axis on the row labeled Target is incumbent.
Predicting Issue-based Negative Campaigning Based on Issue Ownership (Competition)
Table 1 Long description
The table reports two logistic regression models predicting whether a message is an issue-based attack, shown as odds ratios with standard errors and significance. Across both models, the strongest association is target incumbency, linked to dramatically higher odds of being attacked (about twenty-five times higher in both models). Target competition for issue ownership is also consistently associated with higher odds of attack (about one and two-thirds higher in both models). In Model 2, targets that own the issue are less likely to be attacked, while this effect is not statistically clear in Model 1. Sender-side traits (sender owning the issue, being a competitor, or being an incumbent) show no statistically clear main effects in either model. Model 1 shows higher attack odds during campaigns, but in Model 2 the overall campaign indicator is not statistically clear once interactions are included. In Model 2, several campaign interactions are positive and statistically clear, including stronger attack odds when the sender is a competitor and when the target owns or competes for the issue. Issue-level controls indicate higher attack odds when an issue has an owner and when the issue is salient, with salience showing a strong and consistent increase in both models. Model fit improves slightly from Model 1 to Model 2 (lower AIC and BIC), with the same sample size and similar between-group variance and residual clustering.
Odds ratios are represented as exponentiated coefficients.
* p < 0.05.
** p < 0.01.
*** p < 0.001.
Overall, the models do not support H1: in the Belgian multi-party system, during our fieldwork (2022–2024), ownership of an issue did not increase the probability that a post on X by a political actor includes an issue-based attack – once other determinants of an issue-based attack are controlled for. In other words, political actors are not significantly more likely to attack an adversary if they own an issue than when they do not own that issue. In Model 1, issue-based attacks seem more likely to occur when the sender of the post is competing for ownership of the issue at stake than when it is not a competitor – yet the significance level barely reaches p < 0.1, meaning that we cannot fully confirm H2. Findings pertaining to issue ownership competition for the target, however, confirm our expectations. Issue-based attacks are significantly more likely to occur when the target owns the issue at stake than when it does not, leading us to confirm H3. Besides, issue-based attacks have a greater probability of occurring against a target that is a competitor on the issue at stake than against a target which is not a competitor (H4 confirmed). In other words, our case study suggests that issue-based attacks are not more likely when the sender owns or competes for ownership of the issue, but instead when the target and the sender compete on that issue.
Model 1 indicates a significant and positive effect of campaign: parties and party leaders are more likely to deliver issue-based attacks during the campaign than outside that period. Model 2 provides additional insights into how the effect of issue ownership (competition) on the sender and target sides varies between the routine and campaign periods. Figures 2–5 help visualise the interaction effects. Overall, timing (campaign vs routine) does not seem to fundamentally alter the relationship between the sender’s status as owner (Figure 2) or competitor for issue ownership (Figure 3) on the probability that a post will contain an issue-based attack. However, the effect of competition for issue ownership is slightly stronger during the campaign than during routine times (Figure 3). The effect of competition for issue ownership on the side of the target does not fundamentally change during the campaign compared to the routine period, although the effect appears slightly stronger during the campaign (Figure 5): parties and leaders seem to have reinforced their strategy to attack issue competitors during the campaign. Interestingly, we do uncover that campaign modifies the effect of a target’s ownership of the issue on the probability that a social media post will contain an attack (Figure 4). During the electoral campaign, issue-based attacks were more likely (all other variables kept constant) to be directed at parties that owned the issue at stake than towards parties that did not own the issue(s), whereas this strategy did not operate during routine times.
Predictive Margins – Interaction Effect Campaign X Sender Owns the Issue at Stake
Figure 2 Long description
The x- axis represents Campaign from 0 to 1. The y- axis represents Margins of predicted probabilities from 1 to 3. Two lines are shown with a legend: “Sender does not own the issue” and “Sender owns the issue”. At Campaign 0, the “Sender does not own the issue” line is at about 1.7 and the “Sender owns the issue” line is at about 1.8. At Campaign 1, the “Sender does not own the issue” line is at about 2.0 and the “Sender owns the issue” line is at about 2.1. At Campaign 1, a vertical error bar is shown for each line.
Predictive Margins – Interaction Effect Campaign X Sender Is a Competitor for Ownership of the Issue at Stake
Figure 3 Long description
The x- axis represents Campaign from 0 to 1. The y- axis represents Mean predicted values from minus 1 to 3. A legend on the right lists two lines: “Sender is not competitor” and “Sender is competitor”. At Campaign 0, the “Sender is not competitor” line is at about 1.8 and the “Sender is competitor” line is at about 1.7. At Campaign 1, the “Sender is not competitor” line is at about 2.3 with a vertical error bar extending to about 2.7 and the “Sender is competitor” line is at about 2.0 with a vertical error bar extending to about 2.4. Both lines slope upward from Campaign 0 to Campaign 1.
Predictive Margins – Interaction Effect Campaign X Target Owns the Issue at Stake
Figure 4 Long description
The x- axis label is Campaign. The x- axis range is 0 to 1. The y- axis label is Marginal predicted mean. The y- axis range is 0.15 to 0.30. A legend on the right lists two series: Target does not own the issue and Target owns the issue. The Target does not own the issue line passes through (0, 0.18) and (1, 0.26). The Target owns the issue line passes through (0, 0.19) and (1, 0.20). A vertical error bar is shown at x equals 1 for each series.
Predictive Margins I– nteraction Effect Campaign X Target Is a Competitor for Ownership of the Issue at Stake
Figure 5 Long description
The x- axis label is Campaign. The x- axis range is 0 to 1. The y- axis label is Marginal predicted mean. The y- axis range is 1 to 3. A legend on the right lists two lines: Target is not competitor and Target is competitor. The line for Target is not competitor passes through the points: (0, 2.1) and (1, 2.8). The line for Target is competitor passes through the points: (0, 1.6) and (1, 1.9). A vertical error bar is shown at x equals 1 for each line.
The effects of control variables confirm prior research. Incumbency of the target is a central determinant of the likelihood of issue-based attacks on X: the odds of having an issue-based attack are 25 times higher when government parties are targeted than when opposition parties are targeted. This also connects to the context of our case study, where the coalition at the federal level – made up of seven parties – was regularly the target of attacks by the opposition. During the period of investigation, issue-based attacks were more likely to occur for issues for which there was an owner, and for salient issues (i.e. issues on which all parties together communicated in more than 10% of all posts, by month). Other control variables were not significant: parties that are further from the centre are not more likely to attack, nor do we find a difference between official party and party leader accounts.
Discussion
This article set out to examine the relationship between issue ownership competition and issue-based attacks in Belgium on the social media platform X. By analysing a large dataset spanning both routine and official campaign periods, it provides new insights into how issue ownership (competition) influences online political communication strategies beyond elections, in a fragmented multi-party setting. A first takeaway of this study is that the issue ownership of the ‘attacker’ alone does not contribute to predicting negative campaigning. In other words, being an owner (i.e. the party most associated with an issue) or a competitor for ownership of an issue does not make a political party more likely to attack on that issue. This is a crucial distinction that refines existing models of issue-based negative campaigning (Elmelund-Præstekær Reference Elmelund-Præstekær2011; Poljak and Seeberg Reference Poljak and Seeberg2024; Seeberg and Nai Reference Seeberg and Nai2021). In Belgium’s fragmented party landscape, where issue ownership is often shared and contested (Legein and van Haute Reference Legein, van Haute, Delwit and Van Haute2021; Walgrave et al. Reference Walgrave, Tresch and Lefevere2015), our results suggest that negative campaigning strategies are above all driven by the targets’ issue ownership and issue competition. In that context, negative campaigning appears more as a strategy to contest targets’ issue reputations than to defend one’s own.
Second, issue ownership competition on the side of the target indeed emerges as a significant determinant of issue-based negative campaigning. Issue-based attacks are more likely to occur on social media when they target a competitor over the issue at stake – and this holds across the two time periods examined (campaign vs routine). Our study also complements work on issue trespassing (Blomqvist and Green-Pedersen Reference Blomqvist and Green-Pedersen2004; Wagner and Meyer Reference Wagner and Meyer2014), suggesting that attacks are often aimed at disrupting an opponent’s perceived ownership competitor status rather than merely positioning one’s own issue strengths. In general, the characteristics of the target do matter to a great extent in predicting issue-based negative campaigning, especially their incumbency status. Third, by comparing these communication dynamics during both campaign and routine periods, our analysis has uncovered that ‘timing matters’. The effect of a target’s competition over an issue seems slightly more pronounced during the official campaign compared to routine times. Interestingly, findings also indicate that the official campaign seems to be a privileged period to attack those parties which own the issue at stake, whereas this is not the case during the routine period. This demonstrates the importance of considering the context in which negative campaigning takes place. While past research has suggested that electoral settings amplify negativity (Damore Reference Damore2002; Elmelund-Præstekær Reference Elmelund-Præstekær2011), our findings add nuance to this assumption.
Nevertheless, this study has some limitations. First, while our analysis captures a broad time frame, future research could examine long-term trends in negativity, particularly across multiple election cycles, or during government coalition formation periods – which, in Belgium, can last several months and could either tame or prompt negativity. Besides, even though our comparison provides interesting and exploratory insights into communication strategies during routine and campaign times, we must acknowledge that these two periods are unequal in terms of length, with the routine period representing 25 months (and 83% of the posts) out of 30 in the dataset. This should invite us to provide more fine-grained analysis of context effects during shorter periods. Second, while we can elaborate on the underlying motivations that push political actors to attack their opponents, observational data provide limited evidence of these genuine motivations. Interviews with political actors and communication staff could provide additional insights. Third, while X provides a valuable platform to study the dynamics in political communication, it is only one part of the broader media ecosystem. Future studies could compare negativity on social media with traditional campaign channels, such as televised debates or party manifestos, to assess whether online communication differs systematically from other forms of political discourse – at different points in time over the election cycle. Fourth, while our study focuses on Belgium, comparative research could explore whether similar patterns hold in other multi-party systems with a similar degree of fragmentation, such as the Netherlands or Sweden, or in institutional contexts with distinct media systems and varying levels of political competition.
All in all, we believe our study contributes to the broader literature on issue ownership competition, political communication and digital campaigning. Our findings challenge simplistic narratives about negativity being confined to elections, demonstrating that political actors go negative not just to put their own issues on the agenda, but also to claim ownership of the issues that define political debate. Examining negative campaign dynamics over the whole electoral cycle contributes to the often-theorised permanent campaign logic in digital communication (Blumenthal Reference Blumenthal1980; Larsson Reference Larsson2016), showing that while political actors are always engaged in strategic communication, election periods still provide distinct incentives for negativity.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/gov.2026.10047.
Acknowledgements
The authors thank the reviewers of the 2023 Conference of the International Communication Association for their thoughtful comments. They also particularly wish to thank Elise Debroux, ULB Master's student, who worked for the Social Media Lab during the data collection period.
Financial support
This research has been supported by the FNRS (Fonds National de la Recherche Scientifique) and by the Université libre de Bruxelles (FER funding 2023–2024 – Fonds d’Encouragement à la recherche; ARC funding 2024–2027 – Action de Recherche Concertée).
Competing interests
The authors report that there are no competing interests to declare.
Open access
